d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
763fa8ae30
freebsd-skq/sys/dev/re/if_re.c

4053 lines
108 KiB
C
Raw Normal View History

Start each of the license/copyright comments with /*-, minor shuffle of lines
2005-01-06 01:43:34 +00:00
/*-
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
* Copyright (c) 1997, 1998-2003
* Bill Paul <wpaul@windriver.com>. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. All advertising materials mentioning features or use of this software
* must display the following acknowledgement:
* This product includes software developed by Bill Paul.
* 4. Neither the name of the author nor the names of any co-contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
* BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
* THE POSSIBILITY OF SUCH DAMAGE.
*/
Remove duplicate FBSDID's, move others to their right place.
2003-11-14 17:16:58 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/*
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
* RealTek 8139C+/8169/8169S/8110S/8168/8111/8101E PCI NIC driver
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
*
* Written by Bill Paul <wpaul@windriver.com>
* Senior Networking Software Engineer
* Wind River Systems
*/
/*
* This driver is designed to support RealTek's next generation of
* 10/100 and 10/100/1000 PCI ethernet controllers. There are currently
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
* seven devices in this family: the RTL8139C+, the RTL8169, the RTL8169S,
* RTL8110S, the RTL8168, the RTL8111 and the RTL8101E.
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
*
* The 8139C+ is a 10/100 ethernet chip. It is backwards compatible
* with the older 8139 family, however it also supports a special
* C+ mode of operation that provides several new performance enhancing
* features. These include:
*
* o Descriptor based DMA mechanism. Each descriptor represents
* a single packet fragment. Data buffers may be aligned on
* any byte boundary.
*
* o 64-bit DMA
*
* o TCP/IP checksum offload for both RX and TX
*
* o High and normal priority transmit DMA rings
*
* o VLAN tag insertion and extraction
*
* o TCP large send (segmentation offload)
*
* Like the 8139, the 8139C+ also has a built-in 10/100 PHY. The C+
* programming API is fairly straightforward. The RX filtering, EEPROM
* access and PHY access is the same as it is on the older 8139 series
* chips.
*
* The 8169 is a 64-bit 10/100/1000 gigabit ethernet MAC. It has almost the
* same programming API and feature set as the 8139C+ with the following
* differences and additions:
*
* o 1000Mbps mode
*
* o Jumbo frames
*
Whitespace pass
2004-07-06 02:46:53 +00:00
* o GMII and TBI ports/registers for interfacing with copper
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
* or fiber PHYs
*
Whitespace pass
2004-07-06 02:46:53 +00:00
* o RX and TX DMA rings can have up to 1024 descriptors
* (the 8139C+ allows a maximum of 64)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
*
* o Slight differences in register layout from the 8139C+
*
* The TX start and timer interrupt registers are at different locations
* on the 8169 than they are on the 8139C+. Also, the status word in the
* RX descriptor has a slightly different bit layout. The 8169 does not
* have a built-in PHY. Most reference boards use a Marvell 88E1000 'Alaska'
* copper gigE PHY.
*
* The 8169S/8110S 10/100/1000 devices have built-in copper gigE PHYs
* (the 'S' stands for 'single-chip'). These devices have the same
* programming API as the older 8169, but also have some vendor-specific
* registers for the on-board PHY. The 8110S is a LAN-on-motherboard
* part designed to be pin-compatible with the RealTek 8100 10/100 chip.
Whitespace pass
2004-07-06 02:46:53 +00:00
*
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
* This driver takes advantage of the RX and TX checksum offload and
* VLAN tag insertion/extraction features. It also implements TX
* interrupt moderation using the timer interrupt registers, which
* significantly reduces TX interrupt load. There is also support
* for jumbo frames, however the 8169/8169S/8110S can not transmit
fix jumbo frames as much as they can be fixed for re. We now cap the MTU to 7422 since it appears that the 8169S can't transmit anything larger.. The 8169S can receive full jumbo frames, but we don't have an mru to let the upper layers know this... add fixup so that this driver should work on alignment constrained platforms (!i386 && !amd64) MFC after: 5 days
2004-09-28 18:22:24 +00:00
* jumbo frames larger than 7440, so the max MTU possible with this
* driver is 7422 bytes.
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
*/
- Don't pollute opt_global.h with DEVICE_POLLING and introduce opt_device_polling.h - Include opt_device_polling.h into appropriate files. - Embrace with HAVE_KERNEL_OPTION_HEADERS the include in the files that can be compiled as loadable modules. Reviewed by: bde
2005-10-05 10:09:17 +00:00
#ifdef HAVE_KERNEL_OPTION_HEADERS
#include "opt_device_polling.h"
#endif
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
#include <sys/param.h>
#include <sys/endian.h>
#include <sys/systm.h>
#include <sys/sockio.h>
#include <sys/mbuf.h>
#include <sys/malloc.h>
Add missing <sys/module.h> includes
2004-05-30 20:08:47 +00:00
#include <sys/module.h>
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
#include <sys/kernel.h>
#include <sys/socket.h>
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
#include <sys/lock.h>
#include <sys/mutex.h>
Add simple MAC statistics counter reading support. Unfortunately useful counters like rl_missed_pkts is 16 bits quantity which is too small to hold meaningful information happened in a second. This means driver should frequently read these counters in order not to lose accuracy and that approach is too inefficient in driver's view. Moreover it seems there is no way to trigger an interrupt to detect counter near-full or wraparound event as well as lacking clearing the MAC counters. Another limitation of reading the counters from RealTek controllers is lack of interrupt firing at the end of DMA cycle of MAC counter read request such that driver have to poll the end of the DMA which is a time consuming process as well as inefficient. The more severe issue of the MAC counter read request is it takes too long to complete the DMA. All these limitation made maintaining MAC counters in driver impractical. For now, just provide simple sysctl interface to trigger reading the MAC counters. These counters could be used to track down driver issues. Users can read MAC counters maintained in controller with the following command. #sysctl dev.re.0.stats=1 While I'm here add check for validity of dma map and allocated memory before unloading/freeing them. Tested by: rmacklem
2010-11-05 19:28:00 +00:00
#include <sys/sysctl.h>
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
#include <sys/taskqueue.h>
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
#include <net/if.h>
The r48589 promised to remove implicit inclusion of if_var.h soon. Prepare to this event, adding if_var.h to files that do need it. Also, include all includes that now are included due to implicit pollution via if_var.h Sponsored by: Netflix Sponsored by: Nginx, Inc.
2013-10-26 17:58:36 +00:00
#include <net/if_var.h>
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
#include <net/if_arp.h>
#include <net/ethernet.h>
#include <net/if_dl.h>
#include <net/if_media.h>
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
#include <net/if_types.h>
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
#include <net/if_vlan_var.h>
#include <net/bpf.h>
#include <machine/bus.h>
#include <machine/resource.h>
#include <sys/bus.h>
#include <sys/rman.h>
#include <dev/mii/mii.h>
#include <dev/mii/miivar.h>
#include <dev/pci/pcireg.h>
#include <dev/pci/pcivar.h>
Move rl(4) to dev/rl.
2014-09-19 10:32:20 +00:00
#include <dev/rl/if_rlreg.h>
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
MODULE_DEPEND(re, pci, 1, 1, 1);
MODULE_DEPEND(re, ether, 1, 1, 1);
MODULE_DEPEND(re, miibus, 1, 1, 1);
The controller -> device switch happened in 4.0, fix comment
2005-10-22 05:01:11 +00:00
/* "device miibus" required. See GENERIC if you get errors here. */
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
#include "miibus_if.h"
Add MSI support. Ever since switching to adaptive polling re(4) occasionally spews watchdog timeouts on systems with MSI capability. This change is minimal one for supporting MSI and re(4) also needs MSIX support for RTL8111C in future. Because softc structure of re(4) is shared with rl(4), rl(4) was touched to use the modified softc. Reported by: cnst Tested by: cnst Approved by: re (kensmith)
2007-07-24 01:24:03 +00:00
/* Tunables. */
Do not use interrupt taskqueue on controllers with MSI/MSI-X capability. One of reason using interrupt taskqueue in re(4) was to reduce number of TX/RX interrupts under load because re(4) controllers have no good TX/RX interrupt moderation mechanism. Basic TX interrupt moderation is done by hardware for most controllers but RX interrupt moderation through undocumented register showed poor RX performance so it was disabled in r215025. Using taskqueue to handle RX interrupt greatly reduced number of interrupts but re(4) consumed all available CPU cycles to run the taskqueue under high TX/RX network load. This can happen even with RTL810x fast ethernet controller and I believe this is not acceptable for most systems. To mitigate the issue, use one-shot timer register to moderate RX interrupts. The timer register provides programmable one-shot timer and can be used to suppress interrupt generation. The timer runs at 125MHZ on PCIe controllers so the minimum time allowed for the timer is 8ns. Data sheet says the register is 32 bits but experimentation shows only lower 13 bits are valid so maximum time that can be programmed is 65.528us. This yields theoretical maximum number of RX interrupts that could be generated per second is about 15260. Combined with TX completion interrupts re(4) shall generate less than 20k interrupts. This number is still slightly high compared to other intelligent ethernet controllers but system is very responsive even under high network load. Introduce sysctl variable dev.re.%d.int_rx_mod that controls amount of time to delay RX interrupt processing in units of us. Value 0 completely disables RX interrupt moderation. To provide old behavior for controllers that have MSI/MSI-X capability, introduce a new tunable hw.re.intr_filter. If the tunable is set to non-zero value, driver will use interrupt taskqueue. The default value of the tunable is 0. This tunable has no effect on controllers that has no MSI/MSI-X capability or if MSI/MSI-X is explicitly disabled by administrator. While I'm here cleanup interrupt setup/teardown since re(4) uses single MSI/MSI-X message at this moment.
2011-01-26 20:25:40 +00:00
static int intr_filter = 0;
TUNABLE_INT("hw.re.intr_filter", &intr_filter);
Reclaim transmitted frames in re_tick(). This is for PCIe controllers that lose Tx completion interrupts under certain conditions. With this change it's safe to use MSI on PCIe controllers so enable MSI on these controllers.
2009-02-09 04:59:13 +00:00
static int msi_disable = 0;
Add MSI support. Ever since switching to adaptive polling re(4) occasionally spews watchdog timeouts on systems with MSI capability. This change is minimal one for supporting MSI and re(4) also needs MSIX support for RTL8111C in future. Because softc structure of re(4) is shared with rl(4), rl(4) was touched to use the modified softc. Reported by: cnst Tested by: cnst Approved by: re (kensmith)
2007-07-24 01:24:03 +00:00
TUNABLE_INT("hw.re.msi_disable", &msi_disable);
Prefer MSI-X to MSI on controllers that support MSI-X. All recent PCIe controllers(RTL8102E or later and RTL8168/8111C or later) supports either 2 or 4 MSI-X messages. Unfortunately vendor did not publicly release RSS related information yet. However switching to MSI-X is one-step forward to support RSS.
2011-01-25 22:18:00 +00:00
static int msix_disable = 0;
TUNABLE_INT("hw.re.msix_disable", &msix_disable);
Add a new tunable hw.re.prefer_iomap which disables memory register mapping. The tunable is OFF for all controllers except RTL8169SC family. RTL8169SC seems to require more magic to use memory register mapping. r187483 added a fix for RTL8169SCe controller but it does not looke like fix other variants of RTL8169SC. Tested by: Gavin Stone-Tolcher g.stone-tolcher <> its dot uq dot edu dot au
2009-03-09 06:02:55 +00:00
static int prefer_iomap = 0;
TUNABLE_INT("hw.re.prefer_iomap", &prefer_iomap);
Add MSI support. Ever since switching to adaptive polling re(4) occasionally spews watchdog timeouts on systems with MSI capability. This change is minimal one for supporting MSI and re(4) also needs MSIX support for RTL8111C in future. Because softc structure of re(4) is shared with rl(4), rl(4) was touched to use the modified softc. Reported by: cnst Tested by: cnst Approved by: re (kensmith)
2007-07-24 01:24:03 +00:00
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
#define RE_CSUM_FEATURES (CSUM_IP | CSUM_TCP | CSUM_UDP)
/*
* Various supported device vendors/types and their names.
*/
Remove duplicate const specifiers in many drivers (I hope I got all of them, please let me know if not). Most of these are of the form: static const struct bzzt_type { [...list of members...] } const bzzt_devs[] = { [...list of initializers...] }; The second const is unnecessary, as arrays cannot be modified anyway, and if the elements are const, the whole thing is const automatically (e.g. it is placed in .rodata). I have verified this does not change the binary output of a full kernel build (except for build timestamps embedded in the object files). Reviewed by: yongari, marius MFC after: 1 week
2012-11-05 19:16:27 +00:00
static const struct rl_type re_devs[] = {
After the change of r176757 re(4) no longer relys on reading RL_TXCFG register to identify a device in device probe. Reflect the fact by modifing device description with general ethernet controller family. Note, rl_basetype in struct rl_type is not used and the more detailed information is provided with rl_hwrev structure.
2008-07-02 04:55:39 +00:00
{ DLINK_VENDORID, DLINK_DEVICEID_528T, 0,
"D-Link DGE-528(T) Gigabit Ethernet Adapter" },
Add new device id of D-Link DGE-530T Rev. C controller. DGE-503T Rev A1 and B1 is supported by sk(4) but the DGE-530T Rev. C controller is re-branded RealTek 8169 controller. PR: kern/159116 Approved by: re (kib)
2011-07-30 01:06:12 +00:00
{ DLINK_VENDORID, DLINK_DEVICEID_530T_REVC, 0,
"D-Link DGE-530(T) Gigabit Ethernet Adapter" },
After the change of r176757 re(4) no longer relys on reading RL_TXCFG register to identify a device in device probe. Reflect the fact by modifing device description with general ethernet controller family. Note, rl_basetype in struct rl_type is not used and the more detailed information is provided with rl_hwrev structure.
2008-07-02 04:55:39 +00:00
{ RT_VENDORID, RT_DEVICEID_8139, 0,
"RealTek 8139C+ 10/100BaseTX" },
{ RT_VENDORID, RT_DEVICEID_8101E, 0,
Add support for RTL8105E PCIe Fast Ethernet controller. It seems the controller has a kind of embedded controller/memory and vendor applies a large set of magic code via undocumented PHY registers in device initialization stage. I guess it's a firmware image for the embedded controller in RTL8105E since the code is too big compared to other DSP fixups. However I have no idea what that magic code does and what's purpose of the embedded controller. Fortunately driver seems to still work without loading the firmware. While I'm here change device description of RTL810xE controller. H/W donated by: Realtek Semiconductor Corp.
2011-01-26 21:14:20 +00:00
"RealTek 810xE PCIe 10/100baseTX" },
After the change of r176757 re(4) no longer relys on reading RL_TXCFG register to identify a device in device probe. Reflect the fact by modifing device description with general ethernet controller family. Note, rl_basetype in struct rl_type is not used and the more detailed information is provided with rl_hwrev structure.
2008-07-02 04:55:39 +00:00
{ RT_VENDORID, RT_DEVICEID_8168, 0,
Add preliminary support for RTL8168G, RTL8168GU and RTL8411B. RTL8168GU has two variants(GMII and MII) but it uses the same chip revision id. Driver checks PCI device id of controller and sets internal capability flag(i.e. jumbo frame and link speed down in WOL). H/W donated by: RealTek Semiconductor Corp.
2013-10-29 05:30:21 +00:00
"RealTek 8168/8111 B/C/CP/D/DP/E/F/G PCIe Gigabit Ethernet" },
After the change of r176757 re(4) no longer relys on reading RL_TXCFG register to identify a device in device probe. Reflect the fact by modifing device description with general ethernet controller family. Note, rl_basetype in struct rl_type is not used and the more detailed information is provided with rl_hwrev structure.
2008-07-02 04:55:39 +00:00
{ RT_VENDORID, RT_DEVICEID_8169, 0,
Add a new RTL8169 variant, 8169SB(L)/8110SB(L). Reported by: nork Tested by: nork
2008-07-02 08:00:14 +00:00
"RealTek 8169/8169S/8169SB(L)/8110S/8110SB(L) Gigabit Ethernet" },
After the change of r176757 re(4) no longer relys on reading RL_TXCFG register to identify a device in device probe. Reflect the fact by modifing device description with general ethernet controller family. Note, rl_basetype in struct rl_type is not used and the more detailed information is provided with rl_hwrev structure.
2008-07-02 04:55:39 +00:00
{ RT_VENDORID, RT_DEVICEID_8169SC, 0,
"RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" },
{ COREGA_VENDORID, COREGA_DEVICEID_CGLAPCIGT, 0,
"Corega CG-LAPCIGT (RTL8169S) Gigabit Ethernet" },
{ LINKSYS_VENDORID, LINKSYS_DEVICEID_EG1032, 0,
"Linksys EG1032 (RTL8169S) Gigabit Ethernet" },
{ USR_VENDORID, USR_DEVICEID_997902, 0,
"US Robotics 997902 (RTL8169S) Gigabit Ethernet" }
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
};
Remove duplicate const specifiers in many drivers (I hope I got all of them, please let me know if not). Most of these are of the form: static const struct bzzt_type { [...list of members...] } const bzzt_devs[] = { [...list of initializers...] }; The second const is unnecessary, as arrays cannot be modified anyway, and if the elements are const, the whole thing is const automatically (e.g. it is placed in .rodata). I have verified this does not change the binary output of a full kernel build (except for build timestamps embedded in the object files). Reviewed by: yongari, marius MFC after: 1 week
2012-11-05 19:16:27 +00:00
static const struct rl_hwrev re_hwrevs[] = {
Sprinkle some const.
2011-11-02 23:23:19 +00:00
{ RL_HWREV_8139, RL_8139, "", RL_MTU },
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
{ RL_HWREV_8139A, RL_8139, "A", RL_MTU },
{ RL_HWREV_8139AG, RL_8139, "A-G", RL_MTU },
{ RL_HWREV_8139B, RL_8139, "B", RL_MTU },
{ RL_HWREV_8130, RL_8139, "8130", RL_MTU },
{ RL_HWREV_8139C, RL_8139, "C", RL_MTU },
{ RL_HWREV_8139D, RL_8139, "8139D/8100B/8100C", RL_MTU },
{ RL_HWREV_8139CPLUS, RL_8139CPLUS, "C+", RL_MTU },
Change model names of controller RTL_HWREV_8168_SPIN[123] to real ones. s/RL_HWREV_8168_SPIN1/RL_HWREV_8168B_SPIN1/g s/RL_HWREV_8168_SPIN2/RL_HWREV_8168B_SPIN2/g s/RL_HWREV_8168_SPIN3/RL_HWREV_8168B_SPIN3/g No functional changes.
2011-01-18 00:46:10 +00:00
{ RL_HWREV_8168B_SPIN1, RL_8169, "8168", RL_JUMBO_MTU },
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
{ RL_HWREV_8169, RL_8169, "8169", RL_JUMBO_MTU },
{ RL_HWREV_8169S, RL_8169, "8169S", RL_JUMBO_MTU },
{ RL_HWREV_8110S, RL_8169, "8110S", RL_JUMBO_MTU },
{ RL_HWREV_8169_8110SB, RL_8169, "8169SB/8110SB", RL_JUMBO_MTU },
{ RL_HWREV_8169_8110SC, RL_8169, "8169SC/8110SC", RL_JUMBO_MTU },
{ RL_HWREV_8169_8110SBL, RL_8169, "8169SBL/8110SBL", RL_JUMBO_MTU },
{ RL_HWREV_8169_8110SCE, RL_8169, "8169SC/8110SC", RL_JUMBO_MTU },
{ RL_HWREV_8100, RL_8139, "8100", RL_MTU },
{ RL_HWREV_8101, RL_8139, "8101", RL_MTU },
{ RL_HWREV_8100E, RL_8169, "8100E", RL_MTU },
{ RL_HWREV_8101E, RL_8169, "8101E", RL_MTU },
{ RL_HWREV_8102E, RL_8169, "8102E", RL_MTU },
{ RL_HWREV_8102EL, RL_8169, "8102EL", RL_MTU },
{ RL_HWREV_8102EL_SPIN1, RL_8169, "8102EL", RL_MTU },
{ RL_HWREV_8103E, RL_8169, "8103E", RL_MTU },
Add initial support for RTL8401E PCIe Fast Ethernet. PR: 154789
2011-02-16 21:59:42 +00:00
{ RL_HWREV_8401E, RL_8169, "8401E", RL_MTU },
Add preliminary support for RTL8402 PCIe FastEthernet with integrated card reader. H/W donated by: RealTek Semiconductor Corp.
2011-11-16 21:37:45 +00:00
{ RL_HWREV_8402, RL_8169, "8402", RL_MTU },
Add support for RTL8105E PCIe Fast Ethernet controller. It seems the controller has a kind of embedded controller/memory and vendor applies a large set of magic code via undocumented PHY registers in device initialization stage. I guess it's a firmware image for the embedded controller in RTL8105E since the code is too big compared to other DSP fixups. However I have no idea what that magic code does and what's purpose of the embedded controller. Fortunately driver seems to still work without loading the firmware. While I'm here change device description of RTL810xE controller. H/W donated by: Realtek Semiconductor Corp.
2011-01-26 21:14:20 +00:00
{ RL_HWREV_8105E, RL_8169, "8105E", RL_MTU },
Add preliminary support for second generation RTL8105E PCIe FastEthernet. H/W donated by: RealTek Semiconductor Corp.
2011-11-17 21:24:56 +00:00
{ RL_HWREV_8105E_SPIN1, RL_8169, "8105E", RL_MTU },
Add preliminary support for RTL8106E PCIe FastEthernet. H/W donated by: RealTek Semiconductor Corp.
2013-10-21 06:27:20 +00:00
{ RL_HWREV_8106E, RL_8169, "8106E", RL_MTU },
Change model names of controller RTL_HWREV_8168_SPIN[123] to real ones. s/RL_HWREV_8168_SPIN1/RL_HWREV_8168B_SPIN1/g s/RL_HWREV_8168_SPIN2/RL_HWREV_8168B_SPIN2/g s/RL_HWREV_8168_SPIN3/RL_HWREV_8168B_SPIN3/g No functional changes.
2011-01-18 00:46:10 +00:00
{ RL_HWREV_8168B_SPIN2, RL_8169, "8168", RL_JUMBO_MTU },
{ RL_HWREV_8168B_SPIN3, RL_8169, "8168", RL_JUMBO_MTU },
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
{ RL_HWREV_8168C, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K },
{ RL_HWREV_8168C_SPIN2, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K },
{ RL_HWREV_8168CP, RL_8169, "8168CP/8111CP", RL_JUMBO_MTU_6K },
{ RL_HWREV_8168D, RL_8169, "8168D/8111D", RL_JUMBO_MTU_9K },
{ RL_HWREV_8168DP, RL_8169, "8168DP/8111DP", RL_JUMBO_MTU_9K },
{ RL_HWREV_8168E, RL_8169, "8168E/8111E", RL_JUMBO_MTU_9K},
{ RL_HWREV_8168E_VL, RL_8169, "8168E/8111E-VL", RL_JUMBO_MTU_6K},
Add preliminary support for RTL8168EP. Submitted by: Edward O'Callaghan (eocallaghan <> alterapraxis dot com)
2013-10-29 05:37:05 +00:00
{ RL_HWREV_8168EP, RL_8169, "8168EP/8111EP", RL_JUMBO_MTU_9K},
Add preliminary support for RTL8168/8111F PCIe Gigabit ethernet. H/W donated by: RealTek Semiconductor Corp.
2011-11-17 22:07:50 +00:00
{ RL_HWREV_8168F, RL_8169, "8168F/8111F", RL_JUMBO_MTU_9K},
Add preliminary support for RTL8168G, RTL8168GU and RTL8411B. RTL8168GU has two variants(GMII and MII) but it uses the same chip revision id. Driver checks PCI device id of controller and sets internal capability flag(i.e. jumbo frame and link speed down in WOL). H/W donated by: RealTek Semiconductor Corp.
2013-10-29 05:30:21 +00:00
{ RL_HWREV_8168G, RL_8169, "8168G/8111G", RL_JUMBO_MTU_9K},
{ RL_HWREV_8168GU, RL_8169, "8168GU/8111GU", RL_JUMBO_MTU_9K},
Add preliminary support for RTL8168H, tested by Sreenath Battalahalli. MFC after: 1 week
2015-09-13 22:22:30 +00:00
{ RL_HWREV_8168H, RL_8169, "8168H/8111H", RL_JUMBO_MTU_9K},
Add preliminary support for RTL8411 PCIe Gigabit ethernet with integrated card reader. H/W donated by: RealTek Semiconductor Corp.
2011-11-16 22:05:38 +00:00
{ RL_HWREV_8411, RL_8169, "8411", RL_JUMBO_MTU_9K},
Add preliminary support for RTL8168G, RTL8168GU and RTL8411B. RTL8168GU has two variants(GMII and MII) but it uses the same chip revision id. Driver checks PCI device id of controller and sets internal capability flag(i.e. jumbo frame and link speed down in WOL). H/W donated by: RealTek Semiconductor Corp.
2013-10-29 05:30:21 +00:00
{ RL_HWREV_8411B, RL_8169, "8411B", RL_JUMBO_MTU_9K},
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
{ 0, 0, NULL, 0 }
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
};
static int re_probe (device_t);
static int re_attach (device_t);
static int re_detach (device_t);
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
static int re_encap (struct rl_softc *, struct mbuf **);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
static void re_dma_map_addr (void *, bus_dma_segment_t *, int, int);
static int re_allocmem (device_t, struct rl_softc *);
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
static __inline void re_discard_rxbuf
(struct rl_softc *, int);
static int re_newbuf (struct rl_softc *, int);
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
static int re_jumbo_newbuf (struct rl_softc *, int);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
static int re_rx_list_init (struct rl_softc *);
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
static int re_jrx_list_init (struct rl_softc *);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
static int re_tx_list_init (struct rl_softc *);
fix jumbo frames as much as they can be fixed for re. We now cap the MTU to 7422 since it appears that the 8169S can't transmit anything larger.. The 8169S can receive full jumbo frames, but we don't have an mru to let the upper layers know this... add fixup so that this driver should work on alignment constrained platforms (!i386 && !amd64) MFC after: 5 days
2004-09-28 18:22:24 +00:00
#ifdef RE_FIXUP_RX
static __inline void re_fixup_rx
(struct mbuf *);
#endif
When user_frac in the polling subsystem is low it is going to busy the CPU for too long period than necessary. Additively, interfaces are kept polled (in the tick) even if no more packets are available. In order to avoid such situations a new generic mechanism can be implemented in proactive way, keeping track of the time spent on any packet and fragmenting the time for any tick, stopping the processing as soon as possible. In order to implement such mechanism, the polling handler needs to change, returning the number of packets processed. While the intended logic is not part of this patch, the polling KPI is broken by this commit, adding an int return value and the new flag IFCAP_POLLING_NOCOUNT (which will signal that the return value is meaningless for the installed handler and checking should be skipped). Bump __FreeBSD_version in order to signal such situation. Reviewed by: emaste Sponsored by: Sandvine Incorporated
2009-05-30 15:14:44 +00:00
static int re_rxeof (struct rl_softc *, int *);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
static void re_txeof (struct rl_softc *);
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
#ifdef DEVICE_POLLING
When user_frac in the polling subsystem is low it is going to busy the CPU for too long period than necessary. Additively, interfaces are kept polled (in the tick) even if no more packets are available. In order to avoid such situations a new generic mechanism can be implemented in proactive way, keeping track of the time spent on any packet and fragmenting the time for any tick, stopping the processing as soon as possible. In order to implement such mechanism, the polling handler needs to change, returning the number of packets processed. While the intended logic is not part of this patch, the polling KPI is broken by this commit, adding an int return value and the new flag IFCAP_POLLING_NOCOUNT (which will signal that the return value is meaningless for the installed handler and checking should be skipped). Bump __FreeBSD_version in order to signal such situation. Reviewed by: emaste Sponsored by: Sandvine Incorporated
2009-05-30 15:14:44 +00:00
static int re_poll (struct ifnet *, enum poll_cmd, int);
static int re_poll_locked (struct ifnet *, enum poll_cmd, int);
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
#endif
o break newbus api: add a new argument of type driver_filter_t to bus_setup_intr() o add an int return code to all fast handlers o retire INTR_FAST/IH_FAST For more info: http://docs.freebsd.org/cgi/getmsg.cgi?fetch=465712+0+current/freebsd-current Reviewed by: many Approved by: re@
2007-02-23 12:19:07 +00:00
static int re_intr (void *);
Do not use interrupt taskqueue on controllers with MSI/MSI-X capability. One of reason using interrupt taskqueue in re(4) was to reduce number of TX/RX interrupts under load because re(4) controllers have no good TX/RX interrupt moderation mechanism. Basic TX interrupt moderation is done by hardware for most controllers but RX interrupt moderation through undocumented register showed poor RX performance so it was disabled in r215025. Using taskqueue to handle RX interrupt greatly reduced number of interrupts but re(4) consumed all available CPU cycles to run the taskqueue under high TX/RX network load. This can happen even with RTL810x fast ethernet controller and I believe this is not acceptable for most systems. To mitigate the issue, use one-shot timer register to moderate RX interrupts. The timer register provides programmable one-shot timer and can be used to suppress interrupt generation. The timer runs at 125MHZ on PCIe controllers so the minimum time allowed for the timer is 8ns. Data sheet says the register is 32 bits but experimentation shows only lower 13 bits are valid so maximum time that can be programmed is 65.528us. This yields theoretical maximum number of RX interrupts that could be generated per second is about 15260. Combined with TX completion interrupts re(4) shall generate less than 20k interrupts. This number is still slightly high compared to other intelligent ethernet controllers but system is very responsive even under high network load. Introduce sysctl variable dev.re.%d.int_rx_mod that controls amount of time to delay RX interrupt processing in units of us. Value 0 completely disables RX interrupt moderation. To provide old behavior for controllers that have MSI/MSI-X capability, introduce a new tunable hw.re.intr_filter. If the tunable is set to non-zero value, driver will use interrupt taskqueue. The default value of the tunable is 0. This tunable has no effect on controllers that has no MSI/MSI-X capability or if MSI/MSI-X is explicitly disabled by administrator. While I'm here cleanup interrupt setup/teardown since re(4) uses single MSI/MSI-X message at this moment.
2011-01-26 20:25:40 +00:00
static void re_intr_msi (void *);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
static void re_tick (void *);
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
static void re_int_task (void *, int);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
static void re_start (struct ifnet *);
Remove TX taskqueue and directly invoke re_start in interrupt task.
2011-01-25 23:27:28 +00:00
static void re_start_locked (struct ifnet *);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
static int re_ioctl (struct ifnet *, u_long, caddr_t);
static void re_init (void *);
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
static void re_init_locked (struct rl_softc *);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
static void re_stop (struct rl_softc *);
- Use the re_tick() callout instead of if_slowtimo() for driving re_watchdog() in order to avoid races accessing if_timer. - Use bus_get_dma_tag() so re(4) works on platforms requiring it. - Remove invalid BUS_DMA_ALLOCNOW when creating the parent DMA tag and the tags that are used for static memory allocations. - Don't bother to set if_mtu to ETHERMTU, ether_ifattach() does that. - Remove an unused variable in re_intr().
2006-12-20 02:13:59 +00:00
static void re_watchdog (struct rl_softc *);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
static int re_suspend (device_t);
static int re_resume (device_t);
Fix function prototype for device_shutdown method.
2007-11-22 02:45:00 +00:00
static int re_shutdown (device_t);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
static int re_ifmedia_upd (struct ifnet *);
static void re_ifmedia_sts (struct ifnet *, struct ifmediareq *);
static void re_eeprom_putbyte (struct rl_softc *, int);
static void re_eeprom_getword (struct rl_softc *, int, u_int16_t *);
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
static void re_read_eeprom (struct rl_softc *, caddr_t, int, int);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
static int re_gmii_readreg (device_t, int, int);
static int re_gmii_writereg (device_t, int, int, int);
static int re_miibus_readreg (device_t, int, int);
static int re_miibus_writereg (device_t, int, int, int);
static void re_miibus_statchg (device_t);
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
static void re_set_jumbo (struct rl_softc *, int);
- Do not read and write RX configuration register multiple times. - Always program RX configuration register from scratch instead of doing read/modify/write. - Rename re_setmulti() to re_set_rxmode() to be reflect reality. - Simplify hash filter logic a little while I am here. Reviewed by: yongari (early version)
2009-01-20 19:58:01 +00:00
static void re_set_rxmode (struct rl_softc *);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
static void re_reset (struct rl_softc *);
Implement WOL. Tested by: Fabian Keil ( freebsd-listen AT fabienkeli DOT de )
2008-03-03 03:33:58 +00:00
static void re_setwol (struct rl_softc *);
static void re_clrwol (struct rl_softc *);
To save more power, switch to 10/100Mbps link when controller is put into suspend/shutdown. Old PCI controllers performed that operation in firmware but for RTL8111C or newer controllers, it's responsibility of driver. It's not clear whether the firmware of RTL8111B still downgrades its speed to 10/100Mbps so leave it as it was.
2011-11-23 23:29:18 +00:00
static void re_set_linkspeed (struct rl_softc *);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#ifdef DEV_NETMAP /* see ixgbe.c for details */
#include <dev/netmap/if_re_netmap.h>
Sync netmap sources with the version in our private tree. This commit contains large contributions from Giuseppe Lettieri and Stefano Garzarella, is partly supported by grants from Verisign and Cisco, and brings in the following: - fix zerocopy monitor ports and introduce copying monitor ports (the latter are lower performance but give access to all traffic in parallel with the application) - exclusive open mode, useful to implement solutions that recover from crashes of the main netmap client (suggested by Patrick Kelsey) - revised memory allocator in preparation for the 'passthrough mode' (ptnetmap) recently presented at bsdcan. ptnetmap is described in S. Garzarella, G. Lettieri, L. Rizzo; Virtual device passthrough for high speed VM networking, ACM/IEEE ANCS 2015, Oakland (CA) May 2015 http://info.iet.unipi.it/~luigi/research.html - fix rx CRC handing on ixl - add module dependencies for netmap when building drivers as modules - minor simplifications to device-specific routines (*txsync, *rxsync) - general code cleanup (remove unused variables, introduce macros to access rings and remove duplicate code, Applications do not need to be recompiled, unless of course they want to use the new features (monitors and exclusive open). Those willing to try this code on stable/10 can just update the sys/dev/netmap/*, sys/net/netmap* with the version in HEAD and apply the small patches to individual device drivers. MFC after: 1 month Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
MODULE_DEPEND(re, netmap, 1, 1, 1);
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#endif /* !DEV_NETMAP */
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
#ifdef RE_DIAG
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
static int re_diag (struct rl_softc *);
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
#endif
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Add simple MAC statistics counter reading support. Unfortunately useful counters like rl_missed_pkts is 16 bits quantity which is too small to hold meaningful information happened in a second. This means driver should frequently read these counters in order not to lose accuracy and that approach is too inefficient in driver's view. Moreover it seems there is no way to trigger an interrupt to detect counter near-full or wraparound event as well as lacking clearing the MAC counters. Another limitation of reading the counters from RealTek controllers is lack of interrupt firing at the end of DMA cycle of MAC counter read request such that driver have to poll the end of the DMA which is a time consuming process as well as inefficient. The more severe issue of the MAC counter read request is it takes too long to complete the DMA. All these limitation made maintaining MAC counters in driver impractical. For now, just provide simple sysctl interface to trigger reading the MAC counters. These counters could be used to track down driver issues. Users can read MAC counters maintained in controller with the following command. #sysctl dev.re.0.stats=1 While I'm here add check for validity of dma map and allocated memory before unloading/freeing them. Tested by: rmacklem
2010-11-05 19:28:00 +00:00
static void re_add_sysctls (struct rl_softc *);
static int re_sysctl_stats (SYSCTL_HANDLER_ARGS);
Do not use interrupt taskqueue on controllers with MSI/MSI-X capability. One of reason using interrupt taskqueue in re(4) was to reduce number of TX/RX interrupts under load because re(4) controllers have no good TX/RX interrupt moderation mechanism. Basic TX interrupt moderation is done by hardware for most controllers but RX interrupt moderation through undocumented register showed poor RX performance so it was disabled in r215025. Using taskqueue to handle RX interrupt greatly reduced number of interrupts but re(4) consumed all available CPU cycles to run the taskqueue under high TX/RX network load. This can happen even with RTL810x fast ethernet controller and I believe this is not acceptable for most systems. To mitigate the issue, use one-shot timer register to moderate RX interrupts. The timer register provides programmable one-shot timer and can be used to suppress interrupt generation. The timer runs at 125MHZ on PCIe controllers so the minimum time allowed for the timer is 8ns. Data sheet says the register is 32 bits but experimentation shows only lower 13 bits are valid so maximum time that can be programmed is 65.528us. This yields theoretical maximum number of RX interrupts that could be generated per second is about 15260. Combined with TX completion interrupts re(4) shall generate less than 20k interrupts. This number is still slightly high compared to other intelligent ethernet controllers but system is very responsive even under high network load. Introduce sysctl variable dev.re.%d.int_rx_mod that controls amount of time to delay RX interrupt processing in units of us. Value 0 completely disables RX interrupt moderation. To provide old behavior for controllers that have MSI/MSI-X capability, introduce a new tunable hw.re.intr_filter. If the tunable is set to non-zero value, driver will use interrupt taskqueue. The default value of the tunable is 0. This tunable has no effect on controllers that has no MSI/MSI-X capability or if MSI/MSI-X is explicitly disabled by administrator. While I'm here cleanup interrupt setup/teardown since re(4) uses single MSI/MSI-X message at this moment.
2011-01-26 20:25:40 +00:00
static int sysctl_int_range (SYSCTL_HANDLER_ARGS, int, int);
static int sysctl_hw_re_int_mod (SYSCTL_HANDLER_ARGS);
Add simple MAC statistics counter reading support. Unfortunately useful counters like rl_missed_pkts is 16 bits quantity which is too small to hold meaningful information happened in a second. This means driver should frequently read these counters in order not to lose accuracy and that approach is too inefficient in driver's view. Moreover it seems there is no way to trigger an interrupt to detect counter near-full or wraparound event as well as lacking clearing the MAC counters. Another limitation of reading the counters from RealTek controllers is lack of interrupt firing at the end of DMA cycle of MAC counter read request such that driver have to poll the end of the DMA which is a time consuming process as well as inefficient. The more severe issue of the MAC counter read request is it takes too long to complete the DMA. All these limitation made maintaining MAC counters in driver impractical. For now, just provide simple sysctl interface to trigger reading the MAC counters. These counters could be used to track down driver issues. Users can read MAC counters maintained in controller with the following command. #sysctl dev.re.0.stats=1 While I'm here add check for validity of dma map and allocated memory before unloading/freeing them. Tested by: rmacklem
2010-11-05 19:28:00 +00:00
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
static device_method_t re_methods[] = {
/* Device interface */
DEVMETHOD(device_probe, re_probe),
DEVMETHOD(device_attach, re_attach),
DEVMETHOD(device_detach, re_detach),
DEVMETHOD(device_suspend, re_suspend),
DEVMETHOD(device_resume, re_resume),
DEVMETHOD(device_shutdown, re_shutdown),
/* MII interface */
DEVMETHOD(miibus_readreg, re_miibus_readreg),
DEVMETHOD(miibus_writereg, re_miibus_writereg),
DEVMETHOD(miibus_statchg, re_miibus_statchg),
- There's no need to overwrite the default device method with the default one. Interestingly, these are actually the default for quite some time (bus_generic_driver_added(9) since r52045 and bus_generic_print_child(9) since r52045) but even recently added device drivers do this unnecessarily. Discussed with: jhb, marcel - While at it, use DEVMETHOD_END. Discussed with: jhb - Also while at it, use __FBSDID.
2011-11-22 21:28:20 +00:00
DEVMETHOD_END
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
};
static driver_t re_driver = {
"re",
re_methods,
sizeof(struct rl_softc)
};
static devclass_t re_devclass;
DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0);
DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0);
#define EE_SET(x) \
CSR_WRITE_1(sc, RL_EECMD, \
CSR_READ_1(sc, RL_EECMD) | x)
#define EE_CLR(x) \
CSR_WRITE_1(sc, RL_EECMD, \
CSR_READ_1(sc, RL_EECMD) & ~x)
/*
* Send a read command and address to the EEPROM, check for ACK.
*/
static void
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_eeprom_putbyte(struct rl_softc *sc, int addr)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
Remove register keyword.
2008-08-04 03:47:29 +00:00
int d, i;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
d = addr | (RL_9346_READ << sc->rl_eewidth);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/*
* Feed in each bit and strobe the clock.
*/
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) {
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
if (d & i) {
EE_SET(RL_EE_DATAIN);
} else {
EE_CLR(RL_EE_DATAIN);
}
DELAY(100);
EE_SET(RL_EE_CLK);
DELAY(150);
EE_CLR(RL_EE_CLK);
DELAY(100);
}
}
/*
* Read a word of data stored in the EEPROM at address 'addr.'
*/
static void
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_eeprom_getword(struct rl_softc *sc, int addr, u_int16_t *dest)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
Remove register keyword.
2008-08-04 03:47:29 +00:00
int i;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
u_int16_t word = 0;
/*
* Send address of word we want to read.
*/
re_eeprom_putbyte(sc, addr);
/*
* Start reading bits from EEPROM.
*/
for (i = 0x8000; i; i >>= 1) {
EE_SET(RL_EE_CLK);
DELAY(100);
if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT)
word |= i;
EE_CLR(RL_EE_CLK);
DELAY(100);
}
*dest = word;
}
/*
* Read a sequence of words from the EEPROM.
*/
static void
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_read_eeprom(struct rl_softc *sc, caddr_t dest, int off, int cnt)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
int i;
u_int16_t word = 0, *ptr;
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
DELAY(100);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
for (i = 0; i < cnt; i++) {
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
re_eeprom_getword(sc, off + i, &word);
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
ptr = (u_int16_t *)(dest + (i * 2));
Fix re(4) breakge introduced in tree from rev 1.68. This should fix incorrect configuration of station address on big-endian architectures. Reviewed by: wpaul Tested on: sparc64
2006-08-03 00:15:19 +00:00
*ptr = word;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
static int
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_gmii_readreg(device_t dev, int phy, int reg)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
struct rl_softc *sc;
u_int32_t rval;
int i;
sc = device_get_softc(dev);
Add a PHY driver to support the built-in gigE PHY in the 8169S/8110S ethernet chips. This driver is pretty simple, however it contains special DSP initialization code which is needed in order to get the chip to negotiate a gigE link. (This special initialization may not be needed in subsequent chip revs.) Also: - Fix typo in if_rlreg.h (RL_GMEDIASTAT_1000MPS -> RL_GMEDIASTAT_1000MBPS) - Deal with shared interrupts in re_intr(): if interface isn't up, return. - Fix another bug in re_gmii_writereg() (properly apply data field mask) - Allow PHY driver to read the RL_GMEDIASTAT register via the re_gmii_readreg() register (this is register needed to determine real time link/media status).
2003-09-11 03:53:46 +00:00
/* Let the rgephy driver read the GMEDIASTAT register */
if (reg == RL_GMEDIASTAT) {
rval = CSR_READ_1(sc, RL_GMEDIASTAT);
Whitespace pass
2004-07-06 02:46:53 +00:00
return (rval);
Add a PHY driver to support the built-in gigE PHY in the 8169S/8110S ethernet chips. This driver is pretty simple, however it contains special DSP initialization code which is needed in order to get the chip to negotiate a gigE link. (This special initialization may not be needed in subsequent chip revs.) Also: - Fix typo in if_rlreg.h (RL_GMEDIASTAT_1000MPS -> RL_GMEDIASTAT_1000MBPS) - Deal with shared interrupts in re_intr(): if interface isn't up, return. - Fix another bug in re_gmii_writereg() (properly apply data field mask) - Allow PHY driver to read the RL_GMEDIASTAT register via the re_gmii_readreg() register (this is register needed to determine real time link/media status).
2003-09-11 03:53:46 +00:00
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
Sometimes RTL8168B seems to take long time to access GMII registers in device attach phase. Double GMII register access timeout value to fix the issue. Reported by: wkoszek Tested by: wkoszek
2009-01-19 02:31:27 +00:00
for (i = 0; i < RL_PHY_TIMEOUT; i++) {
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
rval = CSR_READ_4(sc, RL_PHYAR);
if (rval & RL_PHYAR_BUSY)
break;
Reduce spin wait time consumed in GMII register access routine. There were a couple of attempts in the past to reduce it since it took more than 1ms. Because mii_tick() periodically polls link status, waiting more than 1ms for each GMII register access was overkill. Unfortunately all previous attempts were failed with various ways on different controllers. This time, add additional 20us dealy at the end of GMII register access which seems to requirement of all RealTek controllers to issue next GMII register access request. This is the same way what Linux does.
2010-11-08 19:15:31 +00:00
DELAY(25);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
Sometimes RTL8168B seems to take long time to access GMII registers in device attach phase. Double GMII register access timeout value to fix the issue. Reported by: wkoszek Tested by: wkoszek
2009-01-19 02:31:27 +00:00
if (i == RL_PHY_TIMEOUT) {
- Consistently use if_printf() only in interface methods: if_start(), if_watchdog, etc., or in functions used only in these methods. In all other functions in the driver use device_printf(). - Use __func__ instead of typing function name. Submitted by: Alex Lyashkov <umka sevcity.net>
2006-09-15 15:16:12 +00:00
device_printf(sc->rl_dev, "PHY read failed\n");
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
return (0);
}
Reduce spin wait time consumed in GMII register access routine. There were a couple of attempts in the past to reduce it since it took more than 1ms. Because mii_tick() periodically polls link status, waiting more than 1ms for each GMII register access was overkill. Unfortunately all previous attempts were failed with various ways on different controllers. This time, add additional 20us dealy at the end of GMII register access which seems to requirement of all RealTek controllers to issue next GMII register access request. This is the same way what Linux does.
2010-11-08 19:15:31 +00:00
/*
* Controller requires a 20us delay to process next MDIO request.
*/
DELAY(20);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
return (rval & RL_PHYAR_PHYDATA);
}
static int
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_gmii_writereg(device_t dev, int phy, int reg, int data)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
struct rl_softc *sc;
u_int32_t rval;
int i;
sc = device_get_softc(dev);
CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) |
Add a PHY driver to support the built-in gigE PHY in the 8169S/8110S ethernet chips. This driver is pretty simple, however it contains special DSP initialization code which is needed in order to get the chip to negotiate a gigE link. (This special initialization may not be needed in subsequent chip revs.) Also: - Fix typo in if_rlreg.h (RL_GMEDIASTAT_1000MPS -> RL_GMEDIASTAT_1000MBPS) - Deal with shared interrupts in re_intr(): if interface isn't up, return. - Fix another bug in re_gmii_writereg() (properly apply data field mask) - Allow PHY driver to read the RL_GMEDIASTAT register via the re_gmii_readreg() register (this is register needed to determine real time link/media status).
2003-09-11 03:53:46 +00:00
(data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Sometimes RTL8168B seems to take long time to access GMII registers in device attach phase. Double GMII register access timeout value to fix the issue. Reported by: wkoszek Tested by: wkoszek
2009-01-19 02:31:27 +00:00
for (i = 0; i < RL_PHY_TIMEOUT; i++) {
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
rval = CSR_READ_4(sc, RL_PHYAR);
if (!(rval & RL_PHYAR_BUSY))
break;
Reduce spin wait time consumed in GMII register access routine. There were a couple of attempts in the past to reduce it since it took more than 1ms. Because mii_tick() periodically polls link status, waiting more than 1ms for each GMII register access was overkill. Unfortunately all previous attempts were failed with various ways on different controllers. This time, add additional 20us dealy at the end of GMII register access which seems to requirement of all RealTek controllers to issue next GMII register access request. This is the same way what Linux does.
2010-11-08 19:15:31 +00:00
DELAY(25);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
Sometimes RTL8168B seems to take long time to access GMII registers in device attach phase. Double GMII register access timeout value to fix the issue. Reported by: wkoszek Tested by: wkoszek
2009-01-19 02:31:27 +00:00
if (i == RL_PHY_TIMEOUT) {
- Consistently use if_printf() only in interface methods: if_start(), if_watchdog, etc., or in functions used only in these methods. In all other functions in the driver use device_printf(). - Use __func__ instead of typing function name. Submitted by: Alex Lyashkov <umka sevcity.net>
2006-09-15 15:16:12 +00:00
device_printf(sc->rl_dev, "PHY write failed\n");
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
return (0);
}
Reduce spin wait time consumed in GMII register access routine. There were a couple of attempts in the past to reduce it since it took more than 1ms. Because mii_tick() periodically polls link status, waiting more than 1ms for each GMII register access was overkill. Unfortunately all previous attempts were failed with various ways on different controllers. This time, add additional 20us dealy at the end of GMII register access which seems to requirement of all RealTek controllers to issue next GMII register access request. This is the same way what Linux does.
2010-11-08 19:15:31 +00:00
/*
* Controller requires a 20us delay to process next MDIO request.
*/
DELAY(20);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
return (0);
}
static int
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_miibus_readreg(device_t dev, int phy, int reg)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
struct rl_softc *sc;
u_int16_t rval = 0;
u_int16_t re8139_reg = 0;
sc = device_get_softc(dev);
if (sc->rl_type == RL_8169) {
rval = re_gmii_readreg(dev, phy, reg);
return (rval);
}
Whitespace pass
2004-07-06 02:46:53 +00:00
switch (reg) {
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
case MII_BMCR:
re8139_reg = RL_BMCR;
break;
case MII_BMSR:
re8139_reg = RL_BMSR;
break;
case MII_ANAR:
re8139_reg = RL_ANAR;
break;
case MII_ANER:
re8139_reg = RL_ANER;
break;
case MII_ANLPAR:
re8139_reg = RL_LPAR;
break;
case MII_PHYIDR1:
case MII_PHYIDR2:
Whitespace pass
2004-07-06 02:46:53 +00:00
return (0);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/*
* Allow the rlphy driver to read the media status
* register. If we have a link partner which does not
* support NWAY, this is the register which will tell
* us the results of parallel detection.
*/
case RL_MEDIASTAT:
rval = CSR_READ_1(sc, RL_MEDIASTAT);
Whitespace pass
2004-07-06 02:46:53 +00:00
return (rval);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
default:
- Consistently use if_printf() only in interface methods: if_start(), if_watchdog, etc., or in functions used only in these methods. In all other functions in the driver use device_printf(). - Use __func__ instead of typing function name. Submitted by: Alex Lyashkov <umka sevcity.net>
2006-09-15 15:16:12 +00:00
device_printf(sc->rl_dev, "bad phy register\n");
Whitespace pass
2004-07-06 02:46:53 +00:00
return (0);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
rval = CSR_READ_2(sc, re8139_reg);
Make 8139C+ work again which was broken since rev 1.68. Ever since rev 1.68 re(4) checks the validity of link in re_start. But rlphy(4) got a garbled data due to a different bit layout used on 8139C+ and it couldn't report correct link state. To fix it, ignore BMCR_LOOP and BMCR_ISO bits which have different meanings on 8139C+. I think this also make dhclient(8) work on 8139C+. Reported by: Gerrit Kuehn <gerrit AT pmp DOT uni-hannover DOT de> Tested by: Gerrit Kuehn <gerrit AT pmp DOT uni-hannover DOT de>
2006-09-08 00:58:02 +00:00
if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
/* 8139C+ has different bit layout. */
rval &= ~(BMCR_LOOP | BMCR_ISO);
}
Whitespace pass
2004-07-06 02:46:53 +00:00
return (rval);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
static int
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_miibus_writereg(device_t dev, int phy, int reg, int data)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
struct rl_softc *sc;
u_int16_t re8139_reg = 0;
int rval = 0;
sc = device_get_softc(dev);
if (sc->rl_type == RL_8169) {
rval = re_gmii_writereg(dev, phy, reg, data);
return (rval);
}
Whitespace pass
2004-07-06 02:46:53 +00:00
switch (reg) {
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
case MII_BMCR:
re8139_reg = RL_BMCR;
Make 8139C+ work again which was broken since rev 1.68. Ever since rev 1.68 re(4) checks the validity of link in re_start. But rlphy(4) got a garbled data due to a different bit layout used on 8139C+ and it couldn't report correct link state. To fix it, ignore BMCR_LOOP and BMCR_ISO bits which have different meanings on 8139C+. I think this also make dhclient(8) work on 8139C+. Reported by: Gerrit Kuehn <gerrit AT pmp DOT uni-hannover DOT de> Tested by: Gerrit Kuehn <gerrit AT pmp DOT uni-hannover DOT de>
2006-09-08 00:58:02 +00:00
if (sc->rl_type == RL_8139CPLUS) {
/* 8139C+ has different bit layout. */
data &= ~(BMCR_LOOP | BMCR_ISO);
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
break;
case MII_BMSR:
re8139_reg = RL_BMSR;
break;
case MII_ANAR:
re8139_reg = RL_ANAR;
break;
case MII_ANER:
re8139_reg = RL_ANER;
break;
case MII_ANLPAR:
re8139_reg = RL_LPAR;
break;
case MII_PHYIDR1:
case MII_PHYIDR2:
Whitespace pass
2004-07-06 02:46:53 +00:00
return (0);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
break;
default:
- Consistently use if_printf() only in interface methods: if_start(), if_watchdog, etc., or in functions used only in these methods. In all other functions in the driver use device_printf(). - Use __func__ instead of typing function name. Submitted by: Alex Lyashkov <umka sevcity.net>
2006-09-15 15:16:12 +00:00
device_printf(sc->rl_dev, "bad phy register\n");
Whitespace pass
2004-07-06 02:46:53 +00:00
return (0);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
CSR_WRITE_2(sc, re8139_reg, data);
Whitespace pass
2004-07-06 02:46:53 +00:00
return (0);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
static void
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_miibus_statchg(device_t dev)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
o Implemented miibus_statchg handler. It detects whether re(4) established a valid link or not. In miibus_statchg handler add a check for established link is valid one for the controller(e.g. 1000baseT is not a valid link for fastethernet controllers.) o Added a flag RE_FLAG_FASTETHER to mark fastethernet controllers. o Added additional check to know whether we've really encountered watchdog timeouts or missed Tx completion interrupts. This change may help to track down the cause of watchdog timeouts. o In interrupt handler, removed a check for link state change interrupt. Not all controllers have the bit and re(4) did not rely on the event for a long time. In addition, re(4) didn't request the interrupt in RL_IMR register. Tested by: rpaulo
2008-12-08 02:34:13 +00:00
struct rl_softc *sc;
struct ifnet *ifp;
struct mii_data *mii;
Eliminate unneeded return keywords.
2004-07-06 02:48:29 +00:00
o Implemented miibus_statchg handler. It detects whether re(4) established a valid link or not. In miibus_statchg handler add a check for established link is valid one for the controller(e.g. 1000baseT is not a valid link for fastethernet controllers.) o Added a flag RE_FLAG_FASTETHER to mark fastethernet controllers. o Added additional check to know whether we've really encountered watchdog timeouts or missed Tx completion interrupts. This change may help to track down the cause of watchdog timeouts. o In interrupt handler, removed a check for link state change interrupt. Not all controllers have the bit and re(4) did not rely on the event for a long time. In addition, re(4) didn't request the interrupt in RL_IMR register. Tested by: rpaulo
2008-12-08 02:34:13 +00:00
sc = device_get_softc(dev);
mii = device_get_softc(sc->rl_miibus);
ifp = sc->rl_ifp;
if (mii == NULL || ifp == NULL ||
(ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
return;
sc->rl_flags &= ~RL_FLAG_LINK;
if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
(IFM_ACTIVE | IFM_AVALID)) {
switch (IFM_SUBTYPE(mii->mii_media_active)) {
case IFM_10_T:
case IFM_100_TX:
sc->rl_flags |= RL_FLAG_LINK;
break;
case IFM_1000_T:
if ((sc->rl_flags & RL_FLAG_FASTETHER) != 0)
break;
sc->rl_flags |= RL_FLAG_LINK;
break;
default:
break;
}
}
/*
* RealTek controllers does not provide any interface to
* Tx/Rx MACs for resolved speed, duplex and flow-control
* parameters.
*/
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
/*
- Do not read and write RX configuration register multiple times. - Always program RX configuration register from scratch instead of doing read/modify/write. - Rename re_setmulti() to re_set_rxmode() to be reflect reality. - Simplify hash filter logic a little while I am here. Reviewed by: yongari (early version)
2009-01-20 19:58:01 +00:00
* Set the RX configuration and 64-bit multicast hash filter.
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
*/
static void
- Do not read and write RX configuration register multiple times. - Always program RX configuration register from scratch instead of doing read/modify/write. - Rename re_setmulti() to re_set_rxmode() to be reflect reality. - Simplify hash filter logic a little while I am here. Reviewed by: yongari (early version)
2009-01-20 19:58:01 +00:00
re_set_rxmode(struct rl_softc *sc)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
struct ifnet *ifp;
struct ifmultiaddr *ifma;
- Do not read and write RX configuration register multiple times. - Always program RX configuration register from scratch instead of doing read/modify/write. - Rename re_setmulti() to re_set_rxmode() to be reflect reality. - Simplify hash filter logic a little while I am here. Reviewed by: yongari (early version)
2009-01-20 19:58:01 +00:00
uint32_t hashes[2] = { 0, 0 };
uint32_t h, rxfilt;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
RL_LOCK_ASSERT(sc);
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
ifp = sc->rl_ifp;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
- Do not read and write RX configuration register multiple times. - Always program RX configuration register from scratch instead of doing read/modify/write. - Rename re_setmulti() to re_set_rxmode() to be reflect reality. - Simplify hash filter logic a little while I am here. Reviewed by: yongari (early version)
2009-01-20 19:58:01 +00:00
rxfilt = RL_RXCFG_CONFIG | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_BROAD;
- Implement the RX EARLYOFF and RXDV GATED bits as done by RealTek's Linux driver as version 8.037.00 for RTL8168{E-VL,EP,F,G,GU} and RTL8111B. This makes reception of packets work with the RTL8168G (HW rev. 0x4c000000) in my Shuttle DS47. - Consistently use RL_MSI_MESSAGES. In joint forces with: yongari MFC after: 5 days
2014-02-05 23:13:40 +00:00
if ((sc->rl_flags & RL_FLAG_EARLYOFF) != 0)
rxfilt |= RL_RXCFG_EARLYOFF;
else if ((sc->rl_flags & RL_FLAG_EARLYOFFV2) != 0)
rxfilt |= RL_RXCFG_EARLYOFFV2;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
- Do not read and write RX configuration register multiple times. - Always program RX configuration register from scratch instead of doing read/modify/write. - Rename re_setmulti() to re_set_rxmode() to be reflect reality. - Simplify hash filter logic a little while I am here. Reviewed by: yongari (early version)
2009-01-20 19:58:01 +00:00
if (ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) {
Always honor promiscuous flag prior to programming Rx multicast filter. This fixes a regression introduced in rev 1.89. PR: 114632 MFC after: 3 days
2007-12-03 01:28:08 +00:00
if (ifp->if_flags & IFF_PROMISC)
rxfilt |= RL_RXCFG_RX_ALLPHYS;
By definition promiscuous mode should see all unicast frames as well as multicast/broadcast frames. Previously re(4) ignored multicast frames in promiscuous mode. The RTL8169 datasheet was not clear how it handles multicast frames in promiscuous mode. PR: kern/118572 MFC after: 3 days
2007-12-20 07:26:20 +00:00
/*
* Unlike other hardwares, we have to explicitly set
* RL_RXCFG_RX_MULTI to receive multicast frames in
* promiscuous mode.
*/
rxfilt |= RL_RXCFG_RX_MULTI;
- Do not read and write RX configuration register multiple times. - Always program RX configuration register from scratch instead of doing read/modify/write. - Rename re_setmulti() to re_set_rxmode() to be reflect reality. - Simplify hash filter logic a little while I am here. Reviewed by: yongari (early version)
2009-01-20 19:58:01 +00:00
hashes[0] = hashes[1] = 0xffffffff;
goto done;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
Use if_maddr_rlock()/if_maddr_runlock() rather than IF_ADDR_LOCK()/ IF_ADDR_UNLOCK() across network device drivers when accessing the per-interface multicast address list, if_multiaddrs. This will allow us to change the locking strategy without affecting our driver programming interface or binary interface. For two wireless drivers, remove unnecessary locking, since they don't actually access the multicast address list. Approved by: re (kib) MFC after: 6 weeks
2009-06-26 11:45:06 +00:00
if_maddr_rlock(ifp);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
if (ifma->ifma_addr->sa_family != AF_LINK)
continue;
Replace handrolled CRC calculation with ether_crc32_[lb]e().
2004-06-09 14:34:04 +00:00
h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
if (h < 32)
hashes[0] |= (1 << h);
else
hashes[1] |= (1 << (h - 32));
}
Use if_maddr_rlock()/if_maddr_runlock() rather than IF_ADDR_LOCK()/ IF_ADDR_UNLOCK() across network device drivers when accessing the per-interface multicast address list, if_multiaddrs. This will allow us to change the locking strategy without affecting our driver programming interface or binary interface. For two wireless drivers, remove unnecessary locking, since they don't actually access the multicast address list. Approved by: re (kib) MFC after: 6 weeks
2009-06-26 11:45:06 +00:00
if_maddr_runlock(ifp);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
- Do not read and write RX configuration register multiple times. - Always program RX configuration register from scratch instead of doing read/modify/write. - Rename re_setmulti() to re_set_rxmode() to be reflect reality. - Simplify hash filter logic a little while I am here. Reviewed by: yongari (early version)
2009-01-20 19:58:01 +00:00
if (hashes[0] != 0 || hashes[1] != 0) {
/*
* For some unfathomable reason, RealTek decided to
* reverse the order of the multicast hash registers
* in the PCI Express parts. This means we have to
* write the hash pattern in reverse order for those
* devices.
*/
Retire RL_FLAG_INVMAR bit to match its comment and reality.
2009-01-20 20:04:09 +00:00
if ((sc->rl_flags & RL_FLAG_PCIE) != 0) {
- Do not read and write RX configuration register multiple times. - Always program RX configuration register from scratch instead of doing read/modify/write. - Rename re_setmulti() to re_set_rxmode() to be reflect reality. - Simplify hash filter logic a little while I am here. Reviewed by: yongari (early version)
2009-01-20 19:58:01 +00:00
h = bswap32(hashes[0]);
hashes[0] = bswap32(hashes[1]);
hashes[1] = h;
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
rxfilt |= RL_RXCFG_RX_MULTI;
- Do not read and write RX configuration register multiple times. - Always program RX configuration register from scratch instead of doing read/modify/write. - Rename re_setmulti() to re_set_rxmode() to be reflect reality. - Simplify hash filter logic a little while I am here. Reviewed by: yongari (early version)
2009-01-20 19:58:01 +00:00
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
It seems multicast filtering of RTL8168F does not work. Workaround the silicon bug by accepting any multicast packets. PR: 193488 MFC After: 1 week
2014-10-21 01:31:24 +00:00
if (sc->rl_hwrev->rl_rev == RL_HWREV_8168F) {
/* Disable multicast filtering due to silicon bug. */
hashes[0] = 0xffffffff;
hashes[1] = 0xffffffff;
}
- Do not read and write RX configuration register multiple times. - Always program RX configuration register from scratch instead of doing read/modify/write. - Rename re_setmulti() to re_set_rxmode() to be reflect reality. - Simplify hash filter logic a little while I am here. Reviewed by: yongari (early version)
2009-01-20 19:58:01 +00:00
done:
CSR_WRITE_4(sc, RL_MAR0, hashes[0]);
CSR_WRITE_4(sc, RL_MAR4, hashes[1]);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
CSR_WRITE_4(sc, RL_RXCFG, rxfilt);
}
static void
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_reset(struct rl_softc *sc)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
Remove register keyword.
2008-08-04 03:47:29 +00:00
int i;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
RL_LOCK_ASSERT(sc);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET);
for (i = 0; i < RL_TIMEOUT; i++) {
DELAY(10);
if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET))
break;
}
if (i == RL_TIMEOUT)
- Consistently use if_printf() only in interface methods: if_start(), if_watchdog, etc., or in functions used only in these methods. In all other functions in the driver use device_printf(). - Use __func__ instead of typing function name. Submitted by: Alex Lyashkov <umka sevcity.net>
2006-09-15 15:16:12 +00:00
device_printf(sc->rl_dev, "reset never completed!\n");
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
- Add support for 8110SCe part. Some magic registers were taken from Linux driver. - Swap hardware revisions for 8110S and 8169S as Linux driver claims. Reviewed by: yongari (early version)
2009-01-20 20:22:28 +00:00
if ((sc->rl_flags & RL_FLAG_MACRESET) != 0)
Don't access undocumented register 0x82 on controllers that have no such register. While here clear undocumented PHY register 0x0B for RTL8110S. Obtained from: RealTek FreeBSD driver
2008-12-11 01:41:38 +00:00
CSR_WRITE_1(sc, 0x82, 1);
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
if (sc->rl_hwrev->rl_rev == RL_HWREV_8169S)
- Add support for 8110SCe part. Some magic registers were taken from Linux driver. - Swap hardware revisions for 8110S and 8169S as Linux driver claims. Reviewed by: yongari (early version)
2009-01-20 20:22:28 +00:00
re_gmii_writereg(sc->rl_dev, 1, 0x0b, 0);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
#ifdef RE_DIAG
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/*
* The following routine is designed to test for a defect on some
* 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
* lines connected to the bus, however for a 32-bit only card, they
* should be pulled high. The result of this defect is that the
* NIC will not work right if you plug it into a 64-bit slot: DMA
* operations will be done with 64-bit transfers, which will fail
* because the 64-bit data lines aren't connected.
*
* There's no way to work around this (short of talking a soldering
* iron to the board), however we can detect it. The method we use
* here is to put the NIC into digital loopback mode, set the receiver
* to promiscuous mode, and then try to send a frame. We then compare
* the frame data we sent to what was received. If the data matches,
* then the NIC is working correctly, otherwise we know the user has
* a defective NIC which has been mistakenly plugged into a 64-bit PCI
* slot. In the latter case, there's no way the NIC can work correctly,
* so we print out a message on the console and abort the device attach.
*/
static int
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_diag(struct rl_softc *sc)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
struct ifnet *ifp = sc->rl_ifp;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
struct mbuf *m0;
struct ether_header *eh;
struct rl_desc *cur_rx;
u_int16_t status;
u_int32_t rxstat;
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
int total_len, i, error = 0, phyaddr;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
u_int8_t dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
u_int8_t src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
/* Allocate a single mbuf */
Mechanically substitute flags from historic mbuf allocator with malloc(9) flags in sys/dev.
2012-12-04 09:32:43 +00:00
MGETHDR(m0, M_NOWAIT, MT_DATA);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
if (m0 == NULL)
Whitespace pass
2004-07-06 02:46:53 +00:00
return (ENOBUFS);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
RL_LOCK(sc);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/*
* Initialize the NIC in test mode. This sets the chip up
* so that it can send and receive frames, but performs the
* following special functions:
* - Puts receiver in promiscuous mode
* - Enables digital loopback mode
* - Leaves interrupts turned off
*/
ifp->if_flags |= IFF_PROMISC;
sc->rl_testmode = 1;
Minimize interface reinitialization by checking IFF_DRV_RUNNING flag. This fixes unnecessary interface UP/DOWNs during getting an IP address via DHCP. Tested by: Warren Block ( wblock<> wonkity dot com )
2009-11-19 23:12:44 +00:00
ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
re_init_locked(sc);
Instead of allocating variables for each events/hardware flags, use a dedicated flag that represents controller capabilities/events. This will simplify many part of code that requires different workaround for each controller revisions and will enhance readability. While I'm here move PHY wakeup code up before mii_phy_probe() which seems to help to wake PHY in some cases.
2008-07-02 06:29:43 +00:00
sc->rl_flags |= RL_FLAG_LINK;
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
if (sc->rl_type == RL_8169)
phyaddr = 1;
else
phyaddr = 0;
re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET);
for (i = 0; i < RL_TIMEOUT; i++) {
status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR);
if (!(status & BMCR_RESET))
break;
}
re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP);
CSR_WRITE_2(sc, RL_ISR, RL_INTRS);
Toggle the interface on and off before starting the diag test. This seems to be necessary for the 8139C+ under certain circumstances, and doesn't appear to hurt the other chips. (In the failure case, the packet would be sent through the TX DMA ring but not get echoed back. I suspect this has something to do with the link state changing unexpectedly.)
2003-09-11 07:54:16 +00:00
DELAY(100000);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/* Put some data in the mbuf */
eh = mtod(m0, struct ether_header *);
bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN);
bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN);
eh->ether_type = htons(ETHERTYPE_IP);
m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
In re_diag(), there's no need for us to call re_start() ourselves: IF_HANDOFF() does it for us behind the scenes. Remove the extra call to re_start() otherwise we try to transmit twice. In re_encap(), fix the code that guards against consuming too many descriptors in the TX ring so that it actually works. With the new 8169S chip, I was able to hit a corner case that drained the free descriptor count all the way to 0. This is not supposed to be possible.
2003-09-18 18:32:15 +00:00
/*
* Queue the packet, start transmission.
* Note: IF_HANDOFF() ultimately calls re_start() for us.
*/
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
- For the 8169 chips, read the station address by forcing an EEPROM autoload and then copying the contends of the station address registers. For some reason, reading the EEPROM on the 8169S doesn't work right. This gets around the problem, and allows us to read the station address correctly on the 8169S. - Insert a delay after initiating packet transmition in re_diag() to allow lots of time for the frame to echo back to the host, and wait for both the 'RX complete' and 'timeout expired' bits in the ISR register to be set. - Deal more intelligently with the fact that the frame length field in the RX descriptor is a different width on the 8139C+ than it is on the 8169/8169S/8110S - For the 8169, you have to set bit 17 in the TX config register to enter digital loopback mode, but for the 8139C+, you have to set both bits 17 and 18. Take this into account so that re_diag() works properly for both types of chips.
2003-09-11 06:56:46 +00:00
CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
RL_UNLOCK(sc);
ALTQ support for re(4). Submitted by: Chris Dionissopoulos, Theo Schlossnagle PR: kern/78681 MFC after: 2 weeks
2005-03-12 17:35:37 +00:00
/* XXX: re_diag must not be called when in ALTQ mode */
In re_diag(), there's no need for us to call re_start() ourselves: IF_HANDOFF() does it for us behind the scenes. Remove the extra call to re_start() otherwise we try to transmit twice. In re_encap(), fix the code that guards against consuming too many descriptors in the TX ring so that it actually works. With the new 8169S chip, I was able to hit a corner case that drained the free descriptor count all the way to 0. This is not supposed to be possible.
2003-09-18 18:32:15 +00:00
IF_HANDOFF(&ifp->if_snd, m0, ifp);
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
RL_LOCK(sc);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
m0 = NULL;
/* Wait for it to propagate through the chip */
- For the 8169 chips, read the station address by forcing an EEPROM autoload and then copying the contends of the station address registers. For some reason, reading the EEPROM on the 8169S doesn't work right. This gets around the problem, and allows us to read the station address correctly on the 8169S. - Insert a delay after initiating packet transmition in re_diag() to allow lots of time for the frame to echo back to the host, and wait for both the 'RX complete' and 'timeout expired' bits in the ISR register to be set. - Deal more intelligently with the fact that the frame length field in the RX descriptor is a different width on the 8139C+ than it is on the 8169/8169S/8110S - For the 8169, you have to set bit 17 in the TX config register to enter digital loopback mode, but for the 8139C+, you have to set both bits 17 and 18. Take this into account so that re_diag() works properly for both types of chips.
2003-09-11 06:56:46 +00:00
DELAY(100000);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
for (i = 0; i < RL_TIMEOUT; i++) {
status = CSR_READ_2(sc, RL_ISR);
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
CSR_WRITE_2(sc, RL_ISR, status);
- For the 8169 chips, read the station address by forcing an EEPROM autoload and then copying the contends of the station address registers. For some reason, reading the EEPROM on the 8169S doesn't work right. This gets around the problem, and allows us to read the station address correctly on the 8169S. - Insert a delay after initiating packet transmition in re_diag() to allow lots of time for the frame to echo back to the host, and wait for both the 'RX complete' and 'timeout expired' bits in the ISR register to be set. - Deal more intelligently with the fact that the frame length field in the RX descriptor is a different width on the 8139C+ than it is on the 8169/8169S/8110S - For the 8169, you have to set bit 17 in the TX config register to enter digital loopback mode, but for the 8139C+, you have to set both bits 17 and 18. Take this into account so that re_diag() works properly for both types of chips.
2003-09-11 06:56:46 +00:00
if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) ==
(RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK))
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
break;
DELAY(10);
}
if (i == RL_TIMEOUT) {
- Consistently use if_printf() only in interface methods: if_start(), if_watchdog, etc., or in functions used only in these methods. In all other functions in the driver use device_printf(). - Use __func__ instead of typing function name. Submitted by: Alex Lyashkov <umka sevcity.net>
2006-09-15 15:16:12 +00:00
device_printf(sc->rl_dev,
"diagnostic failed, failed to receive packet in"
" loopback mode\n");
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
error = EIO;
goto done;
}
/*
* The packet should have been dumped into the first
* entry in the RX DMA ring. Grab it from there.
*/
bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
sc->rl_ldata.rl_rx_list_map,
BUS_DMASYNC_POSTREAD);
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
sc->rl_ldata.rl_rx_desc[0].rx_dmamap,
BUS_DMASYNC_POSTREAD);
bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
sc->rl_ldata.rl_rx_desc[0].rx_dmamap);
m0 = sc->rl_ldata.rl_rx_desc[0].rx_m;
sc->rl_ldata.rl_rx_desc[0].rx_m = NULL;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
eh = mtod(m0, struct ether_header *);
cur_rx = &sc->rl_ldata.rl_rx_list[0];
total_len = RL_RXBYTES(cur_rx);
rxstat = le32toh(cur_rx->rl_cmdstat);
if (total_len != ETHER_MIN_LEN) {
- Consistently use if_printf() only in interface methods: if_start(), if_watchdog, etc., or in functions used only in these methods. In all other functions in the driver use device_printf(). - Use __func__ instead of typing function name. Submitted by: Alex Lyashkov <umka sevcity.net>
2006-09-15 15:16:12 +00:00
device_printf(sc->rl_dev,
"diagnostic failed, received short packet\n");
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
error = EIO;
goto done;
}
/* Test that the received packet data matches what we sent. */
if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
ntohs(eh->ether_type) != ETHERTYPE_IP) {
- Consistently use if_printf() only in interface methods: if_start(), if_watchdog, etc., or in functions used only in these methods. In all other functions in the driver use device_printf(). - Use __func__ instead of typing function name. Submitted by: Alex Lyashkov <umka sevcity.net>
2006-09-15 15:16:12 +00:00
device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n");
device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n",
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
dst, ":", src, ":", ETHERTYPE_IP);
- Consistently use if_printf() only in interface methods: if_start(), if_watchdog, etc., or in functions used only in these methods. In all other functions in the driver use device_printf(). - Use __func__ instead of typing function name. Submitted by: Alex Lyashkov <umka sevcity.net>
2006-09-15 15:16:12 +00:00
device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n",
Sprinkle some const.
2011-11-02 23:23:19 +00:00
eh->ether_dhost, ":", eh->ether_shost, ":",
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
ntohs(eh->ether_type));
- Consistently use if_printf() only in interface methods: if_start(), if_watchdog, etc., or in functions used only in these methods. In all other functions in the driver use device_printf(). - Use __func__ instead of typing function name. Submitted by: Alex Lyashkov <umka sevcity.net>
2006-09-15 15:16:12 +00:00
device_printf(sc->rl_dev, "You may have a defective 32-bit "
"NIC plugged into a 64-bit PCI slot.\n");
device_printf(sc->rl_dev, "Please re-install the NIC in a "
"32-bit slot for proper operation.\n");
device_printf(sc->rl_dev, "Read the re(4) man page for more "
"details.\n");
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
error = EIO;
}
done:
/* Turn interface off, release resources */
sc->rl_testmode = 0;
Instead of allocating variables for each events/hardware flags, use a dedicated flag that represents controller capabilities/events. This will simplify many part of code that requires different workaround for each controller revisions and will enhance readability. While I'm here move PHY wakeup code up before mii_phy_probe() which seems to help to wake PHY in some cases.
2008-07-02 06:29:43 +00:00
sc->rl_flags &= ~RL_FLAG_LINK;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
ifp->if_flags &= ~IFF_PROMISC;
re_stop(sc);
if (m0 != NULL)
m_freem(m0);
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
RL_UNLOCK(sc);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
return (error);
}
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
#endif
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/*
* Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device
* IDs against our list and return a device name if we find a match.
*/
static int
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_probe(device_t dev)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
Sprinkle some const.
2011-11-02 23:23:19 +00:00
const struct rl_type *t;
Don't map memory/IO resource in device probe and just use PCI vendor/revision/sub device id of the hardware to probe it. This is the same way as NetBSD does and it enhances readabilty a lot.
2008-03-03 04:15:08 +00:00
uint16_t devid, vendor;
uint16_t revid, sdevid;
int i;
Remove extra white spaces.
2010-11-05 18:23:43 +00:00
Don't map memory/IO resource in device probe and just use PCI vendor/revision/sub device id of the hardware to probe it. This is the same way as NetBSD does and it enhances readabilty a lot.
2008-03-03 04:15:08 +00:00
vendor = pci_get_vendor(dev);
devid = pci_get_device(dev);
revid = pci_get_revid(dev);
sdevid = pci_get_subdevice(dev);
if (vendor == LINKSYS_VENDORID && devid == LINKSYS_DEVICEID_EG1032) {
if (sdevid != LINKSYS_SUBDEVICE_EG1032_REV3) {
Only allow the sk(4) driver to attach to revision 2 of the LinkSys EG1032 cards and teach the re(4) driver to attach to revision 3 cards. Submitted by: Fredrik Lindberg fli+freebsd-current at shapeshifter dot se MFC after: 2 weeks Reviewed by: imp, mdodd
2005-10-14 18:51:30 +00:00
/*
* Only attach to rev. 3 of the Linksys EG1032 adapter.
Don't map memory/IO resource in device probe and just use PCI vendor/revision/sub device id of the hardware to probe it. This is the same way as NetBSD does and it enhances readabilty a lot.
2008-03-03 04:15:08 +00:00
* Rev. 2 is supported by sk(4).
Only allow the sk(4) driver to attach to revision 2 of the LinkSys EG1032 cards and teach the re(4) driver to attach to revision 3 cards. Submitted by: Fredrik Lindberg fli+freebsd-current at shapeshifter dot se MFC after: 2 weeks Reviewed by: imp, mdodd
2005-10-14 18:51:30 +00:00
*/
Don't map memory/IO resource in device probe and just use PCI vendor/revision/sub device id of the hardware to probe it. This is the same way as NetBSD does and it enhances readabilty a lot.
2008-03-03 04:15:08 +00:00
return (ENXIO);
}
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Don't map memory/IO resource in device probe and just use PCI vendor/revision/sub device id of the hardware to probe it. This is the same way as NetBSD does and it enhances readabilty a lot.
2008-03-03 04:15:08 +00:00
if (vendor == RT_VENDORID && devid == RT_DEVICEID_8139) {
if (revid != 0x20) {
/* 8139, let rl(4) take care of this device. */
return (ENXIO);
}
}
t = re_devs;
for (i = 0; i < sizeof(re_devs) / sizeof(re_devs[0]); i++, t++) {
if (vendor == t->rl_vid && devid == t->rl_did) {
device_set_desc(dev, t->rl_name);
return (BUS_PROBE_DEFAULT);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
}
Whitespace pass
2004-07-06 02:46:53 +00:00
return (ENXIO);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
/*
* Map a single buffer address.
*/
static void
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
Use correct pointer type. This should fix non-functional re(4) driver on 64bit architectures. Approved by: re (scottl)
2005-07-01 04:12:42 +00:00
bus_addr_t *addr;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
if (error)
return;
KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg));
addr = arg;
*addr = segs->ds_addr;
}
static int
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_allocmem(device_t dev, struct rl_softc *sc)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
Enable 64bit DMA addressing for RTL810xE/RTL8168/RTL8111 PCIe controllers. Some old PCI controllers may work with DAC but it was known to be buggy so 64bit DMA addressing is used only on PCIe controllers.
2010-11-05 18:19:54 +00:00
bus_addr_t lowaddr;
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
bus_size_t rx_list_size, tx_list_size;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
int error;
int i;
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
rx_list_size = sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc);
tx_list_size = sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc);
/*
* Allocate the parent bus DMA tag appropriate for PCI.
Some RealTek chips are known to be buggy on DAC handling, so disable DAC by default.
2008-03-23 05:13:45 +00:00
* In order to use DAC, RL_CPLUSCMD_PCI_DAC bit of RL_CPLUS_CMD
* register should be set. However some RealTek chips are known
* to be buggy on DAC handling, therefore disable DAC by limiting
* DMA address space to 32bit. PCIe variants of RealTek chips
Enable 64bit DMA addressing for RTL810xE/RTL8168/RTL8111 PCIe controllers. Some old PCI controllers may work with DAC but it was known to be buggy so 64bit DMA addressing is used only on PCIe controllers.
2010-11-05 18:19:54 +00:00
* may not have the limitation.
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
*/
Enable 64bit DMA addressing for RTL810xE/RTL8168/RTL8111 PCIe controllers. Some old PCI controllers may work with DAC but it was known to be buggy so 64bit DMA addressing is used only on PCIe controllers.
2010-11-05 18:19:54 +00:00
lowaddr = BUS_SPACE_MAXADDR;
if ((sc->rl_flags & RL_FLAG_PCIE) == 0)
lowaddr = BUS_SPACE_MAXADDR_32BIT;
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0,
Enable 64bit DMA addressing for RTL810xE/RTL8168/RTL8111 PCIe controllers. Some old PCI controllers may work with DAC but it was known to be buggy so 64bit DMA addressing is used only on PCIe controllers.
2010-11-05 18:19:54 +00:00
lowaddr, BUS_SPACE_MAXADDR, NULL, NULL,
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
NULL, NULL, &sc->rl_parent_tag);
if (error) {
device_printf(dev, "could not allocate parent DMA tag\n");
return (error);
}
/*
* Allocate map for TX mbufs.
*/
error = bus_dma_tag_create(sc->rl_parent_tag, 1, 0,
BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL,
NULL, MCLBYTES * RL_NTXSEGS, RL_NTXSEGS, 4096, 0,
NULL, NULL, &sc->rl_ldata.rl_tx_mtag);
if (error) {
device_printf(dev, "could not allocate TX DMA tag\n");
return (error);
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/*
* Allocate map for RX mbufs.
*/
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t),
0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
MJUM9BYTES, 1, MJUM9BYTES, 0, NULL, NULL,
&sc->rl_ldata.rl_jrx_mtag);
if (error) {
device_printf(dev,
"could not allocate jumbo RX DMA tag\n");
return (error);
}
}
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t), 0,
BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
MCLBYTES, 1, MCLBYTES, 0, NULL, NULL, &sc->rl_ldata.rl_rx_mtag);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
if (error) {
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
device_printf(dev, "could not allocate RX DMA tag\n");
return (error);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
/*
* Allocate map for TX descriptor list.
*/
error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
NULL, tx_list_size, 1, tx_list_size, 0,
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
NULL, NULL, &sc->rl_ldata.rl_tx_list_tag);
if (error) {
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
device_printf(dev, "could not allocate TX DMA ring tag\n");
return (error);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
/* Allocate DMA'able memory for the TX ring */
Whitespace pass
2004-07-06 02:46:53 +00:00
error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag,
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
(void **)&sc->rl_ldata.rl_tx_list,
BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
Whitespace pass
2004-07-06 02:46:53 +00:00
&sc->rl_ldata.rl_tx_list_map);
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
if (error) {
device_printf(dev, "could not allocate TX DMA ring\n");
return (error);
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/* Load the map for the TX ring. */
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
sc->rl_ldata.rl_tx_list_addr = 0;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag,
sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list,
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
tx_list_size, re_dma_map_addr,
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
&sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT);
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
if (error != 0 || sc->rl_ldata.rl_tx_list_addr == 0) {
device_printf(dev, "could not load TX DMA ring\n");
return (ENOMEM);
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/* Create DMA maps for TX buffers */
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
error = bus_dmamap_create(sc->rl_ldata.rl_tx_mtag, 0,
&sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
if (error) {
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
device_printf(dev, "could not create DMA map for TX\n");
return (error);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
}
/*
* Allocate map for RX descriptor list.
*/
error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN,
0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL,
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
NULL, rx_list_size, 1, rx_list_size, 0,
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
NULL, NULL, &sc->rl_ldata.rl_rx_list_tag);
if (error) {
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
device_printf(dev, "could not create RX DMA ring tag\n");
return (error);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
/* Allocate DMA'able memory for the RX ring */
Whitespace pass
2004-07-06 02:46:53 +00:00
error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag,
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
(void **)&sc->rl_ldata.rl_rx_list,
BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
Whitespace pass
2004-07-06 02:46:53 +00:00
&sc->rl_ldata.rl_rx_list_map);
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
if (error) {
device_printf(dev, "could not allocate RX DMA ring\n");
return (error);
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/* Load the map for the RX ring. */
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
sc->rl_ldata.rl_rx_list_addr = 0;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag,
sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list,
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
rx_list_size, re_dma_map_addr,
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
&sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT);
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
if (error != 0 || sc->rl_ldata.rl_rx_list_addr == 0) {
device_printf(dev, "could not load RX DMA ring\n");
return (ENOMEM);
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/* Create DMA maps for RX buffers */
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
&sc->rl_ldata.rl_jrx_sparemap);
if (error) {
device_printf(dev,
"could not create spare DMA map for jumbo RX\n");
return (error);
}
for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0,
&sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
if (error) {
device_printf(dev,
"could not create DMA map for jumbo RX\n");
return (error);
}
}
}
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
&sc->rl_ldata.rl_rx_sparemap);
if (error) {
device_printf(dev, "could not create spare DMA map for RX\n");
return (error);
}
for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0,
&sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
if (error) {
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
device_printf(dev, "could not create DMA map for RX\n");
return (error);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
}
Add simple MAC statistics counter reading support. Unfortunately useful counters like rl_missed_pkts is 16 bits quantity which is too small to hold meaningful information happened in a second. This means driver should frequently read these counters in order not to lose accuracy and that approach is too inefficient in driver's view. Moreover it seems there is no way to trigger an interrupt to detect counter near-full or wraparound event as well as lacking clearing the MAC counters. Another limitation of reading the counters from RealTek controllers is lack of interrupt firing at the end of DMA cycle of MAC counter read request such that driver have to poll the end of the DMA which is a time consuming process as well as inefficient. The more severe issue of the MAC counter read request is it takes too long to complete the DMA. All these limitation made maintaining MAC counters in driver impractical. For now, just provide simple sysctl interface to trigger reading the MAC counters. These counters could be used to track down driver issues. Users can read MAC counters maintained in controller with the following command. #sysctl dev.re.0.stats=1 While I'm here add check for validity of dma map and allocated memory before unloading/freeing them. Tested by: rmacklem
2010-11-05 19:28:00 +00:00
/* Create DMA map for statistics. */
error = bus_dma_tag_create(sc->rl_parent_tag, RL_DUMP_ALIGN, 0,
BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
sizeof(struct rl_stats), 1, sizeof(struct rl_stats), 0, NULL, NULL,
&sc->rl_ldata.rl_stag);
if (error) {
device_printf(dev, "could not create statistics DMA tag\n");
return (error);
}
/* Allocate DMA'able memory for statistics. */
error = bus_dmamem_alloc(sc->rl_ldata.rl_stag,
(void **)&sc->rl_ldata.rl_stats,
BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
&sc->rl_ldata.rl_smap);
if (error) {
device_printf(dev,
"could not allocate statistics DMA memory\n");
return (error);
}
/* Load the map for statistics. */
sc->rl_ldata.rl_stats_addr = 0;
error = bus_dmamap_load(sc->rl_ldata.rl_stag, sc->rl_ldata.rl_smap,
sc->rl_ldata.rl_stats, sizeof(struct rl_stats), re_dma_map_addr,
&sc->rl_ldata.rl_stats_addr, BUS_DMA_NOWAIT);
if (error != 0 || sc->rl_ldata.rl_stats_addr == 0) {
device_printf(dev, "could not load statistics DMA memory\n");
return (ENOMEM);
}
Whitespace pass
2004-07-06 02:46:53 +00:00
return (0);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
/*
* Attach the interface. Allocate softc structures, do ifmedia
* setup and ethernet/BPF attach.
*/
static int
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_attach(device_t dev)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
u_char eaddr[ETHER_ADDR_LEN];
Fix re(4) breakge introduced in tree from rev 1.68. This should fix incorrect configuration of station address on big-endian architectures. Reviewed by: wpaul Tested on: sparc64
2006-08-03 00:15:19 +00:00
u_int16_t as[ETHER_ADDR_LEN / 2];
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
struct rl_softc *sc;
struct ifnet *ifp;
Sprinkle some const.
2011-11-02 23:23:19 +00:00
const struct rl_hwrev *hw_rev;
Disable PCIe ASPM (Active State Power Management) for all controllers. More and more RealTek controllers started to implement EEE feature. Vendor driver seems to load a kind of firmware for EEE with additional PHY fixups. It is known that the EEE feature may need ASPM support. Unfortunately there is no documentation for EEE of the controller so enabling ASPM may cause more problems.
2011-11-16 23:29:27 +00:00
u_int32_t cap, ctl;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
int hwrev;
Switch to memory space register mapping over IO space. If that mapping fails fall back to traditional IO space access.
2008-07-02 05:21:09 +00:00
u_int16_t devid, re_did = 0;
Convert the PHY drivers to honor the mii_flags passed down and convert the NIC drivers as well as the PHY drivers to take advantage of the mii_attach() introduced in r213878 to get rid of certain hacks. For the most part these were: - Artificially limiting miibus_{read,write}reg methods to certain PHY addresses; we now let mii_attach() only probe the PHY at the desired address(es) instead. - PHY drivers setting MIIF_* flags based on the NIC driver they hang off from, partly even based on grabbing and using the softc of the parent; we now pass these flags down from the NIC to the PHY drivers via mii_attach(). This got us rid of all such hacks except those of brgphy() in combination with bce(4) and bge(4), which is way beyond what can be expressed with simple flags. While at it, I took the opportunity to change the NIC drivers to pass up the error returned by mii_attach() (previously by mii_phy_probe()) and unify the error message used in this case where and as appropriate as mii_attach() actually can fail for a number of reasons, not just because of no PHY(s) being present at the expected address(es). Reviewed by: jhb, yongari
2010-10-15 14:52:11 +00:00
int error = 0, i, phy, rid;
Prefer MSI-X to MSI on controllers that support MSI-X. All recent PCIe controllers(RTL8102E or later and RTL8168/8111C or later) supports either 2 or 4 MSI-X messages. Unfortunately vendor did not publicly release RSS related information yet. However switching to MSI-X is one-step forward to support RSS.
2011-01-25 22:18:00 +00:00
int msic, msixc, reg;
For MSI capable hardwares, enable MSI enable bit in RL_CFG2 register. If MSI was disabled by hw.re.msi_disable tunable expliclty clear the MSI enable bit.
2008-03-23 05:31:35 +00:00
uint8_t cfg;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
sc = device_get_softc(dev);
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
sc->rl_dev = dev;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
MTX_DEF);
- Use if_printf() and device_printf() in re(4) and remove rl_unit from the softc. - Use callout_init_mtx() and rather than timeout/untimeout in both rl(4) and re(4). - Fix locking for ifmedia by locking the driver in the ifmedia handlers rather than in the miibus functions. (re(4) didn't lock the mii stuff at all!) - Fix some locking in re_ioctl(). Note: the two drivers share the same softc declared in if_rlreg.h, so they had to be change simultaneously. MFC after: 1 week Tested by: several on rl(4), none on re(4)
2005-09-29 16:52:19 +00:00
callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/*
* Map control/status registers.
*/
pci_enable_busmaster(dev);
Switch to memory space register mapping over IO space. If that mapping fails fall back to traditional IO space access.
2008-07-02 05:21:09 +00:00
devid = pci_get_device(dev);
Add a new tunable hw.re.prefer_iomap which disables memory register mapping. The tunable is OFF for all controllers except RTL8169SC family. RTL8169SC seems to require more magic to use memory register mapping. r187483 added a fix for RTL8169SCe controller but it does not looke like fix other variants of RTL8169SC. Tested by: Gavin Stone-Tolcher g.stone-tolcher <> its dot uq dot edu dot au
2009-03-09 06:02:55 +00:00
/*
* Prefer memory space register mapping over IO space.
* Because RTL8169SC does not seem to work when memory mapping
Remove extra white spaces.
2010-11-05 18:23:43 +00:00
* is used always activate io mapping.
Add a new tunable hw.re.prefer_iomap which disables memory register mapping. The tunable is OFF for all controllers except RTL8169SC family. RTL8169SC seems to require more magic to use memory register mapping. r187483 added a fix for RTL8169SCe controller but it does not looke like fix other variants of RTL8169SC. Tested by: Gavin Stone-Tolcher g.stone-tolcher <> its dot uq dot edu dot au
2009-03-09 06:02:55 +00:00
*/
if (devid == RT_DEVICEID_8169SC)
prefer_iomap = 1;
if (prefer_iomap == 0) {
sc->rl_res_id = PCIR_BAR(1);
sc->rl_res_type = SYS_RES_MEMORY;
/* RTL8168/8101E seems to use different BARs. */
if (devid == RT_DEVICEID_8168 || devid == RT_DEVICEID_8101E)
sc->rl_res_id = PCIR_BAR(2);
} else {
sc->rl_res_id = PCIR_BAR(0);
sc->rl_res_type = SYS_RES_IOPORT;
}
Switch to memory space register mapping over IO space. If that mapping fails fall back to traditional IO space access.
2008-07-02 05:21:09 +00:00
sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
&sc->rl_res_id, RF_ACTIVE);
Add a new tunable hw.re.prefer_iomap which disables memory register mapping. The tunable is OFF for all controllers except RTL8169SC family. RTL8169SC seems to require more magic to use memory register mapping. r187483 added a fix for RTL8169SCe controller but it does not looke like fix other variants of RTL8169SC. Tested by: Gavin Stone-Tolcher g.stone-tolcher <> its dot uq dot edu dot au
2009-03-09 06:02:55 +00:00
if (sc->rl_res == NULL && prefer_iomap == 0) {
Switch to memory space register mapping over IO space. If that mapping fails fall back to traditional IO space access.
2008-07-02 05:21:09 +00:00
sc->rl_res_id = PCIR_BAR(0);
sc->rl_res_type = SYS_RES_IOPORT;
sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
&sc->rl_res_id, RF_ACTIVE);
Add a new tunable hw.re.prefer_iomap which disables memory register mapping. The tunable is OFF for all controllers except RTL8169SC family. RTL8169SC seems to require more magic to use memory register mapping. r187483 added a fix for RTL8169SCe controller but it does not looke like fix other variants of RTL8169SC. Tested by: Gavin Stone-Tolcher g.stone-tolcher <> its dot uq dot edu dot au
2009-03-09 06:02:55 +00:00
}
if (sc->rl_res == NULL) {
device_printf(dev, "couldn't map ports/memory\n");
error = ENXIO;
goto fail;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
sc->rl_btag = rman_get_bustag(sc->rl_res);
sc->rl_bhandle = rman_get_bushandle(sc->rl_res);
Prefer MSI-X to MSI on controllers that support MSI-X. All recent PCIe controllers(RTL8102E or later and RTL8168/8111C or later) supports either 2 or 4 MSI-X messages. Unfortunately vendor did not publicly release RSS related information yet. However switching to MSI-X is one-step forward to support RSS.
2011-01-25 22:18:00 +00:00
msic = pci_msi_count(dev);
msixc = pci_msix_count(dev);
Disable PCIe ASPM (Active State Power Management) for all controllers. More and more RealTek controllers started to implement EEE feature. Vendor driver seems to load a kind of firmware for EEE with additional PHY fixups. It is known that the EEE feature may need ASPM support. Unfortunately there is no documentation for EEE of the controller so enabling ASPM may cause more problems.
2011-11-16 23:29:27 +00:00
if (pci_find_cap(dev, PCIY_EXPRESS, &reg) == 0) {
It seems that RealTek PCIe controllers require an explicit Tx poll command whenever Tx completion interrupt is raised. The Tx poll bit is cleared when all packets waiting to be transferred have been processed. This means the second Tx poll command can be silently ignored as the Tx poll bit could be still active while processing of previous Tx poll command is in progress. To address the issue re(4) used to invoke the Tx poll command in Tx completion handler whenever it detects there are pending packets in TxQ. However that still does not seem to completely eliminate watchdog timeouts seen on RealTek PCIe controllers. To fix the issue kick Tx poll command only after Tx completion interrupt is raised as this would indicate Tx is now idle state such that it can accept new Tx poll command again. While here apply this workaround for PCIe based controllers as other controllers does not seem to have this limitation. Tested by: Victor Balada Diaz < victor <> bsdes DOT net >
2008-12-17 08:18:11 +00:00
sc->rl_flags |= RL_FLAG_PCIE;
Disable PCIe ASPM (Active State Power Management) for all controllers. More and more RealTek controllers started to implement EEE feature. Vendor driver seems to load a kind of firmware for EEE with additional PHY fixups. It is known that the EEE feature may need ASPM support. Unfortunately there is no documentation for EEE of the controller so enabling ASPM may cause more problems.
2011-11-16 23:29:27 +00:00
sc->rl_expcap = reg;
}
Prefer MSI-X to MSI on controllers that support MSI-X. All recent PCIe controllers(RTL8102E or later and RTL8168/8111C or later) supports either 2 or 4 MSI-X messages. Unfortunately vendor did not publicly release RSS related information yet. However switching to MSI-X is one-step forward to support RSS.
2011-01-25 22:18:00 +00:00
if (bootverbose) {
device_printf(dev, "MSI count : %d\n", msic);
device_printf(dev, "MSI-X count : %d\n", msixc);
}
if (msix_disable > 0)
msixc = 0;
if (msi_disable > 0)
msic = 0;
/* Prefer MSI-X to MSI. */
if (msixc > 0) {
- Implement the RX EARLYOFF and RXDV GATED bits as done by RealTek's Linux driver as version 8.037.00 for RTL8168{E-VL,EP,F,G,GU} and RTL8111B. This makes reception of packets work with the RTL8168G (HW rev. 0x4c000000) in my Shuttle DS47. - Consistently use RL_MSI_MESSAGES. In joint forces with: yongari MFC after: 5 days
2014-02-05 23:13:40 +00:00
msixc = RL_MSI_MESSAGES;
Prefer MSI-X to MSI on controllers that support MSI-X. All recent PCIe controllers(RTL8102E or later and RTL8168/8111C or later) supports either 2 or 4 MSI-X messages. Unfortunately vendor did not publicly release RSS related information yet. However switching to MSI-X is one-step forward to support RSS.
2011-01-25 22:18:00 +00:00
rid = PCIR_BAR(4);
sc->rl_res_pba = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
&rid, RF_ACTIVE);
if (sc->rl_res_pba == NULL) {
device_printf(sc->rl_dev,
"could not allocate MSI-X PBA resource\n");
}
if (sc->rl_res_pba != NULL &&
pci_alloc_msix(dev, &msixc) == 0) {
- Implement the RX EARLYOFF and RXDV GATED bits as done by RealTek's Linux driver as version 8.037.00 for RTL8168{E-VL,EP,F,G,GU} and RTL8111B. This makes reception of packets work with the RTL8168G (HW rev. 0x4c000000) in my Shuttle DS47. - Consistently use RL_MSI_MESSAGES. In joint forces with: yongari MFC after: 5 days
2014-02-05 23:13:40 +00:00
if (msixc == RL_MSI_MESSAGES) {
Prefer MSI-X to MSI on controllers that support MSI-X. All recent PCIe controllers(RTL8102E or later and RTL8168/8111C or later) supports either 2 or 4 MSI-X messages. Unfortunately vendor did not publicly release RSS related information yet. However switching to MSI-X is one-step forward to support RSS.
2011-01-25 22:18:00 +00:00
device_printf(dev, "Using %d MSI-X message\n",
msixc);
sc->rl_flags |= RL_FLAG_MSIX;
} else
pci_release_msi(dev);
}
if ((sc->rl_flags & RL_FLAG_MSIX) == 0) {
if (sc->rl_res_pba != NULL)
bus_release_resource(dev, SYS_RES_MEMORY, rid,
sc->rl_res_pba);
sc->rl_res_pba = NULL;
msixc = 0;
}
Add MSI support. Ever since switching to adaptive polling re(4) occasionally spews watchdog timeouts on systems with MSI capability. This change is minimal one for supporting MSI and re(4) also needs MSIX support for RTL8111C in future. Because softc structure of re(4) is shared with rl(4), rl(4) was touched to use the modified softc. Reported by: cnst Tested by: cnst Approved by: re (kensmith)
2007-07-24 01:24:03 +00:00
}
Prefer MSI-X to MSI on controllers that support MSI-X. All recent PCIe controllers(RTL8102E or later and RTL8168/8111C or later) supports either 2 or 4 MSI-X messages. Unfortunately vendor did not publicly release RSS related information yet. However switching to MSI-X is one-step forward to support RSS.
2011-01-25 22:18:00 +00:00
/* Prefer MSI to INTx. */
if (msixc == 0 && msic > 0) {
- Implement the RX EARLYOFF and RXDV GATED bits as done by RealTek's Linux driver as version 8.037.00 for RTL8168{E-VL,EP,F,G,GU} and RTL8111B. This makes reception of packets work with the RTL8168G (HW rev. 0x4c000000) in my Shuttle DS47. - Consistently use RL_MSI_MESSAGES. In joint forces with: yongari MFC after: 5 days
2014-02-05 23:13:40 +00:00
msic = RL_MSI_MESSAGES;
Add MSI support. Ever since switching to adaptive polling re(4) occasionally spews watchdog timeouts on systems with MSI capability. This change is minimal one for supporting MSI and re(4) also needs MSIX support for RTL8111C in future. Because softc structure of re(4) is shared with rl(4), rl(4) was touched to use the modified softc. Reported by: cnst Tested by: cnst Approved by: re (kensmith)
2007-07-24 01:24:03 +00:00
if (pci_alloc_msi(dev, &msic) == 0) {
if (msic == RL_MSI_MESSAGES) {
Prefer MSI-X to MSI on controllers that support MSI-X. All recent PCIe controllers(RTL8102E or later and RTL8168/8111C or later) supports either 2 or 4 MSI-X messages. Unfortunately vendor did not publicly release RSS related information yet. However switching to MSI-X is one-step forward to support RSS.
2011-01-25 22:18:00 +00:00
device_printf(dev, "Using %d MSI message\n",
Add MSI support. Ever since switching to adaptive polling re(4) occasionally spews watchdog timeouts on systems with MSI capability. This change is minimal one for supporting MSI and re(4) also needs MSIX support for RTL8111C in future. Because softc structure of re(4) is shared with rl(4), rl(4) was touched to use the modified softc. Reported by: cnst Tested by: cnst Approved by: re (kensmith)
2007-07-24 01:24:03 +00:00
msic);
Instead of allocating variables for each events/hardware flags, use a dedicated flag that represents controller capabilities/events. This will simplify many part of code that requires different workaround for each controller revisions and will enhance readability. While I'm here move PHY wakeup code up before mii_phy_probe() which seems to help to wake PHY in some cases.
2008-07-02 06:29:43 +00:00
sc->rl_flags |= RL_FLAG_MSI;
Don't touch MSI enable bit in RL_CFG2 register. For unknown reason clearing MSI enable bit for MSI capable hardwares resulted in Tx problems. MSI enable bit is set only when MSI is requested from user. Tested by: remko
2008-04-15 00:46:15 +00:00
/* Explicitly set MSI enable bit. */
CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
cfg = CSR_READ_1(sc, RL_CFG2);
cfg |= RL_CFG2_MSI;
CSR_WRITE_1(sc, RL_CFG2, cfg);
While accessing EEPROM command register use pre-defined constant instead of hardcoded value.
2008-07-02 05:01:19 +00:00
CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
style(9) Pointed out by: cnst Approved by: re (kensmith)
2007-07-27 00:43:12 +00:00
} else
pci_release_msi(dev);
Add MSI support. Ever since switching to adaptive polling re(4) occasionally spews watchdog timeouts on systems with MSI capability. This change is minimal one for supporting MSI and re(4) also needs MSIX support for RTL8111C in future. Because softc structure of re(4) is shared with rl(4), rl(4) was touched to use the modified softc. Reported by: cnst Tested by: cnst Approved by: re (kensmith)
2007-07-24 01:24:03 +00:00
}
Prefer MSI-X to MSI on controllers that support MSI-X. All recent PCIe controllers(RTL8102E or later and RTL8168/8111C or later) supports either 2 or 4 MSI-X messages. Unfortunately vendor did not publicly release RSS related information yet. However switching to MSI-X is one-step forward to support RSS.
2011-01-25 22:18:00 +00:00
if ((sc->rl_flags & RL_FLAG_MSI) == 0)
msic = 0;
Add MSI support. Ever since switching to adaptive polling re(4) occasionally spews watchdog timeouts on systems with MSI capability. This change is minimal one for supporting MSI and re(4) also needs MSIX support for RTL8111C in future. Because softc structure of re(4) is shared with rl(4), rl(4) was touched to use the modified softc. Reported by: cnst Tested by: cnst Approved by: re (kensmith)
2007-07-24 01:24:03 +00:00
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Add MSI support. Ever since switching to adaptive polling re(4) occasionally spews watchdog timeouts on systems with MSI capability. This change is minimal one for supporting MSI and re(4) also needs MSIX support for RTL8111C in future. Because softc structure of re(4) is shared with rl(4), rl(4) was touched to use the modified softc. Reported by: cnst Tested by: cnst Approved by: re (kensmith)
2007-07-24 01:24:03 +00:00
/* Allocate interrupt */
Prefer MSI-X to MSI on controllers that support MSI-X. All recent PCIe controllers(RTL8102E or later and RTL8168/8111C or later) supports either 2 or 4 MSI-X messages. Unfortunately vendor did not publicly release RSS related information yet. However switching to MSI-X is one-step forward to support RSS.
2011-01-25 22:18:00 +00:00
if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0) {
Add MSI support. Ever since switching to adaptive polling re(4) occasionally spews watchdog timeouts on systems with MSI capability. This change is minimal one for supporting MSI and re(4) also needs MSIX support for RTL8111C in future. Because softc structure of re(4) is shared with rl(4), rl(4) was touched to use the modified softc. Reported by: cnst Tested by: cnst Approved by: re (kensmith)
2007-07-24 01:24:03 +00:00
rid = 0;
sc->rl_irq[0] = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
RF_SHAREABLE | RF_ACTIVE);
if (sc->rl_irq[0] == NULL) {
device_printf(dev, "couldn't allocate IRQ resources\n");
error = ENXIO;
goto fail;
}
} else {
for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) {
sc->rl_irq[i] = bus_alloc_resource_any(dev,
SYS_RES_IRQ, &rid, RF_ACTIVE);
if (sc->rl_irq[i] == NULL) {
device_printf(dev,
Correct comment typos.
2013-06-28 14:41:06 +00:00
"couldn't allocate IRQ resources for "
Add MSI support. Ever since switching to adaptive polling re(4) occasionally spews watchdog timeouts on systems with MSI capability. This change is minimal one for supporting MSI and re(4) also needs MSIX support for RTL8111C in future. Because softc structure of re(4) is shared with rl(4), rl(4) was touched to use the modified softc. Reported by: cnst Tested by: cnst Approved by: re (kensmith)
2007-07-24 01:24:03 +00:00
"message %d\n", rid);
error = ENXIO;
goto fail;
}
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
If MSI is not active make sure to disable MSI in EEPROM.
2008-07-02 06:45:59 +00:00
if ((sc->rl_flags & RL_FLAG_MSI) == 0) {
CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
cfg = CSR_READ_1(sc, RL_CFG2);
if ((cfg & RL_CFG2_MSI) != 0) {
device_printf(dev, "turning off MSI enable bit.\n");
cfg &= ~RL_CFG2_MSI;
CSR_WRITE_1(sc, RL_CFG2, cfg);
}
CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
}
Disable PCIe ASPM (Active State Power Management) for all controllers. More and more RealTek controllers started to implement EEE feature. Vendor driver seems to load a kind of firmware for EEE with additional PHY fixups. It is known that the EEE feature may need ASPM support. Unfortunately there is no documentation for EEE of the controller so enabling ASPM may cause more problems.
2011-11-16 23:29:27 +00:00
/* Disable ASPM L0S/L1. */
if (sc->rl_expcap != 0) {
cap = pci_read_config(dev, sc->rl_expcap +
Align the PCI Express #defines with the style used for the PCI-X #defines. This also has the advantage that it makes the names more compact, iand also allows us to correct the non-uniform naming of the PCIM_LINK_* defines, making them all consistent amongst themselves. This is a mostly mechanical rename: s/PCIR_EXPRESS_/PCIER_/g s/PCIM_EXP_/PCIEM_/g s/PCIM_LINK_/PCIEM_LINK_/g When this is MFC'd, #defines will be added for the old names to assist out-of-tree drivers. Discussed with: jhb MFC after: 1 week
2012-09-18 22:04:59 +00:00
PCIER_LINK_CAP, 2);
if ((cap & PCIEM_LINK_CAP_ASPM) != 0) {
Disable PCIe ASPM (Active State Power Management) for all controllers. More and more RealTek controllers started to implement EEE feature. Vendor driver seems to load a kind of firmware for EEE with additional PHY fixups. It is known that the EEE feature may need ASPM support. Unfortunately there is no documentation for EEE of the controller so enabling ASPM may cause more problems.
2011-11-16 23:29:27 +00:00
ctl = pci_read_config(dev, sc->rl_expcap +
Align the PCI Express #defines with the style used for the PCI-X #defines. This also has the advantage that it makes the names more compact, iand also allows us to correct the non-uniform naming of the PCIM_LINK_* defines, making them all consistent amongst themselves. This is a mostly mechanical rename: s/PCIR_EXPRESS_/PCIER_/g s/PCIM_EXP_/PCIEM_/g s/PCIM_LINK_/PCIEM_LINK_/g When this is MFC'd, #defines will be added for the old names to assist out-of-tree drivers. Discussed with: jhb MFC after: 1 week
2012-09-18 22:04:59 +00:00
PCIER_LINK_CTL, 2);
Switch some PCI register reads from using magic numbers to using the names defined in pcireg.h MFC after: 1 week
2012-09-19 12:27:23 +00:00
if ((ctl & PCIEM_LINK_CTL_ASPMC) != 0) {
ctl &= ~PCIEM_LINK_CTL_ASPMC;
Disable PCIe ASPM (Active State Power Management) for all controllers. More and more RealTek controllers started to implement EEE feature. Vendor driver seems to load a kind of firmware for EEE with additional PHY fixups. It is known that the EEE feature may need ASPM support. Unfortunately there is no documentation for EEE of the controller so enabling ASPM may cause more problems.
2011-11-16 23:29:27 +00:00
pci_write_config(dev, sc->rl_expcap +
Align the PCI Express #defines with the style used for the PCI-X #defines. This also has the advantage that it makes the names more compact, iand also allows us to correct the non-uniform naming of the PCIM_LINK_* defines, making them all consistent amongst themselves. This is a mostly mechanical rename: s/PCIR_EXPRESS_/PCIER_/g s/PCIM_EXP_/PCIEM_/g s/PCIM_LINK_/PCIEM_LINK_/g When this is MFC'd, #defines will be added for the old names to assist out-of-tree drivers. Discussed with: jhb MFC after: 1 week
2012-09-18 22:04:59 +00:00
PCIER_LINK_CTL, ctl, 2);
Disable PCIe ASPM (Active State Power Management) for all controllers. More and more RealTek controllers started to implement EEE feature. Vendor driver seems to load a kind of firmware for EEE with additional PHY fixups. It is known that the EEE feature may need ASPM support. Unfortunately there is no documentation for EEE of the controller so enabling ASPM may cause more problems.
2011-11-16 23:29:27 +00:00
device_printf(dev, "ASPM disabled\n");
}
} else
device_printf(dev, "no ASPM capability\n");
}
- For the 8169 chips, read the station address by forcing an EEPROM autoload and then copying the contends of the station address registers. For some reason, reading the EEPROM on the 8169S doesn't work right. This gets around the problem, and allows us to read the station address correctly on the 8169S. - Insert a delay after initiating packet transmition in re_diag() to allow lots of time for the frame to echo back to the host, and wait for both the 'RX complete' and 'timeout expired' bits in the ISR register to be set. - Deal more intelligently with the fact that the frame length field in the RX descriptor is a different width on the 8139C+ than it is on the 8169/8169S/8110S - For the 8169, you have to set bit 17 in the TX config register to enter digital loopback mode, but for the 8139C+, you have to set both bits 17 and 18. Take this into account so that re_diag() works properly for both types of chips.
2003-09-11 06:56:46 +00:00
hw_rev = re_hwrevs;
Display detected chip revision in device attach. This will help to diagnose revision specific issues in future. Also add 0x to notify that the revision number is in hex form.
2008-07-02 06:55:03 +00:00
hwrev = CSR_READ_4(sc, RL_TXCFG);
- Add support for 8110SCe part. Some magic registers were taken from Linux driver. - Swap hardware revisions for 8110S and 8169S as Linux driver claims. Reviewed by: yongari (early version)
2009-01-20 20:22:28 +00:00
switch (hwrev & 0x70000000) {
case 0x00000000:
case 0x10000000:
device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0xfc800000);
hwrev &= (RL_TXCFG_HWREV | 0x80000000);
break;
default:
device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0x7c800000);
Correct MAC revision bits. Previously it always cleared bit 20 and bit 21.
2013-10-21 06:22:20 +00:00
sc->rl_macrev = hwrev & 0x00700000;
- Add support for 8110SCe part. Some magic registers were taken from Linux driver. - Swap hardware revisions for 8110S and 8169S as Linux driver claims. Reviewed by: yongari (early version)
2009-01-20 20:22:28 +00:00
hwrev &= RL_TXCFG_HWREV;
break;
}
Correct MAC revision bits. Previously it always cleared bit 20 and bit 21.
2013-10-21 06:22:20 +00:00
device_printf(dev, "MAC rev. 0x%08x\n", sc->rl_macrev);
- For the 8169 chips, read the station address by forcing an EEPROM autoload and then copying the contends of the station address registers. For some reason, reading the EEPROM on the 8169S doesn't work right. This gets around the problem, and allows us to read the station address correctly on the 8169S. - Insert a delay after initiating packet transmition in re_diag() to allow lots of time for the frame to echo back to the host, and wait for both the 'RX complete' and 'timeout expired' bits in the ISR register to be set. - Deal more intelligently with the fact that the frame length field in the RX descriptor is a different width on the 8139C+ than it is on the 8169/8169S/8110S - For the 8169, you have to set bit 17 in the TX config register to enter digital loopback mode, but for the 8139C+, you have to set both bits 17 and 18. Take this into account so that re_diag() works properly for both types of chips.
2003-09-11 06:56:46 +00:00
while (hw_rev->rl_desc != NULL) {
if (hw_rev->rl_rev == hwrev) {
sc->rl_type = hw_rev->rl_type;
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
sc->rl_hwrev = hw_rev;
- For the 8169 chips, read the station address by forcing an EEPROM autoload and then copying the contends of the station address registers. For some reason, reading the EEPROM on the 8169S doesn't work right. This gets around the problem, and allows us to read the station address correctly on the 8169S. - Insert a delay after initiating packet transmition in re_diag() to allow lots of time for the frame to echo back to the host, and wait for both the 'RX complete' and 'timeout expired' bits in the ISR register to be set. - Deal more intelligently with the fact that the frame length field in the RX descriptor is a different width on the 8139C+ than it is on the 8169/8169S/8110S - For the 8169, you have to set bit 17 in the TX config register to enter digital loopback mode, but for the 8139C+, you have to set both bits 17 and 18. Take this into account so that re_diag() works properly for both types of chips.
2003-09-11 06:56:46 +00:00
break;
}
hw_rev++;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
if (hw_rev->rl_desc == NULL) {
Display detected chip revision in device attach. This will help to diagnose revision specific issues in future. Also add 0x to notify that the revision number is in hex form.
2008-07-02 06:55:03 +00:00
device_printf(dev, "Unknown H/W revision: 0x%08x\n", hwrev);
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
error = ENXIO;
goto fail;
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Instead of allocating variables for each events/hardware flags, use a dedicated flag that represents controller capabilities/events. This will simplify many part of code that requires different workaround for each controller revisions and will enhance readability. While I'm here move PHY wakeup code up before mii_phy_probe() which seems to help to wake PHY in some cases.
2008-07-02 06:29:43 +00:00
switch (hw_rev->rl_rev) {
case RL_HWREV_8139CPLUS:
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
sc->rl_flags |= RL_FLAG_FASTETHER | RL_FLAG_AUTOPAD;
Instead of allocating variables for each events/hardware flags, use a dedicated flag that represents controller capabilities/events. This will simplify many part of code that requires different workaround for each controller revisions and will enhance readability. While I'm here move PHY wakeup code up before mii_phy_probe() which seems to help to wake PHY in some cases.
2008-07-02 06:29:43 +00:00
break;
case RL_HWREV_8100E:
case RL_HWREV_8101E:
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_FASTETHER;
Instead of allocating variables for each events/hardware flags, use a dedicated flag that represents controller capabilities/events. This will simplify many part of code that requires different workaround for each controller revisions and will enhance readability. While I'm here move PHY wakeup code up before mii_phy_probe() which seems to help to wake PHY in some cases.
2008-07-02 06:29:43 +00:00
break;
Add driver support for RTL8102E and RTL8102EL which is the second generation of RTL810x PCIe fast ethernet controller. Note, Tx/Rx descriptor format is different from that of first generation of RTL8101E series. Jumbo frame is not supported for RTL810x family. Tested by: NAGATA Shinya ( maya AT negeta DOT com )
2008-07-09 01:58:18 +00:00
case RL_HWREV_8102E:
case RL_HWREV_8102EL:
Adding hardware ID for RTL810x PCIe found on HP Pavilion DV2-1022AX. Reviewed by: yongari Approved by: re (kib, kensmith)
2009-07-14 04:35:13 +00:00
case RL_HWREV_8102EL_SPIN1:
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
RL_FLAG_AUTOPAD;
Add driver support for RTL8102E and RTL8102EL which is the second generation of RTL810x PCIe fast ethernet controller. Note, Tx/Rx descriptor format is different from that of first generation of RTL8101E series. Jumbo frame is not supported for RTL810x family. Tested by: NAGATA Shinya ( maya AT negeta DOT com )
2008-07-09 01:58:18 +00:00
break;
Add initial support for RTL8103E PCIe fastethernet. PR: kern/142974
2010-01-27 17:49:27 +00:00
case RL_HWREV_8103E:
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
RL_FLAG_AUTOPAD | RL_FLAG_MACSLEEP;
Add initial support for RTL8103E PCIe fastethernet. PR: kern/142974
2010-01-27 17:49:27 +00:00
break;
Add initial support for RTL8401E PCIe Fast Ethernet. PR: 154789
2011-02-16 21:59:42 +00:00
case RL_HWREV_8401E:
Add support for RTL8105E PCIe Fast Ethernet controller. It seems the controller has a kind of embedded controller/memory and vendor applies a large set of magic code via undocumented PHY registers in device initialization stage. I guess it's a firmware image for the embedded controller in RTL8105E since the code is too big compared to other DSP fixups. However I have no idea what that magic code does and what's purpose of the embedded controller. Fortunately driver seems to still work without loading the firmware. While I'm here change device description of RTL810xE controller. H/W donated by: Realtek Semiconductor Corp.
2011-01-26 21:14:20 +00:00
case RL_HWREV_8105E:
Add preliminary support for second generation RTL8105E PCIe FastEthernet. H/W donated by: RealTek Semiconductor Corp.
2011-11-17 21:24:56 +00:00
case RL_HWREV_8105E_SPIN1:
Add preliminary support for RTL8106E PCIe FastEthernet. H/W donated by: RealTek Semiconductor Corp.
2013-10-21 06:27:20 +00:00
case RL_HWREV_8106E:
Add support for RTL8105E PCIe Fast Ethernet controller. It seems the controller has a kind of embedded controller/memory and vendor applies a large set of magic code via undocumented PHY registers in device initialization stage. I guess it's a firmware image for the embedded controller in RTL8105E since the code is too big compared to other DSP fixups. However I have no idea what that magic code does and what's purpose of the embedded controller. Fortunately driver seems to still work without loading the firmware. While I'm here change device description of RTL810xE controller. H/W donated by: Realtek Semiconductor Corp.
2011-01-26 21:14:20 +00:00
sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD;
break;
Make sure to stop TX MAC before freeing queued TX frames. For RTL8111DP, check if the TX MAC is active by reading RL_GTXSTART register. For RTL8402/8168E-VL/8168F/8411, wait until TX queue is empty.
2011-11-23 22:07:13 +00:00
case RL_HWREV_8402:
sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD |
RL_FLAG_CMDSTOP_WAIT_TXQ;
break;
Change model names of controller RTL_HWREV_8168_SPIN[123] to real ones. s/RL_HWREV_8168_SPIN1/RL_HWREV_8168B_SPIN1/g s/RL_HWREV_8168_SPIN2/RL_HWREV_8168B_SPIN2/g s/RL_HWREV_8168_SPIN3/RL_HWREV_8168B_SPIN3/g No functional changes.
2011-01-18 00:46:10 +00:00
case RL_HWREV_8168B_SPIN1:
case RL_HWREV_8168B_SPIN2:
Make WOL work on RTL8168B. This controller seems to require explicit command to enable Rx MAC prior to entering D3. Tested by: Cyrus Rahman <crahman <> gmail DOT com>
2008-12-11 02:24:11 +00:00
sc->rl_flags |= RL_FLAG_WOLRXENB;
/* FALLTHROUGH */
Change model names of controller RTL_HWREV_8168_SPIN[123] to real ones. s/RL_HWREV_8168_SPIN1/RL_HWREV_8168B_SPIN1/g s/RL_HWREV_8168_SPIN2/RL_HWREV_8168B_SPIN2/g s/RL_HWREV_8168_SPIN3/RL_HWREV_8168B_SPIN3/g No functional changes.
2011-01-18 00:46:10 +00:00
case RL_HWREV_8168B_SPIN3:
Retire RL_FLAG_INVMAR bit to match its comment and reality.
2009-01-20 20:04:09 +00:00
sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_MACSTAT;
Add basic support for RTL8168C, RTL8168CP, RTL8111C and RTL8111CP. ATM Tx/Rx checksum offload is supported but TSO and jumbo frame is not yet supported. Because these newer controllers use different descriptor formats, a flag RL_FLAG_DESCV2 flag was introduced to handle that case in Tx/Rx handler. Also newer controllers seems to require to not touch 'enable Tx/Rx bit' in RL_CPLUS_CMD register so don't blindly try to set that bits. Note, it seems that there is still power-saving related issue where driver fails to attach PHY. Rebooting seems to fix that issue but number of required reboots varys. Many thanks to users that helped during developement. I really appreciate their patient and test/feedbacks.
2008-07-02 07:54:53 +00:00
break;
case RL_HWREV_8168C_SPIN2:
For RTL8168C SPIN2 controllers, make sure to take the controller out of sleep mode prior to accessing to PHY. This should fix device attach failure seen on these controllers. Also enable the sleep mode when device is put into sleep state. PR: kern/123123, kern/123053
2008-12-17 06:01:03 +00:00
sc->rl_flags |= RL_FLAG_MACSLEEP;
/* FALLTHROUGH */
case RL_HWREV_8168C:
Correct MAC revision bits. Previously it always cleared bit 20 and bit 21.
2013-10-21 06:22:20 +00:00
if (sc->rl_macrev == 0x00200000)
For RTL8168C SPIN2 controllers, make sure to take the controller out of sleep mode prior to accessing to PHY. This should fix device attach failure seen on these controllers. Also enable the sleep mode when device is put into sleep state. PR: kern/123123, kern/123053
2008-12-17 06:01:03 +00:00
sc->rl_flags |= RL_FLAG_MACSLEEP;
/* FALLTHROUGH */
Add basic support for RTL8168C, RTL8168CP, RTL8111C and RTL8111CP. ATM Tx/Rx checksum offload is supported but TSO and jumbo frame is not yet supported. Because these newer controllers use different descriptor formats, a flag RL_FLAG_DESCV2 flag was introduced to handle that case in Tx/Rx handler. Also newer controllers seems to require to not touch 'enable Tx/Rx bit' in RL_CPLUS_CMD register so don't blindly try to set that bits. Note, it seems that there is still power-saving related issue where driver fails to attach PHY. Rebooting seems to fix that issue but number of required reboots varys. Many thanks to users that helped during developement. I really appreciate their patient and test/feedbacks.
2008-07-02 07:54:53 +00:00
case RL_HWREV_8168CP:
Retire RL_FLAG_INVMAR bit to match its comment and reality.
2009-01-20 20:04:09 +00:00
sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
For RTL8139C+ controllers, have controller handle padding short checksum offload frames. Software workaround used for broken controllers(RTL8169, RTL8168, RTL8168B) seem to cause watchdog timeouts on RTL8139C+. Introduce a new flag RL_FLAG_AUTOPAD to mark automatic padding feature of controller and set it for RTL8139C+ and controllers that use new descriptor format. This fixes watchdog timeouts seen on RTL8139C+. Reported by: Dimitri Rodis < DimitriR <> integritasystems dot com > Tested by: Dimitri Rodis < DimitriR <> integritasystems dot com >
2009-04-20 07:13:04 +00:00
RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
To save more power, switch to 10/100Mbps link when controller is put into suspend/shutdown. Old PCI controllers performed that operation in firmware but for RTL8111C or newer controllers, it's responsibility of driver. It's not clear whether the firmware of RTL8111B still downgrades its speed to 10/100Mbps so leave it as it was.
2011-11-23 23:29:18 +00:00
RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 | RL_FLAG_WOL_MANLINK;
Instead of allocating variables for each events/hardware flags, use a dedicated flag that represents controller capabilities/events. This will simplify many part of code that requires different workaround for each controller revisions and will enhance readability. While I'm here move PHY wakeup code up before mii_phy_probe() which seems to help to wake PHY in some cases.
2008-07-02 06:29:43 +00:00
break;
For RTL8168/8111D controller, make sure to wake PHY from power down mode. Otherwise, PHY access times out under certain conditions.
2012-02-14 00:54:40 +00:00
case RL_HWREV_8168D:
sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
RL_FLAG_WOL_MANLINK;
break;
Make sure to stop TX MAC before freeing queued TX frames. For RTL8111DP, check if the TX MAC is active by reading RL_GTXSTART register. For RTL8402/8168E-VL/8168F/8411, wait until TX queue is empty.
2011-11-23 22:07:13 +00:00
case RL_HWREV_8168DP:
sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_AUTOPAD |
To save more power, switch to 10/100Mbps link when controller is put into suspend/shutdown. Old PCI controllers performed that operation in firmware but for RTL8111C or newer controllers, it's responsibility of driver. It's not clear whether the firmware of RTL8111B still downgrades its speed to 10/100Mbps so leave it as it was.
2011-11-23 23:29:18 +00:00
RL_FLAG_JUMBOV2 | RL_FLAG_WAIT_TXPOLL | RL_FLAG_WOL_MANLINK;
Make sure to stop TX MAC before freeing queued TX frames. For RTL8111DP, check if the TX MAC is active by reading RL_GTXSTART register. For RTL8402/8168E-VL/8168F/8411, wait until TX queue is empty.
2011-11-23 22:07:13 +00:00
break;
Add preliminary support for 8168E/8111E PCIe controller. While I'm here simplify device description string. Tested by: Michael Beckmann < michael <> apfel dot de > MFC after: 5 days
2010-04-09 22:50:28 +00:00
case RL_HWREV_8168E:
sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM |
RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT |
To save more power, switch to 10/100Mbps link when controller is put into suspend/shutdown. Old PCI controllers performed that operation in firmware but for RTL8111C or newer controllers, it's responsibility of driver. It's not clear whether the firmware of RTL8111B still downgrades its speed to 10/100Mbps so leave it as it was.
2011-11-23 23:29:18 +00:00
RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
RL_FLAG_WOL_MANLINK;
Add preliminary support for 8168E/8111E PCIe controller. While I'm here simplify device description string. Tested by: Michael Beckmann < michael <> apfel dot de > MFC after: 5 days
2010-04-09 22:50:28 +00:00
break;
Add initial support for RTL8168E/8111E-VL PCIe GbE. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 02:23:50 +00:00
case RL_HWREV_8168E_VL:
Add preliminary support for RTL8168/8111F PCIe Gigabit ethernet. H/W donated by: RealTek Semiconductor Corp.
2011-11-17 22:07:50 +00:00
case RL_HWREV_8168F:
- Implement the RX EARLYOFF and RXDV GATED bits as done by RealTek's Linux driver as version 8.037.00 for RTL8168{E-VL,EP,F,G,GU} and RTL8111B. This makes reception of packets work with the RTL8168G (HW rev. 0x4c000000) in my Shuttle DS47. - Consistently use RL_MSI_MESSAGES. In joint forces with: yongari MFC after: 5 days
2014-02-05 23:13:40 +00:00
sc->rl_flags |= RL_FLAG_EARLYOFF;
/* FALLTHROUGH */
Add preliminary support for RTL8411 PCIe Gigabit ethernet with integrated card reader. H/W donated by: RealTek Semiconductor Corp.
2011-11-16 22:05:38 +00:00
case RL_HWREV_8411:
Add initial support for RTL8168E/8111E-VL PCIe GbE. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 02:23:50 +00:00
sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
Make sure to stop TX MAC before freeing queued TX frames. For RTL8111DP, check if the TX MAC is active by reading RL_GTXSTART register. For RTL8402/8168E-VL/8168F/8411, wait until TX queue is empty.
2011-11-23 22:07:13 +00:00
RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
To save more power, switch to 10/100Mbps link when controller is put into suspend/shutdown. Old PCI controllers performed that operation in firmware but for RTL8111C or newer controllers, it's responsibility of driver. It's not clear whether the firmware of RTL8111B still downgrades its speed to 10/100Mbps so leave it as it was.
2011-11-23 23:29:18 +00:00
RL_FLAG_CMDSTOP_WAIT_TXQ | RL_FLAG_WOL_MANLINK;
Add initial support for RTL8168E/8111E-VL PCIe GbE. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 02:23:50 +00:00
break;
- Implement the RX EARLYOFF and RXDV GATED bits as done by RealTek's Linux driver as version 8.037.00 for RTL8168{E-VL,EP,F,G,GU} and RTL8111B. This makes reception of packets work with the RTL8168G (HW rev. 0x4c000000) in my Shuttle DS47. - Consistently use RL_MSI_MESSAGES. In joint forces with: yongari MFC after: 5 days
2014-02-05 23:13:40 +00:00
case RL_HWREV_8168EP:
case RL_HWREV_8168G:
Add preliminary support for RTL8168H, tested by Sreenath Battalahalli. MFC after: 1 week
2015-09-13 22:22:30 +00:00
case RL_HWREV_8168H:
- Implement the RX EARLYOFF and RXDV GATED bits as done by RealTek's Linux driver as version 8.037.00 for RTL8168{E-VL,EP,F,G,GU} and RTL8111B. This makes reception of packets work with the RTL8168G (HW rev. 0x4c000000) in my Shuttle DS47. - Consistently use RL_MSI_MESSAGES. In joint forces with: yongari MFC after: 5 days
2014-02-05 23:13:40 +00:00
case RL_HWREV_8411B:
sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 |
RL_FLAG_CMDSTOP_WAIT_TXQ | RL_FLAG_WOL_MANLINK |
RL_FLAG_EARLYOFFV2 | RL_FLAG_RXDV_GATED;
break;
Add preliminary support for RTL8168G, RTL8168GU and RTL8411B. RTL8168GU has two variants(GMII and MII) but it uses the same chip revision id. Driver checks PCI device id of controller and sets internal capability flag(i.e. jumbo frame and link speed down in WOL). H/W donated by: RealTek Semiconductor Corp.
2013-10-29 05:30:21 +00:00
case RL_HWREV_8168GU:
if (pci_get_device(dev) == RT_DEVICEID_8101E) {
/* RTL8106EUS */
sc->rl_flags |= RL_FLAG_FASTETHER;
} else
sc->rl_flags |= RL_FLAG_JUMBOV2 | RL_FLAG_WOL_MANLINK;
sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
- Implement the RX EARLYOFF and RXDV GATED bits as done by RealTek's Linux driver as version 8.037.00 for RTL8168{E-VL,EP,F,G,GU} and RTL8111B. This makes reception of packets work with the RTL8168G (HW rev. 0x4c000000) in my Shuttle DS47. - Consistently use RL_MSI_MESSAGES. In joint forces with: yongari MFC after: 5 days
2014-02-05 23:13:40 +00:00
RL_FLAG_AUTOPAD | RL_FLAG_CMDSTOP_WAIT_TXQ |
RL_FLAG_EARLYOFFV2 | RL_FLAG_RXDV_GATED;
Add preliminary support for RTL8168G, RTL8168GU and RTL8411B. RTL8168GU has two variants(GMII and MII) but it uses the same chip revision id. Driver checks PCI device id of controller and sets internal capability flag(i.e. jumbo frame and link speed down in WOL). H/W donated by: RealTek Semiconductor Corp.
2013-10-29 05:30:21 +00:00
break;
Instead of allocating variables for each events/hardware flags, use a dedicated flag that represents controller capabilities/events. This will simplify many part of code that requires different workaround for each controller revisions and will enhance readability. While I'm here move PHY wakeup code up before mii_phy_probe() which seems to help to wake PHY in some cases.
2008-07-02 06:29:43 +00:00
case RL_HWREV_8169_8110SB:
Add a new RTL8169 variant, 8169SB(L)/8110SB(L). Reported by: nork Tested by: nork
2008-07-02 08:00:14 +00:00
case RL_HWREV_8169_8110SBL:
- Add support for 8110SCe part. Some magic registers were taken from Linux driver. - Swap hardware revisions for 8110S and 8169S as Linux driver claims. Reviewed by: yongari (early version)
2009-01-20 20:22:28 +00:00
case RL_HWREV_8169_8110SC:
case RL_HWREV_8169_8110SCE:
sc->rl_flags |= RL_FLAG_PHYWAKE;
/* FALLTHROUGH */
case RL_HWREV_8169:
case RL_HWREV_8169S:
case RL_HWREV_8110S:
sc->rl_flags |= RL_FLAG_MACRESET;
Instead of allocating variables for each events/hardware flags, use a dedicated flag that represents controller capabilities/events. This will simplify many part of code that requires different workaround for each controller revisions and will enhance readability. While I'm here move PHY wakeup code up before mii_phy_probe() which seems to help to wake PHY in some cases.
2008-07-02 06:29:43 +00:00
break;
default:
break;
}
Use correct Config registers for RTL8139 family. Unlike RTL8168 and RTL810x family , RTL8139 has different register map for Config registers. While here, follow the lead of re(4) in WOL configuration. - Disable WOL_UCAST and WOL_MCAST capabilities by default. - Config5 register write does not need to unlock EEPROM access on RTL8139 family but unlocking EEPROM access does not affect its operation and make it consistent with re(4). Reported by: Matt Renzelmann mjr <> cs dot wisc dot edu
2012-02-25 04:54:51 +00:00
if (sc->rl_hwrev->rl_rev == RL_HWREV_8139CPLUS) {
sc->rl_cfg0 = RL_8139_CFG0;
sc->rl_cfg1 = RL_8139_CFG1;
sc->rl_cfg2 = 0;
sc->rl_cfg3 = RL_8139_CFG3;
sc->rl_cfg4 = RL_8139_CFG4;
sc->rl_cfg5 = RL_8139_CFG5;
} else {
sc->rl_cfg0 = RL_CFG0;
sc->rl_cfg1 = RL_CFG1;
sc->rl_cfg2 = RL_CFG2;
sc->rl_cfg3 = RL_CFG3;
sc->rl_cfg4 = RL_CFG4;
sc->rl_cfg5 = RL_CFG5;
}
re_reset() should be called only after setting device specific features.
2011-01-13 22:52:57 +00:00
/* Reset the adapter. */
RL_LOCK(sc);
re_reset(sc);
RL_UNLOCK(sc);
Add basic support for RTL8168C, RTL8168CP, RTL8111C and RTL8111CP. ATM Tx/Rx checksum offload is supported but TSO and jumbo frame is not yet supported. Because these newer controllers use different descriptor formats, a flag RL_FLAG_DESCV2 flag was introduced to handle that case in Tx/Rx handler. Also newer controllers seems to require to not touch 'enable Tx/Rx bit' in RL_CPLUS_CMD register so don't blindly try to set that bits. Note, it seems that there is still power-saving related issue where driver fails to attach PHY. Rebooting seems to fix that issue but number of required reboots varys. Many thanks to users that helped during developement. I really appreciate their patient and test/feedbacks.
2008-07-02 07:54:53 +00:00
/* Enable PME. */
CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
Use correct Config registers for RTL8139 family. Unlike RTL8168 and RTL810x family , RTL8139 has different register map for Config registers. While here, follow the lead of re(4) in WOL configuration. - Disable WOL_UCAST and WOL_MCAST capabilities by default. - Config5 register write does not need to unlock EEPROM access on RTL8139 family but unlocking EEPROM access does not affect its operation and make it consistent with re(4). Reported by: Matt Renzelmann mjr <> cs dot wisc dot edu
2012-02-25 04:54:51 +00:00
cfg = CSR_READ_1(sc, sc->rl_cfg1);
Add basic support for RTL8168C, RTL8168CP, RTL8111C and RTL8111CP. ATM Tx/Rx checksum offload is supported but TSO and jumbo frame is not yet supported. Because these newer controllers use different descriptor formats, a flag RL_FLAG_DESCV2 flag was introduced to handle that case in Tx/Rx handler. Also newer controllers seems to require to not touch 'enable Tx/Rx bit' in RL_CPLUS_CMD register so don't blindly try to set that bits. Note, it seems that there is still power-saving related issue where driver fails to attach PHY. Rebooting seems to fix that issue but number of required reboots varys. Many thanks to users that helped during developement. I really appreciate their patient and test/feedbacks.
2008-07-02 07:54:53 +00:00
cfg |= RL_CFG1_PME;
Use correct Config registers for RTL8139 family. Unlike RTL8168 and RTL810x family , RTL8139 has different register map for Config registers. While here, follow the lead of re(4) in WOL configuration. - Disable WOL_UCAST and WOL_MCAST capabilities by default. - Config5 register write does not need to unlock EEPROM access on RTL8139 family but unlocking EEPROM access does not affect its operation and make it consistent with re(4). Reported by: Matt Renzelmann mjr <> cs dot wisc dot edu
2012-02-25 04:54:51 +00:00
CSR_WRITE_1(sc, sc->rl_cfg1, cfg);
cfg = CSR_READ_1(sc, sc->rl_cfg5);
Add basic support for RTL8168C, RTL8168CP, RTL8111C and RTL8111CP. ATM Tx/Rx checksum offload is supported but TSO and jumbo frame is not yet supported. Because these newer controllers use different descriptor formats, a flag RL_FLAG_DESCV2 flag was introduced to handle that case in Tx/Rx handler. Also newer controllers seems to require to not touch 'enable Tx/Rx bit' in RL_CPLUS_CMD register so don't blindly try to set that bits. Note, it seems that there is still power-saving related issue where driver fails to attach PHY. Rebooting seems to fix that issue but number of required reboots varys. Many thanks to users that helped during developement. I really appreciate their patient and test/feedbacks.
2008-07-02 07:54:53 +00:00
cfg &= RL_CFG5_PME_STS;
Use correct Config registers for RTL8139 family. Unlike RTL8168 and RTL810x family , RTL8139 has different register map for Config registers. While here, follow the lead of re(4) in WOL configuration. - Disable WOL_UCAST and WOL_MCAST capabilities by default. - Config5 register write does not need to unlock EEPROM access on RTL8139 family but unlocking EEPROM access does not affect its operation and make it consistent with re(4). Reported by: Matt Renzelmann mjr <> cs dot wisc dot edu
2012-02-25 04:54:51 +00:00
CSR_WRITE_1(sc, sc->rl_cfg5, cfg);
Add basic support for RTL8168C, RTL8168CP, RTL8111C and RTL8111CP. ATM Tx/Rx checksum offload is supported but TSO and jumbo frame is not yet supported. Because these newer controllers use different descriptor formats, a flag RL_FLAG_DESCV2 flag was introduced to handle that case in Tx/Rx handler. Also newer controllers seems to require to not touch 'enable Tx/Rx bit' in RL_CPLUS_CMD register so don't blindly try to set that bits. Note, it seems that there is still power-saving related issue where driver fails to attach PHY. Rebooting seems to fix that issue but number of required reboots varys. Many thanks to users that helped during developement. I really appreciate their patient and test/feedbacks.
2008-07-02 07:54:53 +00:00
CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
- For the 8169 chips, read the station address by forcing an EEPROM autoload and then copying the contends of the station address registers. For some reason, reading the EEPROM on the 8169S doesn't work right. This gets around the problem, and allows us to read the station address correctly on the 8169S. - Insert a delay after initiating packet transmition in re_diag() to allow lots of time for the frame to echo back to the host, and wait for both the 'RX complete' and 'timeout expired' bits in the ISR register to be set. - Deal more intelligently with the fact that the frame length field in the RX descriptor is a different width on the 8139C+ than it is on the 8169/8169S/8110S - For the 8169, you have to set bit 17 in the TX config register to enter digital loopback mode, but for the 8139C+, you have to set both bits 17 and 18. Take this into account so that re_diag() works properly for both types of chips.
2003-09-11 06:56:46 +00:00
Add basic support for RTL8168C, RTL8168CP, RTL8111C and RTL8111CP. ATM Tx/Rx checksum offload is supported but TSO and jumbo frame is not yet supported. Because these newer controllers use different descriptor formats, a flag RL_FLAG_DESCV2 flag was introduced to handle that case in Tx/Rx handler. Also newer controllers seems to require to not touch 'enable Tx/Rx bit' in RL_CPLUS_CMD register so don't blindly try to set that bits. Note, it seems that there is still power-saving related issue where driver fails to attach PHY. Rebooting seems to fix that issue but number of required reboots varys. Many thanks to users that helped during developement. I really appreciate their patient and test/feedbacks.
2008-07-02 07:54:53 +00:00
if ((sc->rl_flags & RL_FLAG_PAR) != 0) {
/*
* XXX Should have a better way to extract station
* address from EEPROM.
*/
for (i = 0; i < ETHER_ADDR_LEN; i++)
eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i);
} else {
sc->rl_eewidth = RL_9356_ADDR_LEN;
re_read_eeprom(sc, (caddr_t)&re_did, 0, 1);
if (re_did != 0x8129)
sc->rl_eewidth = RL_9346_ADDR_LEN;
/*
* Get station address from the EEPROM.
*/
re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3);
for (i = 0; i < ETHER_ADDR_LEN / 2; i++)
as[i] = le16toh(as[i]);
Fix size of the bcopy when extracting ethernet address Obtained from: DragonFly
2012-06-25 05:47:12 +00:00
bcopy(as, eaddr, ETHER_ADDR_LEN);
Add basic support for RTL8168C, RTL8168CP, RTL8111C and RTL8111CP. ATM Tx/Rx checksum offload is supported but TSO and jumbo frame is not yet supported. Because these newer controllers use different descriptor formats, a flag RL_FLAG_DESCV2 flag was introduced to handle that case in Tx/Rx handler. Also newer controllers seems to require to not touch 'enable Tx/Rx bit' in RL_CPLUS_CMD register so don't blindly try to set that bits. Note, it seems that there is still power-saving related issue where driver fails to attach PHY. Rebooting seems to fix that issue but number of required reboots varys. Many thanks to users that helped during developement. I really appreciate their patient and test/feedbacks.
2008-07-02 07:54:53 +00:00
}
- For the 8169 chips, read the station address by forcing an EEPROM autoload and then copying the contends of the station address registers. For some reason, reading the EEPROM on the 8169S doesn't work right. This gets around the problem, and allows us to read the station address correctly on the 8169S. - Insert a delay after initiating packet transmition in re_diag() to allow lots of time for the frame to echo back to the host, and wait for both the 'RX complete' and 'timeout expired' bits in the ISR register to be set. - Deal more intelligently with the fact that the frame length field in the RX descriptor is a different width on the 8139C+ than it is on the 8169/8169S/8110S - For the 8169, you have to set bit 17 in the TX config register to enter digital loopback mode, but for the 8139C+, you have to set both bits 17 and 18. Take this into account so that re_diag() works properly for both types of chips.
2003-09-11 06:56:46 +00:00
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
if (sc->rl_type == RL_8169) {
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
/* Set RX length mask and number of descriptors. */
- For the 8169 chips, read the station address by forcing an EEPROM autoload and then copying the contends of the station address registers. For some reason, reading the EEPROM on the 8169S doesn't work right. This gets around the problem, and allows us to read the station address correctly on the 8169S. - Insert a delay after initiating packet transmition in re_diag() to allow lots of time for the frame to echo back to the host, and wait for both the 'RX complete' and 'timeout expired' bits in the ISR register to be set. - Deal more intelligently with the fact that the frame length field in the RX descriptor is a different width on the 8139C+ than it is on the 8169/8169S/8110S - For the 8169, you have to set bit 17 in the TX config register to enter digital loopback mode, but for the 8139C+, you have to set both bits 17 and 18. Take this into account so that re_diag() works properly for both types of chips.
2003-09-11 06:56:46 +00:00
sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
sc->rl_txstart = RL_GTXSTART;
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
sc->rl_ldata.rl_tx_desc_cnt = RL_8169_TX_DESC_CNT;
sc->rl_ldata.rl_rx_desc_cnt = RL_8169_RX_DESC_CNT;
- For the 8169 chips, read the station address by forcing an EEPROM autoload and then copying the contends of the station address registers. For some reason, reading the EEPROM on the 8169S doesn't work right. This gets around the problem, and allows us to read the station address correctly on the 8169S. - Insert a delay after initiating packet transmition in re_diag() to allow lots of time for the frame to echo back to the host, and wait for both the 'RX complete' and 'timeout expired' bits in the ISR register to be set. - Deal more intelligently with the fact that the frame length field in the RX descriptor is a different width on the 8139C+ than it is on the 8169/8169S/8110S - For the 8169, you have to set bit 17 in the TX config register to enter digital loopback mode, but for the 8139C+, you have to set both bits 17 and 18. Take this into account so that re_diag() works properly for both types of chips.
2003-09-11 06:56:46 +00:00
} else {
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
/* Set RX length mask and number of descriptors. */
- For the 8169 chips, read the station address by forcing an EEPROM autoload and then copying the contends of the station address registers. For some reason, reading the EEPROM on the 8169S doesn't work right. This gets around the problem, and allows us to read the station address correctly on the 8169S. - Insert a delay after initiating packet transmition in re_diag() to allow lots of time for the frame to echo back to the host, and wait for both the 'RX complete' and 'timeout expired' bits in the ISR register to be set. - Deal more intelligently with the fact that the frame length field in the RX descriptor is a different width on the 8139C+ than it is on the 8169/8169S/8110S - For the 8169, you have to set bit 17 in the TX config register to enter digital loopback mode, but for the 8139C+, you have to set both bits 17 and 18. Take this into account so that re_diag() works properly for both types of chips.
2003-09-11 06:56:46 +00:00
sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
sc->rl_txstart = RL_TXSTART;
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
sc->rl_ldata.rl_tx_desc_cnt = RL_8139_TX_DESC_CNT;
sc->rl_ldata.rl_rx_desc_cnt = RL_8139_RX_DESC_CNT;
- For the 8169 chips, read the station address by forcing an EEPROM autoload and then copying the contends of the station address registers. For some reason, reading the EEPROM on the 8169S doesn't work right. This gets around the problem, and allows us to read the station address correctly on the 8169S. - Insert a delay after initiating packet transmition in re_diag() to allow lots of time for the frame to echo back to the host, and wait for both the 'RX complete' and 'timeout expired' bits in the ISR register to be set. - Deal more intelligently with the fact that the frame length field in the RX descriptor is a different width on the 8139C+ than it is on the 8169/8169S/8110S - For the 8169, you have to set bit 17 in the TX config register to enter digital loopback mode, but for the 8139C+, you have to set both bits 17 and 18. Take this into account so that re_diag() works properly for both types of chips.
2003-09-11 06:56:46 +00:00
}
Add a PHY driver to support the built-in gigE PHY in the 8169S/8110S ethernet chips. This driver is pretty simple, however it contains special DSP initialization code which is needed in order to get the chip to negotiate a gigE link. (This special initialization may not be needed in subsequent chip revs.) Also: - Fix typo in if_rlreg.h (RL_GMEDIASTAT_1000MPS -> RL_GMEDIASTAT_1000MBPS) - Deal with shared interrupts in re_intr(): if interface isn't up, return. - Fix another bug in re_gmii_writereg() (properly apply data field mask) - Allow PHY driver to read the RL_GMEDIASTAT register via the re_gmii_readreg() register (this is register needed to determine real time link/media status).
2003-09-11 03:53:46 +00:00
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
error = re_allocmem(dev, sc);
if (error)
goto fail;
Add simple MAC statistics counter reading support. Unfortunately useful counters like rl_missed_pkts is 16 bits quantity which is too small to hold meaningful information happened in a second. This means driver should frequently read these counters in order not to lose accuracy and that approach is too inefficient in driver's view. Moreover it seems there is no way to trigger an interrupt to detect counter near-full or wraparound event as well as lacking clearing the MAC counters. Another limitation of reading the counters from RealTek controllers is lack of interrupt firing at the end of DMA cycle of MAC counter read request such that driver have to poll the end of the DMA which is a time consuming process as well as inefficient. The more severe issue of the MAC counter read request is it takes too long to complete the DMA. All these limitation made maintaining MAC counters in driver impractical. For now, just provide simple sysctl interface to trigger reading the MAC counters. These counters could be used to track down driver issues. Users can read MAC counters maintained in controller with the following command. #sysctl dev.re.0.stats=1 While I'm here add check for validity of dma map and allocated memory before unloading/freeing them. Tested by: rmacklem
2010-11-05 19:28:00 +00:00
re_add_sysctls(sc);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Move if_alloc() up so it's before mii_phy_probe().
2005-06-11 01:37:46 +00:00
ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
if (ifp == NULL) {
- Use if_printf() and device_printf() in re(4) and remove rl_unit from the softc. - Use callout_init_mtx() and rather than timeout/untimeout in both rl(4) and re(4). - Fix locking for ifmedia by locking the driver in the ifmedia handlers rather than in the miibus functions. (re(4) didn't lock the mii stuff at all!) - Fix some locking in re_ioctl(). Note: the two drivers share the same softc declared in if_rlreg.h, so they had to be change simultaneously. MFC after: 1 week Tested by: several on rl(4), none on re(4)
2005-09-29 16:52:19 +00:00
device_printf(dev, "can not if_alloc()\n");
Move if_alloc() up so it's before mii_phy_probe().
2005-06-11 01:37:46 +00:00
error = ENOSPC;
goto fail;
}
For RTL8168C SPIN2 controllers, make sure to take the controller out of sleep mode prior to accessing to PHY. This should fix device attach failure seen on these controllers. Also enable the sleep mode when device is put into sleep state. PR: kern/123123, kern/123053
2008-12-17 06:01:03 +00:00
/* Take controller out of deep sleep mode. */
if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
CSR_WRITE_1(sc, RL_GPIO,
CSR_READ_1(sc, RL_GPIO) | 0x01);
else
CSR_WRITE_1(sc, RL_GPIO,
CSR_READ_1(sc, RL_GPIO) & ~0x01);
}
Instead of allocating variables for each events/hardware flags, use a dedicated flag that represents controller capabilities/events. This will simplify many part of code that requires different workaround for each controller revisions and will enhance readability. While I'm here move PHY wakeup code up before mii_phy_probe() which seems to help to wake PHY in some cases.
2008-07-02 06:29:43 +00:00
/* Take PHY out of power down mode. */
Add initial support for RTL8401E PCIe Fast Ethernet. PR: 154789
2011-02-16 21:59:42 +00:00
if ((sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0) {
Add preliminary support for 8168E/8111E PCIe controller. While I'm here simplify device description string. Tested by: Michael Beckmann < michael <> apfel dot de > MFC after: 5 days
2010-04-09 22:50:28 +00:00
CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) | 0x80);
Add initial support for RTL8401E PCIe Fast Ethernet. PR: 154789
2011-02-16 21:59:42 +00:00
if (hw_rev->rl_rev == RL_HWREV_8401E)
CSR_WRITE_1(sc, 0xD1, CSR_READ_1(sc, 0xD1) & ~0x08);
}
Instead of allocating variables for each events/hardware flags, use a dedicated flag that represents controller capabilities/events. This will simplify many part of code that requires different workaround for each controller revisions and will enhance readability. While I'm here move PHY wakeup code up before mii_phy_probe() which seems to help to wake PHY in some cases.
2008-07-02 06:29:43 +00:00
if ((sc->rl_flags & RL_FLAG_PHYWAKE) != 0) {
re_gmii_writereg(dev, 1, 0x1f, 0);
re_gmii_writereg(dev, 1, 0x0e, 0);
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
ifp->if_softc = sc;
Replace the if_name and if_unit members of struct ifnet with new members if_xname, if_dname, and if_dunit. if_xname is the name of the interface and if_dname/unit are the driver name and instance. This change paves the way for interface renaming and enhanced pseudo device creation and configuration symantics. Approved By: re (in principle) Reviewed By: njl, imp Tested On: i386, amd64, sparc64 Obtained From: NetBSD (if_xname)
2003-10-31 18:32:15 +00:00
if_initname(ifp, device_get_name(dev), device_get_unit(dev));
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
ifp->if_ioctl = re_ioctl;
ifp->if_start = re_start;
Disable TX IP checksum offloading for RTL8168C controllers. The controller in question generates frames with bad IP checksum value if packets contain IP options. For instance, packets generated by ping(8) with record route option have wrong IP checksum value. The controller correctly computes checksum for normal TCP/UDP packets though. There are two known RTL8168/8111C variants in market and the issue I observed happened on RL_HWREV_8168C_SPIN2. I'm not sure RL_HWREV_8168C also has the same issue but it would be better to assume it has the same issue since they shall share same core. RTL8102E which is supposed to be released at the time of RTL8168/8111C announcement does not have the issue. Tested by: Konstantin V. Krotov ( kkv <> insysnet dot ru )
2011-02-04 17:49:55 +00:00
/*
* RTL8168/8111C generates wrong IP checksummed frame if the
Disable TX IP/TCP/UDP checksum offloading for RTL8168C/RTL8168CP. Previously only TX IP checksum offloading was disabled but it's reported that TX checksum offloading for UDP datagrams with IP options also generates corrupted frames. Reporter's controller is RTL8168CP but I guess RTL8168C also have the same issue since it shall share the same core. Reported and tested by: tuexen
2014-05-13 05:19:29 +00:00
* packet has IP options so disable TX checksum offloading.
Disable TX IP checksum offloading for RTL8168C controllers. The controller in question generates frames with bad IP checksum value if packets contain IP options. For instance, packets generated by ping(8) with record route option have wrong IP checksum value. The controller correctly computes checksum for normal TCP/UDP packets though. There are two known RTL8168/8111C variants in market and the issue I observed happened on RL_HWREV_8168C_SPIN2. I'm not sure RL_HWREV_8168C also has the same issue but it would be better to assume it has the same issue since they shall share same core. RTL8102E which is supposed to be released at the time of RTL8168/8111C announcement does not have the issue. Tested by: Konstantin V. Krotov ( kkv <> insysnet dot ru )
2011-02-04 17:49:55 +00:00
*/
if (sc->rl_hwrev->rl_rev == RL_HWREV_8168C ||
Disable TX IP header checksum offloading on RL_HWREV_8168CP. The controller generates wrong checksummed frame if the IP packet has IP options. Submitted by: Alexander Bluhm via brad@openbsd
2013-03-13 02:11:45 +00:00
sc->rl_hwrev->rl_rev == RL_HWREV_8168C_SPIN2 ||
Disable TX IP/TCP/UDP checksum offloading for RTL8168C/RTL8168CP. Previously only TX IP checksum offloading was disabled but it's reported that TX checksum offloading for UDP datagrams with IP options also generates corrupted frames. Reporter's controller is RTL8168CP but I guess RTL8168C also have the same issue since it shall share the same core. Reported and tested by: tuexen
2014-05-13 05:19:29 +00:00
sc->rl_hwrev->rl_rev == RL_HWREV_8168CP) {
ifp->if_hwassist = 0;
ifp->if_capabilities = IFCAP_RXCSUM | IFCAP_TSO4;
} else {
Disable TX IP checksum offloading for RTL8168C controllers. The controller in question generates frames with bad IP checksum value if packets contain IP options. For instance, packets generated by ping(8) with record route option have wrong IP checksum value. The controller correctly computes checksum for normal TCP/UDP packets though. There are two known RTL8168/8111C variants in market and the issue I observed happened on RL_HWREV_8168C_SPIN2. I'm not sure RL_HWREV_8168C also has the same issue but it would be better to assume it has the same issue since they shall share same core. RTL8102E which is supposed to be released at the time of RTL8168/8111C announcement does not have the issue. Tested by: Konstantin V. Krotov ( kkv <> insysnet dot ru )
2011-02-04 17:49:55 +00:00
ifp->if_hwassist = CSUM_IP | CSUM_TCP | CSUM_UDP;
Disable TX IP/TCP/UDP checksum offloading for RTL8168C/RTL8168CP. Previously only TX IP checksum offloading was disabled but it's reported that TX checksum offloading for UDP datagrams with IP options also generates corrupted frames. Reporter's controller is RTL8168CP but I guess RTL8168C also have the same issue since it shall share the same core. Reported and tested by: tuexen
2014-05-13 05:19:29 +00:00
ifp->if_capabilities = IFCAP_HWCSUM | IFCAP_TSO4;
}
Disable TX IP checksum offloading for RTL8168C controllers. The controller in question generates frames with bad IP checksum value if packets contain IP options. For instance, packets generated by ping(8) with record route option have wrong IP checksum value. The controller correctly computes checksum for normal TCP/UDP packets though. There are two known RTL8168/8111C variants in market and the issue I observed happened on RL_HWREV_8168C_SPIN2. I'm not sure RL_HWREV_8168C also has the same issue but it would be better to assume it has the same issue since they shall share same core. RTL8102E which is supposed to be released at the time of RTL8168/8111C announcement does not have the issue. Tested by: Konstantin V. Krotov ( kkv <> insysnet dot ru )
2011-02-04 17:49:55 +00:00
ifp->if_hwassist |= CSUM_TSO;
Fix the following bugs in re(4) - Correct the PCI ID for the 8169SC/8110SC in the device list (I added the macro for it to if_rlreg.h before, but forgot to use it.) - Remove the extra interrupt spinlock I added previously. After giving it some more thought, it's not really needed. - Work around a hardware bug in some versions of the 8169. When sending very small IP datagrams with checksum offload enabled, a conflict can occur between the TX autopadding feature and the hardware checksumming that can corrupt the outbound packet. This is the reason that checksum offload sometimes breaks NFS: if you're using NFS over UDP, and you're very unlucky, you might find yourself doing a fragmented NFS write where the last fragment is smaller than the minimum ethernet frame size (60 bytes). (It's rare, but if you keep NFS running long enough it'll happen.) If checksum offload is enabled, the chip will have to both autopad the fragment and calculate its checksum header. This confuses some revs of the 8169, causing the packet that appears on the wire to be corrupted. (The IP addresses and the checksum field are mangled.) This will cause the NFS write to fail. Unfortunately, when NFS retries, it sends the same write request over and over again, and it keeps failing, so NFS stays wedged. (A simple way to provoke the failure is to connect the failing system to a network with a known good machine and do "ping -s 1473 <badhost>" from the good system. The ping will fail.) Someone had previously worked around this using the heavy-handed approahch of just disabling checksum offload. The correct fix is to manually pad short frames where the TCP/IP stack has requested checksum offloading. This allows us to have checksum offload turned on by default but still let NFS work right. - Not a bug, but change the ID strings for devices with hardware rev 0x30000000 and 0x38000000 to both be 8168B/8111B. According to RealTek, they're both the same device, but 0x30000000 is an earlier silicon spin.
2006-07-30 23:25:21 +00:00
ifp->if_capenable = ifp->if_capabilities;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
ifp->if_init = re_init;
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
ALTQ support for re(4). Submitted by: Chris Dionissopoulos, Theo Schlossnagle PR: kern/78681 MFC after: 2 weeks
2005-03-12 17:35:37 +00:00
ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN;
IFQ_SET_READY(&ifp->if_snd);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc);
Prefer RL_GMEDIASTAT register to RGEPHY_MII_SSR register to extract a link status of PHY when parent driver is re(4). RGEPHY_MII_SSR register does not seem to report correct PHY status on some integrated PHYs used with re(4). Unfortunately, RealTek PHYs have no additional information to differentiate integrated PHYs from external ones so relying on PHY model number is not enough to know that. However, it seems RGEPHY_MII_SSR register exists for external RealTek PHYs so checking parent driver would be good indication to know which PHY was used. In other words, for non-re(4) controllers, the PHY is external one and its revision number is greater than or equal to 2. This change fixes intermittent link UP/DOWN messages reported on RTL8169 controller. Also, mii_attach(9) is tried after setting interface name since rgephy(4) have to know parent driver name. PR: kern/165509
2012-02-28 05:23:29 +00:00
#define RE_PHYAD_INTERNAL 0
/* Do MII setup. */
phy = RE_PHYAD_INTERNAL;
if (sc->rl_type == RL_8169)
phy = 1;
error = mii_attach(dev, &sc->rl_miibus, ifp, re_ifmedia_upd,
re_ifmedia_sts, BMSR_DEFCAPMASK, phy, MII_OFFSET_ANY, MIIF_DOPAUSE);
if (error != 0) {
device_printf(dev, "attaching PHYs failed\n");
goto fail;
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/*
* Call MI attach routine.
*/
ether_ifattach(ifp, eaddr);
o Correctly set IFCAP_VLAN_HWCSUM as re(4) can do VLAN tagging/checksum offloading in hardware. o Correctly set media header length for VLAN.
2006-11-21 04:23:52 +00:00
/* VLAN capability setup */
ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
if (ifp->if_capabilities & IFCAP_HWCSUM)
ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
Implement WOL. Tested by: Fabian Keil ( freebsd-listen AT fabienkeli DOT de )
2008-03-03 03:33:58 +00:00
/* Enable WOL if PM is supported. */
Do a sweep of the tree replacing calls to pci_find_extcap() with calls to pci_find_cap() instead.
2011-03-23 13:10:15 +00:00
if (pci_find_cap(sc->rl_dev, PCIY_PMG, &reg) == 0)
Implement WOL. Tested by: Fabian Keil ( freebsd-listen AT fabienkeli DOT de )
2008-03-03 03:33:58 +00:00
ifp->if_capabilities |= IFCAP_WOL;
o Correctly set IFCAP_VLAN_HWCSUM as re(4) can do VLAN tagging/checksum offloading in hardware. o Correctly set media header length for VLAN.
2006-11-21 04:23:52 +00:00
ifp->if_capenable = ifp->if_capabilities;
Writing access to RL_CFG5 register also requires EEPROM write access. While I'm here, enable WOL through magic packet but disable waking up system via unicast, multicast and broadcast frames. Otherwise, multicast or unicast frame(e.g. ICMP echo request) can wake up system which is not probably wanted behavior on most environments. This was not known as problem because RL_CFG5 register access had not effect until this change. The capability to wake up system with unicast/multicast frames are still set in driver, default off, so users who need that feature can still activate it with ifconfig(8).
2011-11-22 23:19:49 +00:00
ifp->if_capenable &= ~(IFCAP_WOL_UCAST | IFCAP_WOL_MCAST);
Don't enable TSO by default. Users of RTL8169/8110 reported watchdog timeout issues and the root cause seems to stem from silicon bug of controller. Personally I couldn't reproduce it on RTL8169 controller but it seems it's dependent on usage pattern. For newer PCIe based controllers I have no TSO complaints but turning off TSO would be more safe. Users who are sure that their controller works with TSO can still reenable the TSO with ifconfig(8). Reported by: Oliver Lehmann (lehmann at ans-netz dot de), Eugene Butusov (ebutusov at gmail dot com)
2008-08-04 02:34:40 +00:00
/*
Disable TSO for all Realtek controllers. Experimentation showed RTL8111C generated corrupted frames where TCP option header was broken. All other sample controllers I have did not show such problem so it could be RTL8111C specific issue. Because there are too many variants it's hard to tell how many controllers have such issue. Just disable TSO by default but have user override it.
2011-01-25 19:05:46 +00:00
* Don't enable TSO by default. It is known to generate
* corrupted TCP segments(bad TCP options) under certain
Correct comment typos.
2013-06-28 14:41:06 +00:00
* circumstances.
Don't enable TSO by default. Users of RTL8169/8110 reported watchdog timeout issues and the root cause seems to stem from silicon bug of controller. Personally I couldn't reproduce it on RTL8169 controller but it seems it's dependent on usage pattern. For newer PCIe based controllers I have no TSO complaints but turning off TSO would be more safe. Users who are sure that their controller works with TSO can still reenable the TSO with ifconfig(8). Reported by: Oliver Lehmann (lehmann at ans-netz dot de), Eugene Butusov (ebutusov at gmail dot com)
2008-08-04 02:34:40 +00:00
*/
Disable TSO for all Realtek controllers. Experimentation showed RTL8111C generated corrupted frames where TCP option header was broken. All other sample controllers I have did not show such problem so it could be RTL8111C specific issue. Because there are too many variants it's hard to tell how many controllers have such issue. Just disable TSO by default but have user override it.
2011-01-25 19:05:46 +00:00
ifp->if_hwassist &= ~CSUM_TSO;
ifp->if_capenable &= ~(IFCAP_TSO4 | IFCAP_VLAN_HWTSO);
o Correctly set IFCAP_VLAN_HWCSUM as re(4) can do VLAN tagging/checksum offloading in hardware. o Correctly set media header length for VLAN.
2006-11-21 04:23:52 +00:00
#ifdef DEVICE_POLLING
ifp->if_capabilities |= IFCAP_POLLING;
#endif
/*
* Tell the upper layer(s) we support long frames.
* Must appear after the call to ether_ifattach() because
* ether_ifattach() sets ifi_hdrlen to the default value.
*/
Use define from if_var.h to access a field inside struct if_data, that resides in struct ifnet. Sponsored by: Nginx, Inc.
2014-08-30 19:55:54 +00:00
ifp->if_hdrlen = sizeof(struct ether_vlan_header);
o Correctly set IFCAP_VLAN_HWCSUM as re(4) can do VLAN tagging/checksum offloading in hardware. o Correctly set media header length for VLAN.
2006-11-21 04:23:52 +00:00
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#ifdef DEV_NETMAP
re_netmap_attach(sc);
#endif /* DEV_NETMAP */
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
#ifdef RE_DIAG
/*
* Perform hardware diagnostic on the original RTL8169.
* Some 32-bit cards were incorrectly wired and would
* malfunction if plugged into a 64-bit slot.
*/
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
if (hwrev == RL_HWREV_8169) {
error = re_diag(sc);
if (error) {
device_printf(dev,
"attach aborted due to hardware diag failure\n");
ether_ifdetach(ifp);
goto fail;
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
#endif
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Do not use interrupt taskqueue on controllers with MSI/MSI-X capability. One of reason using interrupt taskqueue in re(4) was to reduce number of TX/RX interrupts under load because re(4) controllers have no good TX/RX interrupt moderation mechanism. Basic TX interrupt moderation is done by hardware for most controllers but RX interrupt moderation through undocumented register showed poor RX performance so it was disabled in r215025. Using taskqueue to handle RX interrupt greatly reduced number of interrupts but re(4) consumed all available CPU cycles to run the taskqueue under high TX/RX network load. This can happen even with RTL810x fast ethernet controller and I believe this is not acceptable for most systems. To mitigate the issue, use one-shot timer register to moderate RX interrupts. The timer register provides programmable one-shot timer and can be used to suppress interrupt generation. The timer runs at 125MHZ on PCIe controllers so the minimum time allowed for the timer is 8ns. Data sheet says the register is 32 bits but experimentation shows only lower 13 bits are valid so maximum time that can be programmed is 65.528us. This yields theoretical maximum number of RX interrupts that could be generated per second is about 15260. Combined with TX completion interrupts re(4) shall generate less than 20k interrupts. This number is still slightly high compared to other intelligent ethernet controllers but system is very responsive even under high network load. Introduce sysctl variable dev.re.%d.int_rx_mod that controls amount of time to delay RX interrupt processing in units of us. Value 0 completely disables RX interrupt moderation. To provide old behavior for controllers that have MSI/MSI-X capability, introduce a new tunable hw.re.intr_filter. If the tunable is set to non-zero value, driver will use interrupt taskqueue. The default value of the tunable is 0. This tunable has no effect on controllers that has no MSI/MSI-X capability or if MSI/MSI-X is explicitly disabled by administrator. While I'm here cleanup interrupt setup/teardown since re(4) uses single MSI/MSI-X message at this moment.
2011-01-26 20:25:40 +00:00
#ifdef RE_TX_MODERATION
intr_filter = 1;
#endif
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/* Hook interrupt last to avoid having to lock softc */
Do not use interrupt taskqueue on controllers with MSI/MSI-X capability. One of reason using interrupt taskqueue in re(4) was to reduce number of TX/RX interrupts under load because re(4) controllers have no good TX/RX interrupt moderation mechanism. Basic TX interrupt moderation is done by hardware for most controllers but RX interrupt moderation through undocumented register showed poor RX performance so it was disabled in r215025. Using taskqueue to handle RX interrupt greatly reduced number of interrupts but re(4) consumed all available CPU cycles to run the taskqueue under high TX/RX network load. This can happen even with RTL810x fast ethernet controller and I believe this is not acceptable for most systems. To mitigate the issue, use one-shot timer register to moderate RX interrupts. The timer register provides programmable one-shot timer and can be used to suppress interrupt generation. The timer runs at 125MHZ on PCIe controllers so the minimum time allowed for the timer is 8ns. Data sheet says the register is 32 bits but experimentation shows only lower 13 bits are valid so maximum time that can be programmed is 65.528us. This yields theoretical maximum number of RX interrupts that could be generated per second is about 15260. Combined with TX completion interrupts re(4) shall generate less than 20k interrupts. This number is still slightly high compared to other intelligent ethernet controllers but system is very responsive even under high network load. Introduce sysctl variable dev.re.%d.int_rx_mod that controls amount of time to delay RX interrupt processing in units of us. Value 0 completely disables RX interrupt moderation. To provide old behavior for controllers that have MSI/MSI-X capability, introduce a new tunable hw.re.intr_filter. If the tunable is set to non-zero value, driver will use interrupt taskqueue. The default value of the tunable is 0. This tunable has no effect on controllers that has no MSI/MSI-X capability or if MSI/MSI-X is explicitly disabled by administrator. While I'm here cleanup interrupt setup/teardown since re(4) uses single MSI/MSI-X message at this moment.
2011-01-26 20:25:40 +00:00
if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
intr_filter == 0) {
error = bus_setup_intr(dev, sc->rl_irq[0],
INTR_TYPE_NET | INTR_MPSAFE, NULL, re_intr_msi, sc,
&sc->rl_intrhand[0]);
} else {
Add MSI support. Ever since switching to adaptive polling re(4) occasionally spews watchdog timeouts on systems with MSI capability. This change is minimal one for supporting MSI and re(4) also needs MSIX support for RTL8111C in future. Because softc structure of re(4) is shared with rl(4), rl(4) was touched to use the modified softc. Reported by: cnst Tested by: cnst Approved by: re (kensmith)
2007-07-24 01:24:03 +00:00
error = bus_setup_intr(dev, sc->rl_irq[0],
INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc,
&sc->rl_intrhand[0]);
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
if (error) {
- Use if_printf() and device_printf() in re(4) and remove rl_unit from the softc. - Use callout_init_mtx() and rather than timeout/untimeout in both rl(4) and re(4). - Fix locking for ifmedia by locking the driver in the ifmedia handlers rather than in the miibus functions. (re(4) didn't lock the mii stuff at all!) - Fix some locking in re_ioctl(). Note: the two drivers share the same softc declared in if_rlreg.h, so they had to be change simultaneously. MFC after: 1 week Tested by: several on rl(4), none on re(4)
2005-09-29 16:52:19 +00:00
device_printf(dev, "couldn't set up irq\n");
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
ether_ifdetach(ifp);
}
fail:
- Use the re_tick() callout instead of if_slowtimo() for driving re_watchdog() in order to avoid races accessing if_timer. - Use bus_get_dma_tag() so re(4) works on platforms requiring it. - Remove invalid BUS_DMA_ALLOCNOW when creating the parent DMA tag and the tags that are used for static memory allocations. - Don't bother to set if_mtu to ETHERMTU, ether_ifattach() does that. - Remove an unused variable in re_intr().
2006-12-20 02:13:59 +00:00
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
if (error)
re_detach(dev);
return (error);
}
/*
* Shutdown hardware and free up resources. This can be called any
* time after the mutex has been initialized. It is called in both
* the error case in attach and the normal detach case so it needs
* to be careful about only freeing resources that have actually been
* allocated.
*/
static int
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_detach(device_t dev)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
struct rl_softc *sc;
struct ifnet *ifp;
Add MSI support. Ever since switching to adaptive polling re(4) occasionally spews watchdog timeouts on systems with MSI capability. This change is minimal one for supporting MSI and re(4) also needs MSIX support for RTL8111C in future. Because softc structure of re(4) is shared with rl(4), rl(4) was touched to use the modified softc. Reported by: cnst Tested by: cnst Approved by: re (kensmith)
2007-07-24 01:24:03 +00:00
int i, rid;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
sc = device_get_softc(dev);
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
ifp = sc->rl_ifp;
trim trailing white space.. call the re mutex by it's name.. MFC after: 3 days
2004-09-20 06:33:37 +00:00
KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
/* These should only be active if attach succeeded */
re_detach() fixes: - Fixed if_free() logic screw-up that can either result in freeing a NULL pointer or leaking "struct ifnet". - Move if_free() after re_stop(); the latter accesses "struct ifnet". This bug was masked by a previous bug. - Restore the fix for a panic on detach caused by racing with BPF detach code by Bill by moving ether_ifdetach() after re_stop() and resetting IFF_UP; this got screwed up in revs. 1.30 and 1.36.
2005-09-15 18:59:34 +00:00
if (device_is_attached(dev)) {
In device detach don't access ifnet structure unless device is attached. This fixes NULL pointer dereference when polling(9) is active and unsupported hardware is detected in device attach.
2008-07-02 06:41:46 +00:00
#ifdef DEVICE_POLLING
if (ifp->if_capenable & IFCAP_POLLING)
ether_poll_deregister(ifp);
#endif
re_detach() fixes: - Fixed if_free() logic screw-up that can either result in freeing a NULL pointer or leaking "struct ifnet". - Move if_free() after re_stop(); the latter accesses "struct ifnet". This bug was masked by a previous bug. - Restore the fix for a panic on detach caused by racing with BPF detach code by Bill by moving ether_ifdetach() after re_stop() and resetting IFF_UP; this got screwed up in revs. 1.30 and 1.36.
2005-09-15 18:59:34 +00:00
RL_LOCK(sc);
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
#if 0
re_detach() fixes: - Fixed if_free() logic screw-up that can either result in freeing a NULL pointer or leaking "struct ifnet". - Move if_free() after re_stop(); the latter accesses "struct ifnet". This bug was masked by a previous bug. - Restore the fix for a panic on detach caused by racing with BPF detach code by Bill by moving ether_ifdetach() after re_stop() and resetting IFF_UP; this got screwed up in revs. 1.30 and 1.36.
2005-09-15 18:59:34 +00:00
sc->suspended = 1;
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
#endif
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
re_stop(sc);
re_detach() fixes: - Fixed if_free() logic screw-up that can either result in freeing a NULL pointer or leaking "struct ifnet". - Move if_free() after re_stop(); the latter accesses "struct ifnet". This bug was masked by a previous bug. - Restore the fix for a panic on detach caused by racing with BPF detach code by Bill by moving ether_ifdetach() after re_stop() and resetting IFF_UP; this got screwed up in revs. 1.30 and 1.36.
2005-09-15 18:59:34 +00:00
RL_UNLOCK(sc);
- Use if_printf() and device_printf() in re(4) and remove rl_unit from the softc. - Use callout_init_mtx() and rather than timeout/untimeout in both rl(4) and re(4). - Fix locking for ifmedia by locking the driver in the ifmedia handlers rather than in the miibus functions. (re(4) didn't lock the mii stuff at all!) - Fix some locking in re_ioctl(). Note: the two drivers share the same softc declared in if_rlreg.h, so they had to be change simultaneously. MFC after: 1 week Tested by: several on rl(4), none on re(4)
2005-09-29 16:52:19 +00:00
callout_drain(&sc->rl_stat_callout);
Use taskqueue_drain() to wait for any pending tasks to complete rather than just pausing for a second.
2007-02-27 18:45:37 +00:00
taskqueue_drain(taskqueue_fast, &sc->rl_inttask);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/*
* Force off the IFF_UP flag here, in case someone
* still had a BPF descriptor attached to this
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
* interface. If they do, ether_ifdetach() will cause
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
* the BPF code to try and clear the promisc mode
* flag, which will bubble down to re_ioctl(),
* which will try to call re_init() again. This will
* turn the NIC back on and restart the MII ticker,
* which will panic the system when the kernel tries
* to invoke the re_tick() function that isn't there
* anymore.
*/
ifp->if_flags &= ~IFF_UP;
re_detach() fixes: - Fixed if_free() logic screw-up that can either result in freeing a NULL pointer or leaking "struct ifnet". - Move if_free() after re_stop(); the latter accesses "struct ifnet". This bug was masked by a previous bug. - Restore the fix for a panic on detach caused by racing with BPF detach code by Bill by moving ether_ifdetach() after re_stop() and resetting IFF_UP; this got screwed up in revs. 1.30 and 1.36.
2005-09-15 18:59:34 +00:00
ether_ifdetach(ifp);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
if (sc->rl_miibus)
device_delete_child(dev, sc->rl_miibus);
bus_generic_detach(dev);
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
/*
* The rest is resource deallocation, so we should already be
* stopped here.
*/
Do not use interrupt taskqueue on controllers with MSI/MSI-X capability. One of reason using interrupt taskqueue in re(4) was to reduce number of TX/RX interrupts under load because re(4) controllers have no good TX/RX interrupt moderation mechanism. Basic TX interrupt moderation is done by hardware for most controllers but RX interrupt moderation through undocumented register showed poor RX performance so it was disabled in r215025. Using taskqueue to handle RX interrupt greatly reduced number of interrupts but re(4) consumed all available CPU cycles to run the taskqueue under high TX/RX network load. This can happen even with RTL810x fast ethernet controller and I believe this is not acceptable for most systems. To mitigate the issue, use one-shot timer register to moderate RX interrupts. The timer register provides programmable one-shot timer and can be used to suppress interrupt generation. The timer runs at 125MHZ on PCIe controllers so the minimum time allowed for the timer is 8ns. Data sheet says the register is 32 bits but experimentation shows only lower 13 bits are valid so maximum time that can be programmed is 65.528us. This yields theoretical maximum number of RX interrupts that could be generated per second is about 15260. Combined with TX completion interrupts re(4) shall generate less than 20k interrupts. This number is still slightly high compared to other intelligent ethernet controllers but system is very responsive even under high network load. Introduce sysctl variable dev.re.%d.int_rx_mod that controls amount of time to delay RX interrupt processing in units of us. Value 0 completely disables RX interrupt moderation. To provide old behavior for controllers that have MSI/MSI-X capability, introduce a new tunable hw.re.intr_filter. If the tunable is set to non-zero value, driver will use interrupt taskqueue. The default value of the tunable is 0. This tunable has no effect on controllers that has no MSI/MSI-X capability or if MSI/MSI-X is explicitly disabled by administrator. While I'm here cleanup interrupt setup/teardown since re(4) uses single MSI/MSI-X message at this moment.
2011-01-26 20:25:40 +00:00
if (sc->rl_intrhand[0] != NULL) {
bus_teardown_intr(dev, sc->rl_irq[0], sc->rl_intrhand[0]);
sc->rl_intrhand[0] = NULL;
Add MSI support. Ever since switching to adaptive polling re(4) occasionally spews watchdog timeouts on systems with MSI capability. This change is minimal one for supporting MSI and re(4) also needs MSIX support for RTL8111C in future. Because softc structure of re(4) is shared with rl(4), rl(4) was touched to use the modified softc. Reported by: cnst Tested by: cnst Approved by: re (kensmith)
2007-07-24 01:24:03 +00:00
}
Make sure we don't dereference a null pointer
2012-05-11 07:12:18 +00:00
if (ifp != NULL) {
#ifdef DEV_NETMAP
netmap_detach(ifp);
#endif /* DEV_NETMAP */
Make sure that we call if_free(ifp) after bus_teardown_intr. Since we could get an interrupt after we free the ifp, and the interrupt handler depended on the ifp being still alive, this could, in theory, cause a crash. Eliminate this possibility by moving the if_free to after the bus_teardown_intr() call.
2005-09-19 03:10:21 +00:00
if_free(ifp);
Make sure we don't dereference a null pointer
2012-05-11 07:12:18 +00:00
}
Do not use interrupt taskqueue on controllers with MSI/MSI-X capability. One of reason using interrupt taskqueue in re(4) was to reduce number of TX/RX interrupts under load because re(4) controllers have no good TX/RX interrupt moderation mechanism. Basic TX interrupt moderation is done by hardware for most controllers but RX interrupt moderation through undocumented register showed poor RX performance so it was disabled in r215025. Using taskqueue to handle RX interrupt greatly reduced number of interrupts but re(4) consumed all available CPU cycles to run the taskqueue under high TX/RX network load. This can happen even with RTL810x fast ethernet controller and I believe this is not acceptable for most systems. To mitigate the issue, use one-shot timer register to moderate RX interrupts. The timer register provides programmable one-shot timer and can be used to suppress interrupt generation. The timer runs at 125MHZ on PCIe controllers so the minimum time allowed for the timer is 8ns. Data sheet says the register is 32 bits but experimentation shows only lower 13 bits are valid so maximum time that can be programmed is 65.528us. This yields theoretical maximum number of RX interrupts that could be generated per second is about 15260. Combined with TX completion interrupts re(4) shall generate less than 20k interrupts. This number is still slightly high compared to other intelligent ethernet controllers but system is very responsive even under high network load. Introduce sysctl variable dev.re.%d.int_rx_mod that controls amount of time to delay RX interrupt processing in units of us. Value 0 completely disables RX interrupt moderation. To provide old behavior for controllers that have MSI/MSI-X capability, introduce a new tunable hw.re.intr_filter. If the tunable is set to non-zero value, driver will use interrupt taskqueue. The default value of the tunable is 0. This tunable has no effect on controllers that has no MSI/MSI-X capability or if MSI/MSI-X is explicitly disabled by administrator. While I'm here cleanup interrupt setup/teardown since re(4) uses single MSI/MSI-X message at this moment.
2011-01-26 20:25:40 +00:00
if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
rid = 0;
else
rid = 1;
if (sc->rl_irq[0] != NULL) {
bus_release_resource(dev, SYS_RES_IRQ, rid, sc->rl_irq[0]);
sc->rl_irq[0] = NULL;
Add MSI support. Ever since switching to adaptive polling re(4) occasionally spews watchdog timeouts on systems with MSI capability. This change is minimal one for supporting MSI and re(4) also needs MSIX support for RTL8111C in future. Because softc structure of re(4) is shared with rl(4), rl(4) was touched to use the modified softc. Reported by: cnst Tested by: cnst Approved by: re (kensmith)
2007-07-24 01:24:03 +00:00
}
Do not use interrupt taskqueue on controllers with MSI/MSI-X capability. One of reason using interrupt taskqueue in re(4) was to reduce number of TX/RX interrupts under load because re(4) controllers have no good TX/RX interrupt moderation mechanism. Basic TX interrupt moderation is done by hardware for most controllers but RX interrupt moderation through undocumented register showed poor RX performance so it was disabled in r215025. Using taskqueue to handle RX interrupt greatly reduced number of interrupts but re(4) consumed all available CPU cycles to run the taskqueue under high TX/RX network load. This can happen even with RTL810x fast ethernet controller and I believe this is not acceptable for most systems. To mitigate the issue, use one-shot timer register to moderate RX interrupts. The timer register provides programmable one-shot timer and can be used to suppress interrupt generation. The timer runs at 125MHZ on PCIe controllers so the minimum time allowed for the timer is 8ns. Data sheet says the register is 32 bits but experimentation shows only lower 13 bits are valid so maximum time that can be programmed is 65.528us. This yields theoretical maximum number of RX interrupts that could be generated per second is about 15260. Combined with TX completion interrupts re(4) shall generate less than 20k interrupts. This number is still slightly high compared to other intelligent ethernet controllers but system is very responsive even under high network load. Introduce sysctl variable dev.re.%d.int_rx_mod that controls amount of time to delay RX interrupt processing in units of us. Value 0 completely disables RX interrupt moderation. To provide old behavior for controllers that have MSI/MSI-X capability, introduce a new tunable hw.re.intr_filter. If the tunable is set to non-zero value, driver will use interrupt taskqueue. The default value of the tunable is 0. This tunable has no effect on controllers that has no MSI/MSI-X capability or if MSI/MSI-X is explicitly disabled by administrator. While I'm here cleanup interrupt setup/teardown since re(4) uses single MSI/MSI-X message at this moment.
2011-01-26 20:25:40 +00:00
if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0)
pci_release_msi(dev);
Prefer MSI-X to MSI on controllers that support MSI-X. All recent PCIe controllers(RTL8102E or later and RTL8168/8111C or later) supports either 2 or 4 MSI-X messages. Unfortunately vendor did not publicly release RSS related information yet. However switching to MSI-X is one-step forward to support RSS.
2011-01-25 22:18:00 +00:00
if (sc->rl_res_pba) {
rid = PCIR_BAR(4);
bus_release_resource(dev, SYS_RES_MEMORY, rid, sc->rl_res_pba);
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
if (sc->rl_res)
Switch to memory space register mapping over IO space. If that mapping fails fall back to traditional IO space access.
2008-07-02 05:21:09 +00:00
bus_release_resource(dev, sc->rl_res_type, sc->rl_res_id,
sc->rl_res);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/* Unload and free the RX DMA ring memory and map */
if (sc->rl_ldata.rl_rx_list_tag) {
Fix various NIC drivers to properly cleanup static DMA resources. In particular, don't check the value of the bus_dma map against NULL to determine if either bus_dmamem_alloc() or bus_dmamap_load() succeeded. Instead, assume that bus_dmamap_load() succeeeded (and thus that bus_dmamap_unload() should be called) if the bus address for a resource is non-zero, and assume that bus_dmamem_alloc() succeeded (and thus that bus_dmamem_free() should be called) if the virtual address for a resource is not NULL. In many cases these bugs could result in leaks when a driver was detached. Reviewed by: yongari MFC after: 2 weeks
2014-06-11 14:53:58 +00:00
if (sc->rl_ldata.rl_rx_list_addr)
Add simple MAC statistics counter reading support. Unfortunately useful counters like rl_missed_pkts is 16 bits quantity which is too small to hold meaningful information happened in a second. This means driver should frequently read these counters in order not to lose accuracy and that approach is too inefficient in driver's view. Moreover it seems there is no way to trigger an interrupt to detect counter near-full or wraparound event as well as lacking clearing the MAC counters. Another limitation of reading the counters from RealTek controllers is lack of interrupt firing at the end of DMA cycle of MAC counter read request such that driver have to poll the end of the DMA which is a time consuming process as well as inefficient. The more severe issue of the MAC counter read request is it takes too long to complete the DMA. All these limitation made maintaining MAC counters in driver impractical. For now, just provide simple sysctl interface to trigger reading the MAC counters. These counters could be used to track down driver issues. Users can read MAC counters maintained in controller with the following command. #sysctl dev.re.0.stats=1 While I'm here add check for validity of dma map and allocated memory before unloading/freeing them. Tested by: rmacklem
2010-11-05 19:28:00 +00:00
bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
sc->rl_ldata.rl_rx_list_map);
Fix various NIC drivers to properly cleanup static DMA resources. In particular, don't check the value of the bus_dma map against NULL to determine if either bus_dmamem_alloc() or bus_dmamap_load() succeeded. Instead, assume that bus_dmamap_load() succeeeded (and thus that bus_dmamap_unload() should be called) if the bus address for a resource is non-zero, and assume that bus_dmamem_alloc() succeeded (and thus that bus_dmamem_free() should be called) if the virtual address for a resource is not NULL. In many cases these bugs could result in leaks when a driver was detached. Reviewed by: yongari MFC after: 2 weeks
2014-06-11 14:53:58 +00:00
if (sc->rl_ldata.rl_rx_list)
Add simple MAC statistics counter reading support. Unfortunately useful counters like rl_missed_pkts is 16 bits quantity which is too small to hold meaningful information happened in a second. This means driver should frequently read these counters in order not to lose accuracy and that approach is too inefficient in driver's view. Moreover it seems there is no way to trigger an interrupt to detect counter near-full or wraparound event as well as lacking clearing the MAC counters. Another limitation of reading the counters from RealTek controllers is lack of interrupt firing at the end of DMA cycle of MAC counter read request such that driver have to poll the end of the DMA which is a time consuming process as well as inefficient. The more severe issue of the MAC counter read request is it takes too long to complete the DMA. All these limitation made maintaining MAC counters in driver impractical. For now, just provide simple sysctl interface to trigger reading the MAC counters. These counters could be used to track down driver issues. Users can read MAC counters maintained in controller with the following command. #sysctl dev.re.0.stats=1 While I'm here add check for validity of dma map and allocated memory before unloading/freeing them. Tested by: rmacklem
2010-11-05 19:28:00 +00:00
bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag,
sc->rl_ldata.rl_rx_list,
sc->rl_ldata.rl_rx_list_map);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag);
}
/* Unload and free the TX DMA ring memory and map */
if (sc->rl_ldata.rl_tx_list_tag) {
Fix various NIC drivers to properly cleanup static DMA resources. In particular, don't check the value of the bus_dma map against NULL to determine if either bus_dmamem_alloc() or bus_dmamap_load() succeeded. Instead, assume that bus_dmamap_load() succeeeded (and thus that bus_dmamap_unload() should be called) if the bus address for a resource is non-zero, and assume that bus_dmamem_alloc() succeeded (and thus that bus_dmamem_free() should be called) if the virtual address for a resource is not NULL. In many cases these bugs could result in leaks when a driver was detached. Reviewed by: yongari MFC after: 2 weeks
2014-06-11 14:53:58 +00:00
if (sc->rl_ldata.rl_tx_list_addr)
Add simple MAC statistics counter reading support. Unfortunately useful counters like rl_missed_pkts is 16 bits quantity which is too small to hold meaningful information happened in a second. This means driver should frequently read these counters in order not to lose accuracy and that approach is too inefficient in driver's view. Moreover it seems there is no way to trigger an interrupt to detect counter near-full or wraparound event as well as lacking clearing the MAC counters. Another limitation of reading the counters from RealTek controllers is lack of interrupt firing at the end of DMA cycle of MAC counter read request such that driver have to poll the end of the DMA which is a time consuming process as well as inefficient. The more severe issue of the MAC counter read request is it takes too long to complete the DMA. All these limitation made maintaining MAC counters in driver impractical. For now, just provide simple sysctl interface to trigger reading the MAC counters. These counters could be used to track down driver issues. Users can read MAC counters maintained in controller with the following command. #sysctl dev.re.0.stats=1 While I'm here add check for validity of dma map and allocated memory before unloading/freeing them. Tested by: rmacklem
2010-11-05 19:28:00 +00:00
bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
sc->rl_ldata.rl_tx_list_map);
Fix various NIC drivers to properly cleanup static DMA resources. In particular, don't check the value of the bus_dma map against NULL to determine if either bus_dmamem_alloc() or bus_dmamap_load() succeeded. Instead, assume that bus_dmamap_load() succeeeded (and thus that bus_dmamap_unload() should be called) if the bus address for a resource is non-zero, and assume that bus_dmamem_alloc() succeeded (and thus that bus_dmamem_free() should be called) if the virtual address for a resource is not NULL. In many cases these bugs could result in leaks when a driver was detached. Reviewed by: yongari MFC after: 2 weeks
2014-06-11 14:53:58 +00:00
if (sc->rl_ldata.rl_tx_list)
Add simple MAC statistics counter reading support. Unfortunately useful counters like rl_missed_pkts is 16 bits quantity which is too small to hold meaningful information happened in a second. This means driver should frequently read these counters in order not to lose accuracy and that approach is too inefficient in driver's view. Moreover it seems there is no way to trigger an interrupt to detect counter near-full or wraparound event as well as lacking clearing the MAC counters. Another limitation of reading the counters from RealTek controllers is lack of interrupt firing at the end of DMA cycle of MAC counter read request such that driver have to poll the end of the DMA which is a time consuming process as well as inefficient. The more severe issue of the MAC counter read request is it takes too long to complete the DMA. All these limitation made maintaining MAC counters in driver impractical. For now, just provide simple sysctl interface to trigger reading the MAC counters. These counters could be used to track down driver issues. Users can read MAC counters maintained in controller with the following command. #sysctl dev.re.0.stats=1 While I'm here add check for validity of dma map and allocated memory before unloading/freeing them. Tested by: rmacklem
2010-11-05 19:28:00 +00:00
bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag,
sc->rl_ldata.rl_tx_list,
sc->rl_ldata.rl_tx_list_map);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag);
}
/* Destroy all the RX and TX buffer maps */
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
if (sc->rl_ldata.rl_tx_mtag) {
Make sure to check validity of dma maps before destroying.
2011-01-13 23:00:28 +00:00
for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
if (sc->rl_ldata.rl_tx_desc[i].tx_dmamap)
bus_dmamap_destroy(sc->rl_ldata.rl_tx_mtag,
sc->rl_ldata.rl_tx_desc[i].tx_dmamap);
}
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
bus_dma_tag_destroy(sc->rl_ldata.rl_tx_mtag);
}
if (sc->rl_ldata.rl_rx_mtag) {
Make sure to check validity of dma maps before destroying.
2011-01-13 23:00:28 +00:00
for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
if (sc->rl_ldata.rl_rx_desc[i].rx_dmamap)
bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
sc->rl_ldata.rl_rx_desc[i].rx_dmamap);
}
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
if (sc->rl_ldata.rl_rx_sparemap)
bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag,
sc->rl_ldata.rl_rx_sparemap);
bus_dma_tag_destroy(sc->rl_ldata.rl_rx_mtag);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
if (sc->rl_ldata.rl_jrx_mtag) {
for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
if (sc->rl_ldata.rl_jrx_desc[i].rx_dmamap)
bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
sc->rl_ldata.rl_jrx_desc[i].rx_dmamap);
}
if (sc->rl_ldata.rl_jrx_sparemap)
bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag,
sc->rl_ldata.rl_jrx_sparemap);
bus_dma_tag_destroy(sc->rl_ldata.rl_jrx_mtag);
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/* Unload and free the stats buffer and map */
if (sc->rl_ldata.rl_stag) {
Fix various NIC drivers to properly cleanup static DMA resources. In particular, don't check the value of the bus_dma map against NULL to determine if either bus_dmamem_alloc() or bus_dmamap_load() succeeded. Instead, assume that bus_dmamap_load() succeeeded (and thus that bus_dmamap_unload() should be called) if the bus address for a resource is non-zero, and assume that bus_dmamem_alloc() succeeded (and thus that bus_dmamem_free() should be called) if the virtual address for a resource is not NULL. In many cases these bugs could result in leaks when a driver was detached. Reviewed by: yongari MFC after: 2 weeks
2014-06-11 14:53:58 +00:00
if (sc->rl_ldata.rl_stats_addr)
Add simple MAC statistics counter reading support. Unfortunately useful counters like rl_missed_pkts is 16 bits quantity which is too small to hold meaningful information happened in a second. This means driver should frequently read these counters in order not to lose accuracy and that approach is too inefficient in driver's view. Moreover it seems there is no way to trigger an interrupt to detect counter near-full or wraparound event as well as lacking clearing the MAC counters. Another limitation of reading the counters from RealTek controllers is lack of interrupt firing at the end of DMA cycle of MAC counter read request such that driver have to poll the end of the DMA which is a time consuming process as well as inefficient. The more severe issue of the MAC counter read request is it takes too long to complete the DMA. All these limitation made maintaining MAC counters in driver impractical. For now, just provide simple sysctl interface to trigger reading the MAC counters. These counters could be used to track down driver issues. Users can read MAC counters maintained in controller with the following command. #sysctl dev.re.0.stats=1 While I'm here add check for validity of dma map and allocated memory before unloading/freeing them. Tested by: rmacklem
2010-11-05 19:28:00 +00:00
bus_dmamap_unload(sc->rl_ldata.rl_stag,
sc->rl_ldata.rl_smap);
Fix various NIC drivers to properly cleanup static DMA resources. In particular, don't check the value of the bus_dma map against NULL to determine if either bus_dmamem_alloc() or bus_dmamap_load() succeeded. Instead, assume that bus_dmamap_load() succeeeded (and thus that bus_dmamap_unload() should be called) if the bus address for a resource is non-zero, and assume that bus_dmamem_alloc() succeeded (and thus that bus_dmamem_free() should be called) if the virtual address for a resource is not NULL. In many cases these bugs could result in leaks when a driver was detached. Reviewed by: yongari MFC after: 2 weeks
2014-06-11 14:53:58 +00:00
if (sc->rl_ldata.rl_stats)
Add simple MAC statistics counter reading support. Unfortunately useful counters like rl_missed_pkts is 16 bits quantity which is too small to hold meaningful information happened in a second. This means driver should frequently read these counters in order not to lose accuracy and that approach is too inefficient in driver's view. Moreover it seems there is no way to trigger an interrupt to detect counter near-full or wraparound event as well as lacking clearing the MAC counters. Another limitation of reading the counters from RealTek controllers is lack of interrupt firing at the end of DMA cycle of MAC counter read request such that driver have to poll the end of the DMA which is a time consuming process as well as inefficient. The more severe issue of the MAC counter read request is it takes too long to complete the DMA. All these limitation made maintaining MAC counters in driver impractical. For now, just provide simple sysctl interface to trigger reading the MAC counters. These counters could be used to track down driver issues. Users can read MAC counters maintained in controller with the following command. #sysctl dev.re.0.stats=1 While I'm here add check for validity of dma map and allocated memory before unloading/freeing them. Tested by: rmacklem
2010-11-05 19:28:00 +00:00
bus_dmamem_free(sc->rl_ldata.rl_stag,
sc->rl_ldata.rl_stats, sc->rl_ldata.rl_smap);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
bus_dma_tag_destroy(sc->rl_ldata.rl_stag);
}
if (sc->rl_parent_tag)
bus_dma_tag_destroy(sc->rl_parent_tag);
mtx_destroy(&sc->rl_mtx);
Whitespace pass
2004-07-06 02:46:53 +00:00
return (0);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
static __inline void
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_discard_rxbuf(struct rl_softc *sc, int idx)
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
{
struct rl_desc *desc;
struct rl_rxdesc *rxd;
uint32_t cmdstat;
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
if (sc->rl_ifp->if_mtu > RL_MTU &&
(sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
rxd = &sc->rl_ldata.rl_jrx_desc[idx];
else
rxd = &sc->rl_ldata.rl_rx_desc[idx];
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
desc = &sc->rl_ldata.rl_rx_list[idx];
desc->rl_vlanctl = 0;
cmdstat = rxd->rx_size;
if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
cmdstat |= RL_RDESC_CMD_EOR;
desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
static int
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_newbuf(struct rl_softc *sc, int idx)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
struct mbuf *m;
struct rl_rxdesc *rxd;
bus_dma_segment_t segs[1];
bus_dmamap_t map;
struct rl_desc *desc;
uint32_t cmdstat;
int error, nsegs;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Mechanically substitute flags from historic mbuf allocator with malloc(9) flags in sys/dev.
2012-12-04 09:32:43 +00:00
m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
if (m == NULL)
return (ENOBUFS);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
fix jumbo frames as much as they can be fixed for re. We now cap the MTU to 7422 since it appears that the 8169S can't transmit anything larger.. The 8169S can receive full jumbo frames, but we don't have an mru to let the upper layers know this... add fixup so that this driver should work on alignment constrained platforms (!i386 && !amd64) MFC after: 5 days
2004-09-28 18:22:24 +00:00
m->m_len = m->m_pkthdr.len = MCLBYTES;
#ifdef RE_FIXUP_RX
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/*
fix jumbo frames as much as they can be fixed for re. We now cap the MTU to 7422 since it appears that the 8169S can't transmit anything larger.. The 8169S can receive full jumbo frames, but we don't have an mru to let the upper layers know this... add fixup so that this driver should work on alignment constrained platforms (!i386 && !amd64) MFC after: 5 days
2004-09-28 18:22:24 +00:00
* This is part of an evil trick to deal with non-x86 platforms.
* The RealTek chip requires RX buffers to be aligned on 64-bit
* boundaries, but that will hose non-x86 machines. To get around
* this, we leave some empty space at the start of each buffer
* and for non-x86 hosts, we copy the buffer back six bytes
* to achieve word alignment. This is slightly more efficient
* than allocating a new buffer, copying the contents, and
* discarding the old buffer.
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
*/
fix jumbo frames as much as they can be fixed for re. We now cap the MTU to 7422 since it appears that the 8169S can't transmit anything larger.. The 8169S can receive full jumbo frames, but we don't have an mru to let the upper layers know this... add fixup so that this driver should work on alignment constrained platforms (!i386 && !amd64) MFC after: 5 days
2004-09-28 18:22:24 +00:00
m_adj(m, RE_ETHER_ALIGN);
#endif
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_rx_mtag,
sc->rl_ldata.rl_rx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
if (error != 0) {
m_freem(m);
return (ENOBUFS);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
rxd = &sc->rl_ldata.rl_rx_desc[idx];
if (rxd->rx_m != NULL) {
bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
BUS_DMASYNC_POSTREAD);
bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap);
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
rxd->rx_m = m;
map = rxd->rx_dmamap;
rxd->rx_dmamap = sc->rl_ldata.rl_rx_sparemap;
rxd->rx_size = segs[0].ds_len;
sc->rl_ldata.rl_rx_sparemap = map;
bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap,
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
BUS_DMASYNC_PREREAD);
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
desc = &sc->rl_ldata.rl_rx_list[idx];
desc->rl_vlanctl = 0;
desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
cmdstat = segs[0].ds_len;
if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
cmdstat |= RL_RDESC_CMD_EOR;
desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
Whitespace pass
2004-07-06 02:46:53 +00:00
return (0);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
static int
re_jumbo_newbuf(struct rl_softc *sc, int idx)
{
struct mbuf *m;
struct rl_rxdesc *rxd;
bus_dma_segment_t segs[1];
bus_dmamap_t map;
struct rl_desc *desc;
uint32_t cmdstat;
int error, nsegs;
Mechanically substitute flags from historic mbuf allocator with malloc(9) flags in sys/dev.
2012-12-04 09:32:43 +00:00
m = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, MJUM9BYTES);
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
if (m == NULL)
return (ENOBUFS);
m->m_len = m->m_pkthdr.len = MJUM9BYTES;
#ifdef RE_FIXUP_RX
m_adj(m, RE_ETHER_ALIGN);
#endif
error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_jrx_mtag,
sc->rl_ldata.rl_jrx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
if (error != 0) {
m_freem(m);
return (ENOBUFS);
}
KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs));
rxd = &sc->rl_ldata.rl_jrx_desc[idx];
if (rxd->rx_m != NULL) {
bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
BUS_DMASYNC_POSTREAD);
bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap);
}
rxd->rx_m = m;
map = rxd->rx_dmamap;
rxd->rx_dmamap = sc->rl_ldata.rl_jrx_sparemap;
rxd->rx_size = segs[0].ds_len;
sc->rl_ldata.rl_jrx_sparemap = map;
bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap,
BUS_DMASYNC_PREREAD);
desc = &sc->rl_ldata.rl_rx_list[idx];
desc->rl_vlanctl = 0;
desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr));
desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr));
cmdstat = segs[0].ds_len;
if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1)
cmdstat |= RL_RDESC_CMD_EOR;
desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN);
return (0);
}
fix jumbo frames as much as they can be fixed for re. We now cap the MTU to 7422 since it appears that the 8169S can't transmit anything larger.. The 8169S can receive full jumbo frames, but we don't have an mru to let the upper layers know this... add fixup so that this driver should work on alignment constrained platforms (!i386 && !amd64) MFC after: 5 days
2004-09-28 18:22:24 +00:00
#ifdef RE_FIXUP_RX
static __inline void
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_fixup_rx(struct mbuf *m)
fix jumbo frames as much as they can be fixed for re. We now cap the MTU to 7422 since it appears that the 8169S can't transmit anything larger.. The 8169S can receive full jumbo frames, but we don't have an mru to let the upper layers know this... add fixup so that this driver should work on alignment constrained platforms (!i386 && !amd64) MFC after: 5 days
2004-09-28 18:22:24 +00:00
{
int i;
uint16_t *src, *dst;
src = mtod(m, uint16_t *);
dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src;
for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
*dst++ = *src++;
m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN;
}
#endif
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
static int
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_tx_list_init(struct rl_softc *sc)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
struct rl_desc *desc;
int i;
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
RL_LOCK_ASSERT(sc);
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
bzero(sc->rl_ldata.rl_tx_list,
sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc));
for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++)
sc->rl_ldata.rl_tx_desc[i].tx_m = NULL;
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#ifdef DEV_NETMAP
re_netmap_tx_init(sc);
#endif /* DEV_NETMAP */
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
/* Set EOR. */
desc = &sc->rl_ldata.rl_tx_list[sc->rl_ldata.rl_tx_desc_cnt - 1];
desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOR);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
sc->rl_ldata.rl_tx_list_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
sc->rl_ldata.rl_tx_prodidx = 0;
sc->rl_ldata.rl_tx_considx = 0;
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
sc->rl_ldata.rl_tx_free = sc->rl_ldata.rl_tx_desc_cnt;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Whitespace pass
2004-07-06 02:46:53 +00:00
return (0);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
static int
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_rx_list_init(struct rl_softc *sc)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
int error, i;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
bzero(sc->rl_ldata.rl_rx_list,
sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
sc->rl_ldata.rl_rx_desc[i].rx_m = NULL;
if ((error = re_newbuf(sc, i)) != 0)
return (error);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#ifdef DEV_NETMAP
re_netmap_rx_init(sc);
#endif /* DEV_NETMAP */
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/* Flush the RX descriptors */
bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
sc->rl_ldata.rl_rx_list_map,
BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
sc->rl_ldata.rl_rx_prodidx = 0;
sc->rl_head = sc->rl_tail = NULL;
Do not use interrupt taskqueue on controllers with MSI/MSI-X capability. One of reason using interrupt taskqueue in re(4) was to reduce number of TX/RX interrupts under load because re(4) controllers have no good TX/RX interrupt moderation mechanism. Basic TX interrupt moderation is done by hardware for most controllers but RX interrupt moderation through undocumented register showed poor RX performance so it was disabled in r215025. Using taskqueue to handle RX interrupt greatly reduced number of interrupts but re(4) consumed all available CPU cycles to run the taskqueue under high TX/RX network load. This can happen even with RTL810x fast ethernet controller and I believe this is not acceptable for most systems. To mitigate the issue, use one-shot timer register to moderate RX interrupts. The timer register provides programmable one-shot timer and can be used to suppress interrupt generation. The timer runs at 125MHZ on PCIe controllers so the minimum time allowed for the timer is 8ns. Data sheet says the register is 32 bits but experimentation shows only lower 13 bits are valid so maximum time that can be programmed is 65.528us. This yields theoretical maximum number of RX interrupts that could be generated per second is about 15260. Combined with TX completion interrupts re(4) shall generate less than 20k interrupts. This number is still slightly high compared to other intelligent ethernet controllers but system is very responsive even under high network load. Introduce sysctl variable dev.re.%d.int_rx_mod that controls amount of time to delay RX interrupt processing in units of us. Value 0 completely disables RX interrupt moderation. To provide old behavior for controllers that have MSI/MSI-X capability, introduce a new tunable hw.re.intr_filter. If the tunable is set to non-zero value, driver will use interrupt taskqueue. The default value of the tunable is 0. This tunable has no effect on controllers that has no MSI/MSI-X capability or if MSI/MSI-X is explicitly disabled by administrator. While I'm here cleanup interrupt setup/teardown since re(4) uses single MSI/MSI-X message at this moment.
2011-01-26 20:25:40 +00:00
sc->rl_int_rx_act = 0;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Whitespace pass
2004-07-06 02:46:53 +00:00
return (0);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
static int
re_jrx_list_init(struct rl_softc *sc)
{
int error, i;
bzero(sc->rl_ldata.rl_rx_list,
sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc));
for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
sc->rl_ldata.rl_jrx_desc[i].rx_m = NULL;
if ((error = re_jumbo_newbuf(sc, i)) != 0)
return (error);
}
bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
sc->rl_ldata.rl_rx_list_map,
BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD);
sc->rl_ldata.rl_rx_prodidx = 0;
sc->rl_head = sc->rl_tail = NULL;
Do not use interrupt taskqueue on controllers with MSI/MSI-X capability. One of reason using interrupt taskqueue in re(4) was to reduce number of TX/RX interrupts under load because re(4) controllers have no good TX/RX interrupt moderation mechanism. Basic TX interrupt moderation is done by hardware for most controllers but RX interrupt moderation through undocumented register showed poor RX performance so it was disabled in r215025. Using taskqueue to handle RX interrupt greatly reduced number of interrupts but re(4) consumed all available CPU cycles to run the taskqueue under high TX/RX network load. This can happen even with RTL810x fast ethernet controller and I believe this is not acceptable for most systems. To mitigate the issue, use one-shot timer register to moderate RX interrupts. The timer register provides programmable one-shot timer and can be used to suppress interrupt generation. The timer runs at 125MHZ on PCIe controllers so the minimum time allowed for the timer is 8ns. Data sheet says the register is 32 bits but experimentation shows only lower 13 bits are valid so maximum time that can be programmed is 65.528us. This yields theoretical maximum number of RX interrupts that could be generated per second is about 15260. Combined with TX completion interrupts re(4) shall generate less than 20k interrupts. This number is still slightly high compared to other intelligent ethernet controllers but system is very responsive even under high network load. Introduce sysctl variable dev.re.%d.int_rx_mod that controls amount of time to delay RX interrupt processing in units of us. Value 0 completely disables RX interrupt moderation. To provide old behavior for controllers that have MSI/MSI-X capability, introduce a new tunable hw.re.intr_filter. If the tunable is set to non-zero value, driver will use interrupt taskqueue. The default value of the tunable is 0. This tunable has no effect on controllers that has no MSI/MSI-X capability or if MSI/MSI-X is explicitly disabled by administrator. While I'm here cleanup interrupt setup/teardown since re(4) uses single MSI/MSI-X message at this moment.
2011-01-26 20:25:40 +00:00
sc->rl_int_rx_act = 0;
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
return (0);
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/*
* RX handler for C+ and 8169. For the gigE chips, we support
* the reception of jumbo frames that have been fragmented
* across multiple 2K mbuf cluster buffers.
*/
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
static int
When user_frac in the polling subsystem is low it is going to busy the CPU for too long period than necessary. Additively, interfaces are kept polled (in the tick) even if no more packets are available. In order to avoid such situations a new generic mechanism can be implemented in proactive way, keeping track of the time spent on any packet and fragmenting the time for any tick, stopping the processing as soon as possible. In order to implement such mechanism, the polling handler needs to change, returning the number of packets processed. While the intended logic is not part of this patch, the polling KPI is broken by this commit, adding an int return value and the new flag IFCAP_POLLING_NOCOUNT (which will signal that the return value is meaningless for the installed handler and checking should be skipped). Bump __FreeBSD_version in order to signal such situation. Reviewed by: emaste Sponsored by: Sandvine Incorporated
2009-05-30 15:14:44 +00:00
re_rxeof(struct rl_softc *sc, int *rx_npktsp)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
struct mbuf *m;
struct ifnet *ifp;
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
int i, rxerr, total_len;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
struct rl_desc *cur_rx;
u_int32_t rxstat, rxvlan;
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
int jumbo, maxpkt = 16, rx_npkts = 0;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Drop the driver lock around calls to if_input to avoid a LOR when the packets are immediately returned for sending (e.g. when bridging or packet forwarding). There are more efficient ways to do this but for now use the least intrusive approach. Reviewed by: imp, rwatson
2003-11-14 19:00:32 +00:00
RL_LOCK_ASSERT(sc);
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
ifp = sc->rl_ifp;
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#ifdef DEV_NETMAP
update to the latest netmap snapshot. This includes the following: - use separate memory regions for VALE ports - locking fixes - some simplifications in the NIC-specific routines - performance improvements for the VALE switch - some new features in the pkt-gen test program - documentation updates There are small API changes that require programs to be recompiled (NETMAP_API has been bumped so you will detect old binaries at runtime). In particular: - struct netmap_slot now is 16 bytes to support an extra pointer, which may save one data copy when using VALE ports or VMs; - the struct netmap_if has two extra fields; MFC after: 3 days
2013-11-01 21:21:14 +00:00
if (netmap_rx_irq(ifp, 0, &rx_npkts))
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
return 0;
#endif /* DEV_NETMAP */
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
if (ifp->if_mtu > RL_MTU && (sc->rl_flags & RL_FLAG_JUMBOV2) != 0)
jumbo = 1;
else
jumbo = 0;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/* Invalidate the descriptor memory */
bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
sc->rl_ldata.rl_rx_list_map,
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
for (i = sc->rl_ldata.rl_rx_prodidx; maxpkt > 0;
i = RL_RX_DESC_NXT(sc, i)) {
Add a check to know whether driver is still running after reacquiring driver lock in Rx handler. re(4) drops a driver lock before passing received frame to upper stack and reacquire the lock. During the time window ioctl calls could be executed and if the ioctl was interface down request, driver will stop the controller and free allocated mbufs. After that when driver comes back to Rx handler again it does not know what was happend so it could access free mbufs which in turn cause panic. Reported by: Norbert Papke < npapk <> acm dot org > Tested by: Norbert Papke < npapk <> acm dot org >
2009-11-02 18:35:05 +00:00
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
break;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
cur_rx = &sc->rl_ldata.rl_rx_list[i];
rxstat = le32toh(cur_rx->rl_cmdstat);
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
if ((rxstat & RL_RDESC_STAT_OWN) != 0)
break;
total_len = rxstat & sc->rl_rxlenmask;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
rxvlan = le32toh(cur_rx->rl_vlanctl);
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
if (jumbo != 0)
m = sc->rl_ldata.rl_jrx_desc[i].rx_m;
else
m = sc->rl_ldata.rl_rx_desc[i].rx_m;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
(rxstat & (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) !=
(RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) {
/*
* RTL8168C or later controllers do not
* support multi-fragment packet.
*/
re_discard_rxbuf(sc, i);
continue;
} else if ((rxstat & RL_RDESC_STAT_EOF) == 0) {
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
if (re_newbuf(sc, i) != 0) {
/*
* If this is part of a multi-fragment packet,
* discard all the pieces.
*/
if (sc->rl_head != NULL) {
m_freem(sc->rl_head);
sc->rl_head = sc->rl_tail = NULL;
}
re_discard_rxbuf(sc, i);
continue;
}
fix jumbo frames as much as they can be fixed for re. We now cap the MTU to 7422 since it appears that the 8169S can't transmit anything larger.. The 8169S can receive full jumbo frames, but we don't have an mru to let the upper layers know this... add fixup so that this driver should work on alignment constrained platforms (!i386 && !amd64) MFC after: 5 days
2004-09-28 18:22:24 +00:00
m->m_len = RE_RX_DESC_BUFLEN;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
if (sc->rl_head == NULL)
sc->rl_head = sc->rl_tail = m;
else {
m->m_flags &= ~M_PKTHDR;
sc->rl_tail->m_next = m;
sc->rl_tail = m;
}
continue;
}
/*
* NOTE: for the 8139C+, the frame length field
* is always 12 bits in size, but for the gigE chips,
* it is 13 bits (since the max RX frame length is 16K).
* Unfortunately, all 32 bits in the status word
* were already used, so to make room for the extra
* length bit, RealTek took out the 'frame alignment
* error' bit and shifted the other status bits
* over one slot. The OWN, EOR, FS and LS bits are
* still in the same places. We have already extracted
* the frame length and checked the OWN bit, so rather
* than using an alternate bit mapping, we shift the
* status bits one space to the right so we can evaluate
* them using the 8169 status as though it was in the
* same format as that of the 8139C+.
*/
if (sc->rl_type == RL_8169)
rxstat >>= 1;
fix jumbo frames as much as they can be fixed for re. We now cap the MTU to 7422 since it appears that the 8169S can't transmit anything larger.. The 8169S can receive full jumbo frames, but we don't have an mru to let the upper layers know this... add fixup so that this driver should work on alignment constrained platforms (!i386 && !amd64) MFC after: 5 days
2004-09-28 18:22:24 +00:00
/*
* if total_len > 2^13-1, both _RXERRSUM and _GIANT will be
* set, but if CRC is clear, it will still be a valid frame.
*/
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
if ((rxstat & RL_RDESC_STAT_RXERRSUM) != 0) {
rxerr = 1;
if ((sc->rl_flags & RL_FLAG_JUMBOV2) == 0 &&
total_len > 8191 &&
(rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT)
rxerr = 0;
if (rxerr != 0) {
Mechanically convert to if_inc_counter().
2014-09-19 03:51:26 +00:00
if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
/*
* If this is part of a multi-fragment packet,
* discard all the pieces.
*/
if (sc->rl_head != NULL) {
m_freem(sc->rl_head);
sc->rl_head = sc->rl_tail = NULL;
}
re_discard_rxbuf(sc, i);
continue;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
}
/*
* If allocating a replacement mbuf fails,
* reload the current one.
*/
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
if (jumbo != 0)
rxerr = re_jumbo_newbuf(sc, i);
else
rxerr = re_newbuf(sc, i);
if (rxerr != 0) {
Mechanically convert to if_inc_counter().
2014-09-19 03:51:26 +00:00
if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
if (sc->rl_head != NULL) {
m_freem(sc->rl_head);
sc->rl_head = sc->rl_tail = NULL;
}
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
re_discard_rxbuf(sc, i);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
continue;
}
if (sc->rl_head != NULL) {
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
if (jumbo != 0)
m->m_len = total_len;
else {
m->m_len = total_len % RE_RX_DESC_BUFLEN;
if (m->m_len == 0)
m->m_len = RE_RX_DESC_BUFLEN;
}
Whitespace pass
2004-07-06 02:46:53 +00:00
/*
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
* Special case: if there's 4 bytes or less
* in this buffer, the mbuf can be discarded:
* the last 4 bytes is the CRC, which we don't
* care about anyway.
*/
if (m->m_len <= ETHER_CRC_LEN) {
sc->rl_tail->m_len -=
(ETHER_CRC_LEN - m->m_len);
m_freem(m);
} else {
m->m_len -= ETHER_CRC_LEN;
m->m_flags &= ~M_PKTHDR;
sc->rl_tail->m_next = m;
}
m = sc->rl_head;
sc->rl_head = sc->rl_tail = NULL;
m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
} else
m->m_pkthdr.len = m->m_len =
(total_len - ETHER_CRC_LEN);
fix jumbo frames as much as they can be fixed for re. We now cap the MTU to 7422 since it appears that the 8169S can't transmit anything larger.. The 8169S can receive full jumbo frames, but we don't have an mru to let the upper layers know this... add fixup so that this driver should work on alignment constrained platforms (!i386 && !amd64) MFC after: 5 days
2004-09-28 18:22:24 +00:00
#ifdef RE_FIXUP_RX
re_fixup_rx(m);
#endif
Mechanically convert to if_inc_counter().
2014-09-19 03:51:26 +00:00
if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
m->m_pkthdr.rcvif = ifp;
/* Do RX checksumming if enabled */
if (ifp->if_capenable & IFCAP_RXCSUM) {
Add basic support for RTL8168C, RTL8168CP, RTL8111C and RTL8111CP. ATM Tx/Rx checksum offload is supported but TSO and jumbo frame is not yet supported. Because these newer controllers use different descriptor formats, a flag RL_FLAG_DESCV2 flag was introduced to handle that case in Tx/Rx handler. Also newer controllers seems to require to not touch 'enable Tx/Rx bit' in RL_CPLUS_CMD register so don't blindly try to set that bits. Note, it seems that there is still power-saving related issue where driver fails to attach PHY. Rebooting seems to fix that issue but number of required reboots varys. Many thanks to users that helped during developement. I really appreciate their patient and test/feedbacks.
2008-07-02 07:54:53 +00:00
if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
/* Check IP header checksum */
if (rxstat & RL_RDESC_STAT_PROTOID)
m->m_pkthdr.csum_flags |=
CSUM_IP_CHECKED;
if (!(rxstat & RL_RDESC_STAT_IPSUMBAD))
m->m_pkthdr.csum_flags |=
CSUM_IP_VALID;
/* Check TCP/UDP checksum */
if ((RL_TCPPKT(rxstat) &&
!(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
(RL_UDPPKT(rxstat) &&
!(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
m->m_pkthdr.csum_flags |=
CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
m->m_pkthdr.csum_data = 0xffff;
}
} else {
/*
* RTL8168C/RTL816CP/RTL8111C/RTL8111CP
*/
if ((rxstat & RL_RDESC_STAT_PROTOID) &&
(rxvlan & RL_RDESC_IPV4))
m->m_pkthdr.csum_flags |=
CSUM_IP_CHECKED;
if (!(rxstat & RL_RDESC_STAT_IPSUMBAD) &&
(rxvlan & RL_RDESC_IPV4))
m->m_pkthdr.csum_flags |=
CSUM_IP_VALID;
if (((rxstat & RL_RDESC_STAT_TCP) &&
!(rxstat & RL_RDESC_STAT_TCPSUMBAD)) ||
((rxstat & RL_RDESC_STAT_UDP) &&
!(rxstat & RL_RDESC_STAT_UDPSUMBAD))) {
m->m_pkthdr.csum_flags |=
CSUM_DATA_VALID|CSUM_PSEUDO_HDR;
m->m_pkthdr.csum_data = 0xffff;
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
}
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
maxpkt--;
- Fix VLAN_INPUT_TAG() macro, so that it doesn't touch mtag in case if memory allocation failed. - Remove fourth argument from VLAN_INPUT_TAG(), that was used incorrectly in almost all drivers. Indicate failure with mbuf value of NULL. In collaboration with: yongari, ru, sam
2005-12-18 18:24:27 +00:00
if (rxvlan & RL_RDESC_VLANCTL_TAG) {
Move ethernet VLAN tags from mtags to its own mbuf packet header field m_pkthdr.ether_vlan. The presence of the M_VLANTAG flag on the mbuf signifies the presence and validity of its content. Drivers that support hardware VLAN tag stripping fill in the received VLAN tag (containing both vlan and priority information) into the ether_vtag mbuf packet header field: m->m_pkthdr.ether_vtag = vlan_id; /* ntohs()? */ m->m_flags |= M_VLANTAG; to mark the packet m with the specified VLAN tag. On output the driver should check the mbuf for the M_VLANTAG flag to see if a VLAN tag is present and valid: if (m->m_flags & M_VLANTAG) { ... = m->m_pkthdr.ether_vtag; /* htons()? */ ... pass tag to hardware ... } VLAN tags are stored in host byte order. Byte swapping may be necessary. (Note: This driver conversion was mechanic and did not add or remove any byte swapping in the drivers.) Remove zone_mtag_vlan UMA zone and MTAG_VLAN definition. No more tag memory allocation have to be done. Reviewed by: thompsa, yar Sponsored by: TCP/IP Optimization Fundraise 2005
2006-09-17 13:33:30 +00:00
m->m_pkthdr.ether_vtag =
Fix VLAN hardware tag insertion/stripping on big-endian architectures. Reported by: naddy Tested on: sparc64 MFC after: 1 week
2008-08-13 03:40:08 +00:00
bswap16((rxvlan & RL_RDESC_VLANCTL_DATA));
Move ethernet VLAN tags from mtags to its own mbuf packet header field m_pkthdr.ether_vlan. The presence of the M_VLANTAG flag on the mbuf signifies the presence and validity of its content. Drivers that support hardware VLAN tag stripping fill in the received VLAN tag (containing both vlan and priority information) into the ether_vtag mbuf packet header field: m->m_pkthdr.ether_vtag = vlan_id; /* ntohs()? */ m->m_flags |= M_VLANTAG; to mark the packet m with the specified VLAN tag. On output the driver should check the mbuf for the M_VLANTAG flag to see if a VLAN tag is present and valid: if (m->m_flags & M_VLANTAG) { ... = m->m_pkthdr.ether_vtag; /* htons()? */ ... pass tag to hardware ... } VLAN tags are stored in host byte order. Byte swapping may be necessary. (Note: This driver conversion was mechanic and did not add or remove any byte swapping in the drivers.) Remove zone_mtag_vlan UMA zone and MTAG_VLAN definition. No more tag memory allocation have to be done. Reviewed by: thompsa, yar Sponsored by: TCP/IP Optimization Fundraise 2005
2006-09-17 13:33:30 +00:00
m->m_flags |= M_VLANTAG;
- Fix VLAN_INPUT_TAG() macro, so that it doesn't touch mtag in case if memory allocation failed. - Remove fourth argument from VLAN_INPUT_TAG(), that was used incorrectly in almost all drivers. Indicate failure with mbuf value of NULL. In collaboration with: yongari, ru, sam
2005-12-18 18:24:27 +00:00
}
Drop the driver lock around calls to if_input to avoid a LOR when the packets are immediately returned for sending (e.g. when bridging or packet forwarding). There are more efficient ways to do this but for now use the least intrusive approach. Reviewed by: imp, rwatson
2003-11-14 19:00:32 +00:00
RL_UNLOCK(sc);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
(*ifp->if_input)(ifp, m);
Drop the driver lock around calls to if_input to avoid a LOR when the packets are immediately returned for sending (e.g. when bridging or packet forwarding). There are more efficient ways to do this but for now use the least intrusive approach. Reviewed by: imp, rwatson
2003-11-14 19:00:32 +00:00
RL_LOCK(sc);
When user_frac in the polling subsystem is low it is going to busy the CPU for too long period than necessary. Additively, interfaces are kept polled (in the tick) even if no more packets are available. In order to avoid such situations a new generic mechanism can be implemented in proactive way, keeping track of the time spent on any packet and fragmenting the time for any tick, stopping the processing as soon as possible. In order to implement such mechanism, the polling handler needs to change, returning the number of packets processed. While the intended logic is not part of this patch, the polling KPI is broken by this commit, adding an int return value and the new flag IFCAP_POLLING_NOCOUNT (which will signal that the return value is meaningless for the installed handler and checking should be skipped). Bump __FreeBSD_version in order to signal such situation. Reviewed by: emaste Sponsored by: Sandvine Incorporated
2009-05-30 15:14:44 +00:00
rx_npkts++;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
/* Flush the RX DMA ring */
bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag,
sc->rl_ldata.rl_rx_list_map,
BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
sc->rl_ldata.rl_rx_prodidx = i;
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
When user_frac in the polling subsystem is low it is going to busy the CPU for too long period than necessary. Additively, interfaces are kept polled (in the tick) even if no more packets are available. In order to avoid such situations a new generic mechanism can be implemented in proactive way, keeping track of the time spent on any packet and fragmenting the time for any tick, stopping the processing as soon as possible. In order to implement such mechanism, the polling handler needs to change, returning the number of packets processed. While the intended logic is not part of this patch, the polling KPI is broken by this commit, adding an int return value and the new flag IFCAP_POLLING_NOCOUNT (which will signal that the return value is meaningless for the installed handler and checking should be skipped). Bump __FreeBSD_version in order to signal such situation. Reviewed by: emaste Sponsored by: Sandvine Incorporated
2009-05-30 15:14:44 +00:00
if (rx_npktsp != NULL)
*rx_npktsp = rx_npkts;
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
if (maxpkt)
style(9).
2010-11-05 18:24:50 +00:00
return (EAGAIN);
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
style(9).
2010-11-05 18:24:50 +00:00
return (0);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
static void
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_txeof(struct rl_softc *sc)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
struct ifnet *ifp;
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
struct rl_txdesc *txd;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
u_int32_t txstat;
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
int cons;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
cons = sc->rl_ldata.rl_tx_considx;
if (cons == sc->rl_ldata.rl_tx_prodidx)
return;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
ifp = sc->rl_ifp;
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#ifdef DEV_NETMAP
update to the latest netmap snapshot. This includes the following: - use separate memory regions for VALE ports - locking fixes - some simplifications in the NIC-specific routines - performance improvements for the VALE switch - some new features in the pkt-gen test program - documentation updates There are small API changes that require programs to be recompiled (NETMAP_API has been bumped so you will detect old binaries at runtime). In particular: - struct netmap_slot now is 16 bytes to support an extra pointer, which may save one data copy when using VALE ports or VMs; - the struct netmap_if has two extra fields; MFC after: 3 days
2013-11-01 21:21:14 +00:00
if (netmap_tx_irq(ifp, 0))
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
return;
#endif /* DEV_NETMAP */
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/* Invalidate the TX descriptor list */
bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
sc->rl_ldata.rl_tx_list_map,
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
for (; cons != sc->rl_ldata.rl_tx_prodidx;
cons = RL_TX_DESC_NXT(sc, cons)) {
txstat = le32toh(sc->rl_ldata.rl_tx_list[cons].rl_cmdstat);
if (txstat & RL_TDESC_STAT_OWN)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
break;
/*
* We only stash mbufs in the last descriptor
* in a fragment chain, which also happens to
* be the only place where the TX status bits
* are valid.
*/
if (txstat & RL_TDESC_CMD_EOF) {
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
txd = &sc->rl_ldata.rl_tx_desc[cons];
bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
txd->tx_dmamap);
KASSERT(txd->tx_m != NULL,
("%s: freeing NULL mbufs!", __func__));
m_freem(txd->tx_m);
txd->tx_m = NULL;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
if (txstat & (RL_TDESC_STAT_EXCESSCOL|
RL_TDESC_STAT_COLCNT))
Mechanically convert to if_inc_counter().
2014-09-19 03:51:26 +00:00
if_inc_counter(ifp, IFCOUNTER_COLLISIONS, 1);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
if (txstat & RL_TDESC_STAT_TXERRSUM)
Mechanically convert to if_inc_counter().
2014-09-19 03:51:26 +00:00
if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
else
Mechanically convert to if_inc_counter().
2014-09-19 03:51:26 +00:00
if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
sc->rl_ldata.rl_tx_free++;
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
sc->rl_ldata.rl_tx_considx = cons;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/* No changes made to the TX ring, so no flush needed */
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt) {
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
#ifdef RE_TX_MODERATION
o In re_newbuf() and re_encap() if re_dma_map_desc() aborts the mapping operation as it ran out of free descriptors or if there are too many segments in the first place, call bus_dmamap_unload() in order to unload the already loaded segments. For trying to map the defragmented mbuf (chain) in re_encap() this introduces re_dma_map_desc() setting arg.rl_maxsegs to 0 as a new failure mode. Previously we just ignored this case, corrupting our view of the TX ring. o In re_txeof(): - Don't clear IFF_DRV_OACTIVE unless there are at least 4 free TX descriptors. Further down the road re_encap() will bail if there aren't at least 4 free TX descriptors, causing re_start() to abort and prepend the dequeued mbuf again so it makes no sense to pretend we could process mbufs again when in fact we won't. While at it replace this magic 4 with a macro RL_TX_DESC_THLD throughout this driver. - Don't cancel the watchdog timeout as soon as there's at least one free TX descriptor but instead only if all descriptors have been handled. It's perfectly normal, especially in the DEVICE_POLLING case, that re_txeof() is called when only a part of the enqueued TX descriptors have been handled, causing the watchdog to be disarmed prematurely. o In re_encap(): - If m_defrag() fails just drop the packet like other NIC drivers do. This should only happen when there's a mbuf shortage, in which case it was possible to end up with an IFQ full of packets which couldn't be processed as they couldn't be defragmented as they were taking up all the mbufs themselves. This includes adjusting re_start() to not trying to prepend the mbuf (chain) if re_encap() has freed it. - Remove dupe initialization of members of struct rl_dmaload_arg to values that didn't change since trying to process the fragmented mbuf chain. While at it remove an unused member from struct rl_dmaload_arg. o In re_start() remove a abandoned, banal comment. The corresponding code was moved to re_attach() some time ago. With these changes re(4) now survives one day (until stopped) of hammering out packets here. Reviewed by: yongari MFC after: 2 weeks
2007-01-16 20:35:23 +00:00
/*
* If not all descriptors have been reaped yet, reload
* the timer so that we will eventually get another
* interrupt that will cause us to re-enter this routine.
* This is done in case the transmitter has gone idle.
*/
Whitespace pass
2004-07-06 02:46:53 +00:00
CSR_WRITE_4(sc, RL_TIMERCNT, 1);
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
#endif
o In re_newbuf() and re_encap() if re_dma_map_desc() aborts the mapping operation as it ran out of free descriptors or if there are too many segments in the first place, call bus_dmamap_unload() in order to unload the already loaded segments. For trying to map the defragmented mbuf (chain) in re_encap() this introduces re_dma_map_desc() setting arg.rl_maxsegs to 0 as a new failure mode. Previously we just ignored this case, corrupting our view of the TX ring. o In re_txeof(): - Don't clear IFF_DRV_OACTIVE unless there are at least 4 free TX descriptors. Further down the road re_encap() will bail if there aren't at least 4 free TX descriptors, causing re_start() to abort and prepend the dequeued mbuf again so it makes no sense to pretend we could process mbufs again when in fact we won't. While at it replace this magic 4 with a macro RL_TX_DESC_THLD throughout this driver. - Don't cancel the watchdog timeout as soon as there's at least one free TX descriptor but instead only if all descriptors have been handled. It's perfectly normal, especially in the DEVICE_POLLING case, that re_txeof() is called when only a part of the enqueued TX descriptors have been handled, causing the watchdog to be disarmed prematurely. o In re_encap(): - If m_defrag() fails just drop the packet like other NIC drivers do. This should only happen when there's a mbuf shortage, in which case it was possible to end up with an IFQ full of packets which couldn't be processed as they couldn't be defragmented as they were taking up all the mbufs themselves. This includes adjusting re_start() to not trying to prepend the mbuf (chain) if re_encap() has freed it. - Remove dupe initialization of members of struct rl_dmaload_arg to values that didn't change since trying to process the fragmented mbuf chain. While at it remove an unused member from struct rl_dmaload_arg. o In re_start() remove a abandoned, banal comment. The corresponding code was moved to re_attach() some time ago. With these changes re(4) now survives one day (until stopped) of hammering out packets here. Reviewed by: yongari MFC after: 2 weeks
2007-01-16 20:35:23 +00:00
} else
sc->rl_watchdog_timer = 0;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
static void
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_tick(void *xsc)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
struct rl_softc *sc;
- Use if_printf() and device_printf() in re(4) and remove rl_unit from the softc. - Use callout_init_mtx() and rather than timeout/untimeout in both rl(4) and re(4). - Fix locking for ifmedia by locking the driver in the ifmedia handlers rather than in the miibus functions. (re(4) didn't lock the mii stuff at all!) - Fix some locking in re_ioctl(). Note: the two drivers share the same softc declared in if_rlreg.h, so they had to be change simultaneously. MFC after: 1 week Tested by: several on rl(4), none on re(4)
2005-09-29 16:52:19 +00:00
struct mii_data *mii;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
sc = xsc;
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
RL_LOCK_ASSERT(sc);
- Use the re_tick() callout instead of if_slowtimo() for driving re_watchdog() in order to avoid races accessing if_timer. - Use bus_get_dma_tag() so re(4) works on platforms requiring it. - Remove invalid BUS_DMA_ALLOCNOW when creating the parent DMA tag and the tags that are used for static memory allocations. - Don't bother to set if_mtu to ETHERMTU, ether_ifattach() does that. - Remove an unused variable in re_intr().
2006-12-20 02:13:59 +00:00
mii = device_get_softc(sc->rl_miibus);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
mii_tick(mii);
Since we don't request reset for rlphy(4), the link state 'UP' event from mii(4) may not be delivered if valid link was already established. To address the issue, check current link state after driving MII_TICK. This should fix a regression introduced in r185753 on fast ethernet controllers. Reported by: csjp, Bruce Cran < bruce <> cran DOT org DOT uk > Tested by: csjp, Bruce Cran (initial version)
2008-12-22 00:46:22 +00:00
if ((sc->rl_flags & RL_FLAG_LINK) == 0)
re_miibus_statchg(sc->rl_dev);
Reclaim transmitted frames in re_tick(). This is for PCIe controllers that lose Tx completion interrupts under certain conditions. With this change it's safe to use MSI on PCIe controllers so enable MSI on these controllers.
2009-02-09 04:59:13 +00:00
/*
* Reclaim transmitted frames here. Technically it is not
* necessary to do here but it ensures periodic reclamation
* regardless of Tx completion interrupt which seems to be
Remove extra white spaces.
2010-11-05 18:23:43 +00:00
* lost on PCIe based controllers under certain situations.
Reclaim transmitted frames in re_tick(). This is for PCIe controllers that lose Tx completion interrupts under certain conditions. With this change it's safe to use MSI on PCIe controllers so enable MSI on these controllers.
2009-02-09 04:59:13 +00:00
*/
re_txeof(sc);
o Implemented miibus_statchg handler. It detects whether re(4) established a valid link or not. In miibus_statchg handler add a check for established link is valid one for the controller(e.g. 1000baseT is not a valid link for fastethernet controllers.) o Added a flag RE_FLAG_FASTETHER to mark fastethernet controllers. o Added additional check to know whether we've really encountered watchdog timeouts or missed Tx completion interrupts. This change may help to track down the cause of watchdog timeouts. o In interrupt handler, removed a check for link state change interrupt. Not all controllers have the bit and re(4) did not rely on the event for a long time. In addition, re(4) didn't request the interrupt in RL_IMR register. Tested by: rpaulo
2008-12-08 02:34:13 +00:00
re_watchdog(sc);
- Use if_printf() and device_printf() in re(4) and remove rl_unit from the softc. - Use callout_init_mtx() and rather than timeout/untimeout in both rl(4) and re(4). - Fix locking for ifmedia by locking the driver in the ifmedia handlers rather than in the miibus functions. (re(4) didn't lock the mii stuff at all!) - Fix some locking in re_ioctl(). Note: the two drivers share the same softc declared in if_rlreg.h, so they had to be change simultaneously. MFC after: 1 week Tested by: several on rl(4), none on re(4)
2005-09-29 16:52:19 +00:00
callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
#ifdef DEVICE_POLLING
When user_frac in the polling subsystem is low it is going to busy the CPU for too long period than necessary. Additively, interfaces are kept polled (in the tick) even if no more packets are available. In order to avoid such situations a new generic mechanism can be implemented in proactive way, keeping track of the time spent on any packet and fragmenting the time for any tick, stopping the processing as soon as possible. In order to implement such mechanism, the polling handler needs to change, returning the number of packets processed. While the intended logic is not part of this patch, the polling KPI is broken by this commit, adding an int return value and the new flag IFCAP_POLLING_NOCOUNT (which will signal that the return value is meaningless for the installed handler and checking should be skipped). Bump __FreeBSD_version in order to signal such situation. Reviewed by: emaste Sponsored by: Sandvine Incorporated
2009-05-30 15:14:44 +00:00
static int
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
struct rl_softc *sc = ifp->if_softc;
When user_frac in the polling subsystem is low it is going to busy the CPU for too long period than necessary. Additively, interfaces are kept polled (in the tick) even if no more packets are available. In order to avoid such situations a new generic mechanism can be implemented in proactive way, keeping track of the time spent on any packet and fragmenting the time for any tick, stopping the processing as soon as possible. In order to implement such mechanism, the polling handler needs to change, returning the number of packets processed. While the intended logic is not part of this patch, the polling KPI is broken by this commit, adding an int return value and the new flag IFCAP_POLLING_NOCOUNT (which will signal that the return value is meaningless for the installed handler and checking should be skipped). Bump __FreeBSD_version in order to signal such situation. Reviewed by: emaste Sponsored by: Sandvine Incorporated
2009-05-30 15:14:44 +00:00
int rx_npkts = 0;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
RL_LOCK(sc);
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
if (ifp->if_drv_flags & IFF_DRV_RUNNING)
When user_frac in the polling subsystem is low it is going to busy the CPU for too long period than necessary. Additively, interfaces are kept polled (in the tick) even if no more packets are available. In order to avoid such situations a new generic mechanism can be implemented in proactive way, keeping track of the time spent on any packet and fragmenting the time for any tick, stopping the processing as soon as possible. In order to implement such mechanism, the polling handler needs to change, returning the number of packets processed. While the intended logic is not part of this patch, the polling KPI is broken by this commit, adding an int return value and the new flag IFCAP_POLLING_NOCOUNT (which will signal that the return value is meaningless for the installed handler and checking should be skipped). Bump __FreeBSD_version in order to signal such situation. Reviewed by: emaste Sponsored by: Sandvine Incorporated
2009-05-30 15:14:44 +00:00
rx_npkts = re_poll_locked(ifp, cmd, count);
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
RL_UNLOCK(sc);
When user_frac in the polling subsystem is low it is going to busy the CPU for too long period than necessary. Additively, interfaces are kept polled (in the tick) even if no more packets are available. In order to avoid such situations a new generic mechanism can be implemented in proactive way, keeping track of the time spent on any packet and fragmenting the time for any tick, stopping the processing as soon as possible. In order to implement such mechanism, the polling handler needs to change, returning the number of packets processed. While the intended logic is not part of this patch, the polling KPI is broken by this commit, adding an int return value and the new flag IFCAP_POLLING_NOCOUNT (which will signal that the return value is meaningless for the installed handler and checking should be skipped). Bump __FreeBSD_version in order to signal such situation. Reviewed by: emaste Sponsored by: Sandvine Incorporated
2009-05-30 15:14:44 +00:00
return (rx_npkts);
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
}
When user_frac in the polling subsystem is low it is going to busy the CPU for too long period than necessary. Additively, interfaces are kept polled (in the tick) even if no more packets are available. In order to avoid such situations a new generic mechanism can be implemented in proactive way, keeping track of the time spent on any packet and fragmenting the time for any tick, stopping the processing as soon as possible. In order to implement such mechanism, the polling handler needs to change, returning the number of packets processed. While the intended logic is not part of this patch, the polling KPI is broken by this commit, adding an int return value and the new flag IFCAP_POLLING_NOCOUNT (which will signal that the return value is meaningless for the installed handler and checking should be skipped). Bump __FreeBSD_version in order to signal such situation. Reviewed by: emaste Sponsored by: Sandvine Incorporated
2009-05-30 15:14:44 +00:00
static int
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
{
struct rl_softc *sc = ifp->if_softc;
When user_frac in the polling subsystem is low it is going to busy the CPU for too long period than necessary. Additively, interfaces are kept polled (in the tick) even if no more packets are available. In order to avoid such situations a new generic mechanism can be implemented in proactive way, keeping track of the time spent on any packet and fragmenting the time for any tick, stopping the processing as soon as possible. In order to implement such mechanism, the polling handler needs to change, returning the number of packets processed. While the intended logic is not part of this patch, the polling KPI is broken by this commit, adding an int return value and the new flag IFCAP_POLLING_NOCOUNT (which will signal that the return value is meaningless for the installed handler and checking should be skipped). Bump __FreeBSD_version in order to signal such situation. Reviewed by: emaste Sponsored by: Sandvine Incorporated
2009-05-30 15:14:44 +00:00
int rx_npkts;
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
RL_LOCK_ASSERT(sc);
trim trailing white space.. call the re mutex by it's name.. MFC after: 3 days
2004-09-20 06:33:37 +00:00
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
sc->rxcycles = count;
When user_frac in the polling subsystem is low it is going to busy the CPU for too long period than necessary. Additively, interfaces are kept polled (in the tick) even if no more packets are available. In order to avoid such situations a new generic mechanism can be implemented in proactive way, keeping track of the time spent on any packet and fragmenting the time for any tick, stopping the processing as soon as possible. In order to implement such mechanism, the polling handler needs to change, returning the number of packets processed. While the intended logic is not part of this patch, the polling KPI is broken by this commit, adding an int return value and the new flag IFCAP_POLLING_NOCOUNT (which will signal that the return value is meaningless for the installed handler and checking should be skipped). Bump __FreeBSD_version in order to signal such situation. Reviewed by: emaste Sponsored by: Sandvine Incorporated
2009-05-30 15:14:44 +00:00
re_rxeof(sc, &rx_npkts);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
re_txeof(sc);
Unbreak build with POLLING. I should really listen and test with NOTES instead of the module build.
2005-03-13 01:54:41 +00:00
if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
Remove TX taskqueue and directly invoke re_start in interrupt task.
2011-01-25 23:27:28 +00:00
re_start_locked(ifp);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
u_int16_t status;
status = CSR_READ_2(sc, RL_ISR);
if (status == 0xffff)
When user_frac in the polling subsystem is low it is going to busy the CPU for too long period than necessary. Additively, interfaces are kept polled (in the tick) even if no more packets are available. In order to avoid such situations a new generic mechanism can be implemented in proactive way, keeping track of the time spent on any packet and fragmenting the time for any tick, stopping the processing as soon as possible. In order to implement such mechanism, the polling handler needs to change, returning the number of packets processed. While the intended logic is not part of this patch, the polling KPI is broken by this commit, adding an int return value and the new flag IFCAP_POLLING_NOCOUNT (which will signal that the return value is meaningless for the installed handler and checking should be skipped). Bump __FreeBSD_version in order to signal such situation. Reviewed by: emaste Sponsored by: Sandvine Incorporated
2009-05-30 15:14:44 +00:00
return (rx_npkts);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
if (status)
CSR_WRITE_2(sc, RL_ISR, status);
It seems that RealTek PCIe controllers require an explicit Tx poll command whenever Tx completion interrupt is raised. The Tx poll bit is cleared when all packets waiting to be transferred have been processed. This means the second Tx poll command can be silently ignored as the Tx poll bit could be still active while processing of previous Tx poll command is in progress. To address the issue re(4) used to invoke the Tx poll command in Tx completion handler whenever it detects there are pending packets in TxQ. However that still does not seem to completely eliminate watchdog timeouts seen on RealTek PCIe controllers. To fix the issue kick Tx poll command only after Tx completion interrupt is raised as this would indicate Tx is now idle state such that it can accept new Tx poll command again. While here apply this workaround for PCIe based controllers as other controllers does not seem to have this limitation. Tested by: Victor Balada Diaz < victor <> bsdes DOT net >
2008-12-17 08:18:11 +00:00
if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
(sc->rl_flags & RL_FLAG_PCIE))
CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/*
* XXX check behaviour on receiver stalls.
*/
Minimize interface reinitialization by checking IFF_DRV_RUNNING flag. This fixes unnecessary interface UP/DOWNs during getting an IP address via DHCP. Tested by: Warren Block ( wblock<> wonkity dot com )
2009-11-19 23:12:44 +00:00
if (status & RL_ISR_SYSTEM_ERR) {
ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
re_init_locked(sc);
Minimize interface reinitialization by checking IFF_DRV_RUNNING flag. This fixes unnecessary interface UP/DOWNs during getting an IP address via DHCP. Tested by: Warren Block ( wblock<> wonkity dot com )
2009-11-19 23:12:44 +00:00
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
When user_frac in the polling subsystem is low it is going to busy the CPU for too long period than necessary. Additively, interfaces are kept polled (in the tick) even if no more packets are available. In order to avoid such situations a new generic mechanism can be implemented in proactive way, keeping track of the time spent on any packet and fragmenting the time for any tick, stopping the processing as soon as possible. In order to implement such mechanism, the polling handler needs to change, returning the number of packets processed. While the intended logic is not part of this patch, the polling KPI is broken by this commit, adding an int return value and the new flag IFCAP_POLLING_NOCOUNT (which will signal that the return value is meaningless for the installed handler and checking should be skipped). Bump __FreeBSD_version in order to signal such situation. Reviewed by: emaste Sponsored by: Sandvine Incorporated
2009-05-30 15:14:44 +00:00
return (rx_npkts);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
#endif /* DEVICE_POLLING */
o break newbus api: add a new argument of type driver_filter_t to bus_setup_intr() o add an int return code to all fast handlers o retire INTR_FAST/IH_FAST For more info: http://docs.freebsd.org/cgi/getmsg.cgi?fetch=465712+0+current/freebsd-current Reviewed by: many Approved by: re@
2007-02-23 12:19:07 +00:00
static int
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_intr(void *arg)
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
{
struct rl_softc *sc;
uint16_t status;
sc = arg;
status = CSR_READ_2(sc, RL_ISR);
Fix the following bugs in re(4) - Correct the PCI ID for the 8169SC/8110SC in the device list (I added the macro for it to if_rlreg.h before, but forgot to use it.) - Remove the extra interrupt spinlock I added previously. After giving it some more thought, it's not really needed. - Work around a hardware bug in some versions of the 8169. When sending very small IP datagrams with checksum offload enabled, a conflict can occur between the TX autopadding feature and the hardware checksumming that can corrupt the outbound packet. This is the reason that checksum offload sometimes breaks NFS: if you're using NFS over UDP, and you're very unlucky, you might find yourself doing a fragmented NFS write where the last fragment is smaller than the minimum ethernet frame size (60 bytes). (It's rare, but if you keep NFS running long enough it'll happen.) If checksum offload is enabled, the chip will have to both autopad the fragment and calculate its checksum header. This confuses some revs of the 8169, causing the packet that appears on the wire to be corrupted. (The IP addresses and the checksum field are mangled.) This will cause the NFS write to fail. Unfortunately, when NFS retries, it sends the same write request over and over again, and it keeps failing, so NFS stays wedged. (A simple way to provoke the failure is to connect the failing system to a network with a known good machine and do "ping -s 1473 <badhost>" from the good system. The ping will fail.) Someone had previously worked around this using the heavy-handed approahch of just disabling checksum offload. The correct fix is to manually pad short frames where the TCP/IP stack has requested checksum offloading. This allows us to have checksum offload turned on by default but still let NFS work right. - Not a bug, but change the ID strings for devices with hardware rev 0x30000000 and 0x38000000 to both be 8168B/8111B. According to RealTek, they're both the same device, but 0x30000000 is an earlier silicon spin.
2006-07-30 23:25:21 +00:00
if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0)
o break newbus api: add a new argument of type driver_filter_t to bus_setup_intr() o add an int return code to all fast handlers o retire INTR_FAST/IH_FAST For more info: http://docs.freebsd.org/cgi/getmsg.cgi?fetch=465712+0+current/freebsd-current Reviewed by: many Approved by: re@
2007-02-23 12:19:07 +00:00
return (FILTER_STRAY);
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
CSR_WRITE_2(sc, RL_IMR, 0);
taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
o break newbus api: add a new argument of type driver_filter_t to bus_setup_intr() o add an int return code to all fast handlers o retire INTR_FAST/IH_FAST For more info: http://docs.freebsd.org/cgi/getmsg.cgi?fetch=465712+0+current/freebsd-current Reviewed by: many Approved by: re@
2007-02-23 12:19:07 +00:00
return (FILTER_HANDLED);
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
}
static void
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_int_task(void *arg, int npending)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
struct rl_softc *sc;
struct ifnet *ifp;
u_int16_t status;
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
int rval = 0;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
sc = arg;
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
ifp = sc->rl_ifp;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
RL_LOCK(sc);
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
status = CSR_READ_2(sc, RL_ISR);
CSR_WRITE_2(sc, RL_ISR, status);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
if (sc->suspended ||
(ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
RL_UNLOCK(sc);
return;
}
Add a PHY driver to support the built-in gigE PHY in the 8169S/8110S ethernet chips. This driver is pretty simple, however it contains special DSP initialization code which is needed in order to get the chip to negotiate a gigE link. (This special initialization may not be needed in subsequent chip revs.) Also: - Fix typo in if_rlreg.h (RL_GMEDIASTAT_1000MPS -> RL_GMEDIASTAT_1000MBPS) - Deal with shared interrupts in re_intr(): if interface isn't up, return. - Fix another bug in re_gmii_writereg() (properly apply data field mask) - Allow PHY driver to read the RL_GMEDIASTAT register via the re_gmii_readreg() register (this is register needed to determine real time link/media status).
2003-09-11 03:53:46 +00:00
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
#ifdef DEVICE_POLLING
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
if (ifp->if_capenable & IFCAP_POLLING) {
RL_UNLOCK(sc);
return;
}
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
#endif
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
When user_frac in the polling subsystem is low it is going to busy the CPU for too long period than necessary. Additively, interfaces are kept polled (in the tick) even if no more packets are available. In order to avoid such situations a new generic mechanism can be implemented in proactive way, keeping track of the time spent on any packet and fragmenting the time for any tick, stopping the processing as soon as possible. In order to implement such mechanism, the polling handler needs to change, returning the number of packets processed. While the intended logic is not part of this patch, the polling KPI is broken by this commit, adding an int return value and the new flag IFCAP_POLLING_NOCOUNT (which will signal that the return value is meaningless for the installed handler and checking should be skipped). Bump __FreeBSD_version in order to signal such situation. Reviewed by: emaste Sponsored by: Sandvine Incorporated
2009-05-30 15:14:44 +00:00
rval = re_rxeof(sc, NULL);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
It seems that RealTek PCIe controllers require an explicit Tx poll command whenever Tx completion interrupt is raised. The Tx poll bit is cleared when all packets waiting to be transferred have been processed. This means the second Tx poll command can be silently ignored as the Tx poll bit could be still active while processing of previous Tx poll command is in progress. To address the issue re(4) used to invoke the Tx poll command in Tx completion handler whenever it detects there are pending packets in TxQ. However that still does not seem to completely eliminate watchdog timeouts seen on RealTek PCIe controllers. To fix the issue kick Tx poll command only after Tx completion interrupt is raised as this would indicate Tx is now idle state such that it can accept new Tx poll command again. While here apply this workaround for PCIe based controllers as other controllers does not seem to have this limitation. Tested by: Victor Balada Diaz < victor <> bsdes DOT net >
2008-12-17 08:18:11 +00:00
/*
* Some chips will ignore a second TX request issued
* while an existing transmission is in progress. If
* the transmitter goes idle but there are still
* packets waiting to be sent, we need to restart the
* channel here to flush them out. This only seems to
* be required with the PCIe devices.
*/
if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
(sc->rl_flags & RL_FLAG_PCIE))
CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
Rearrange conditional compilation directives. This makes syntax highlighting work in vim.
2008-08-04 04:00:10 +00:00
if (status & (
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
#ifdef RE_TX_MODERATION
Rearrange conditional compilation directives. This makes syntax highlighting work in vim.
2008-08-04 04:00:10 +00:00
RL_ISR_TIMEOUT_EXPIRED|
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
#else
Rearrange conditional compilation directives. This makes syntax highlighting work in vim.
2008-08-04 04:00:10 +00:00
RL_ISR_TX_OK|
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
#endif
RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL))
re_txeof(sc);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Minimize interface reinitialization by checking IFF_DRV_RUNNING flag. This fixes unnecessary interface UP/DOWNs during getting an IP address via DHCP. Tested by: Warren Block ( wblock<> wonkity dot com )
2009-11-19 23:12:44 +00:00
if (status & RL_ISR_SYSTEM_ERR) {
ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
re_init_locked(sc);
Minimize interface reinitialization by checking IFF_DRV_RUNNING flag. This fixes unnecessary interface UP/DOWNs during getting an IP address via DHCP. Tested by: Warren Block ( wblock<> wonkity dot com )
2009-11-19 23:12:44 +00:00
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
Remove TX taskqueue and directly invoke re_start in interrupt task.
2011-01-25 23:27:28 +00:00
re_start_locked(ifp);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
RL_UNLOCK(sc);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) {
taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
return;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
Do not use interrupt taskqueue on controllers with MSI/MSI-X capability. One of reason using interrupt taskqueue in re(4) was to reduce number of TX/RX interrupts under load because re(4) controllers have no good TX/RX interrupt moderation mechanism. Basic TX interrupt moderation is done by hardware for most controllers but RX interrupt moderation through undocumented register showed poor RX performance so it was disabled in r215025. Using taskqueue to handle RX interrupt greatly reduced number of interrupts but re(4) consumed all available CPU cycles to run the taskqueue under high TX/RX network load. This can happen even with RTL810x fast ethernet controller and I believe this is not acceptable for most systems. To mitigate the issue, use one-shot timer register to moderate RX interrupts. The timer register provides programmable one-shot timer and can be used to suppress interrupt generation. The timer runs at 125MHZ on PCIe controllers so the minimum time allowed for the timer is 8ns. Data sheet says the register is 32 bits but experimentation shows only lower 13 bits are valid so maximum time that can be programmed is 65.528us. This yields theoretical maximum number of RX interrupts that could be generated per second is about 15260. Combined with TX completion interrupts re(4) shall generate less than 20k interrupts. This number is still slightly high compared to other intelligent ethernet controllers but system is very responsive even under high network load. Introduce sysctl variable dev.re.%d.int_rx_mod that controls amount of time to delay RX interrupt processing in units of us. Value 0 completely disables RX interrupt moderation. To provide old behavior for controllers that have MSI/MSI-X capability, introduce a new tunable hw.re.intr_filter. If the tunable is set to non-zero value, driver will use interrupt taskqueue. The default value of the tunable is 0. This tunable has no effect on controllers that has no MSI/MSI-X capability or if MSI/MSI-X is explicitly disabled by administrator. While I'm here cleanup interrupt setup/teardown since re(4) uses single MSI/MSI-X message at this moment.
2011-01-26 20:25:40 +00:00
static void
re_intr_msi(void *xsc)
{
struct rl_softc *sc;
struct ifnet *ifp;
uint16_t intrs, status;
sc = xsc;
RL_LOCK(sc);
ifp = sc->rl_ifp;
#ifdef DEVICE_POLLING
if (ifp->if_capenable & IFCAP_POLLING) {
RL_UNLOCK(sc);
return;
}
#endif
/* Disable interrupts. */
CSR_WRITE_2(sc, RL_IMR, 0);
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
RL_UNLOCK(sc);
return;
}
intrs = RL_INTRS_CPLUS;
status = CSR_READ_2(sc, RL_ISR);
CSR_WRITE_2(sc, RL_ISR, status);
if (sc->rl_int_rx_act > 0) {
intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
RL_ISR_RX_OVERRUN);
status &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW |
RL_ISR_RX_OVERRUN);
}
if (status & (RL_ISR_TIMEOUT_EXPIRED | RL_ISR_RX_OK | RL_ISR_RX_ERR |
RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) {
re_rxeof(sc, NULL);
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
if (sc->rl_int_rx_mod != 0 &&
(status & (RL_ISR_RX_OK | RL_ISR_RX_ERR |
RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) != 0) {
/* Rearm one-shot timer. */
CSR_WRITE_4(sc, RL_TIMERCNT, 1);
intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR |
RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN);
sc->rl_int_rx_act = 1;
} else {
intrs |= RL_ISR_RX_OK | RL_ISR_RX_ERR |
RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN;
sc->rl_int_rx_act = 0;
}
}
}
/*
* Some chips will ignore a second TX request issued
* while an existing transmission is in progress. If
* the transmitter goes idle but there are still
* packets waiting to be sent, we need to restart the
* channel here to flush them out. This only seems to
* be required with the PCIe devices.
*/
if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) &&
(sc->rl_flags & RL_FLAG_PCIE))
CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
if (status & (RL_ISR_TX_OK | RL_ISR_TX_ERR | RL_ISR_TX_DESC_UNAVAIL))
re_txeof(sc);
if (status & RL_ISR_SYSTEM_ERR) {
ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
re_init_locked(sc);
}
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
re_start_locked(ifp);
CSR_WRITE_2(sc, RL_IMR, intrs);
}
RL_UNLOCK(sc);
}
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
static int
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_encap(struct rl_softc *sc, struct mbuf **m_head)
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
{
struct rl_txdesc *txd, *txd_last;
bus_dma_segment_t segs[RL_NTXSEGS];
bus_dmamap_t map;
struct mbuf *m_new;
struct rl_desc *desc;
int nsegs, prod;
int i, error, ei, si;
int padlen;
VLAN hardware tag information should be set for all desciptors of a multi-descriptor transmission attempt. Datasheet said nothing about this requirements. This should fix a long-standing VLAN hardware tagging issues with re(4). Reported by: Giulio Ferro ( auryn AT zirakzigil DOT org ) Tested by: Giulio Ferro ( auryn AT zirakzigil DOT org )
2008-03-23 05:06:16 +00:00
uint32_t cmdstat, csum_flags, vlanctl;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
RL_LOCK_ASSERT(sc);
Fix build.
2008-01-15 03:47:24 +00:00
M_ASSERTPKTHDR((*m_head));
Another small update to the re(4) driver: - Change the workaround for the autopad/checksum offload bug so that instead of lying about the map size, we actually create a properly padded mbuf and map it as usual. The other trick works, but is ugly. This approach also gives us a chance to zero the pad space to avoid possibly leaking data. - With the PCIe devices, it looks issuing a TX command while there's already a transmission in progress doesn't have any effect. In other words, if you send two packets in rapid succession, the second one may end up sitting in the TX DMA ring until another transmit command is issued later in the future. Basically, if re_txeof() sees that there are still descriptors outstanding, it needs to manually resume the TX DMA channel by issuing another TX command to make sure all transmissions are flushed out. (The PCI devices seem to keep the TX channel moving until all descriptors have been consumed. I'm not sure why the PCIe devices behave differently.) (You can see this issue if you do the following test: plug an re(4) interface into another host via crossover cable, and from the other host do 'ping -c 2 <host with re(4) NIC>' to prime the ARP cache, then do 'ping -c 1 -s 1473 <host with re(4) NIC>'. You're supposed to see two packets sent in response, but you may only see one. If you do 'ping -c 1 -s 1473 <host with re(4) NIC>' again, you'll see two packets, but one will be the missing fragment from the last ping, followed by one of the fragments from this ping.) - Add the PCI ID for the US Robotics 997902 NIC, which is based on the RTL8169S. - Add a tsleep() of 1 second in re_detach() after the interrupt handler is disconnected. This should allow any tasks queued up by the ISR to drain. Now, I know you're supposed to use taskqueue_drain() for this, but something about the way taskqueue_drain() works with taskqueue_fast queues doesn't seem quite right, and I refuse to be tricked into fixing it.
2006-08-01 17:18:25 +00:00
/*
* With some of the RealTek chips, using the checksum offload
* support in conjunction with the autopadding feature results
* in the transmission of corrupt frames. For example, if we
* need to send a really small IP fragment that's less than 60
* bytes in size, and IP header checksumming is enabled, the
* resulting ethernet frame that appears on the wire will
In revision 1.70, 1.71 and 1.84 re(4) tried to workaround checksum offload bugs by manual padding for short IP/UDP frames. Unfortunately it seems that these workaround does not work reliably on newer PCIe variants of RealTek chips. To workaround the hardware bug, always pad short frames if Tx IP checksum offload is requested. It seems that the hardware has a bug in IP checksum offload handling. NetBSD manually pads short frames only when the length of IP frame is less than 28 bytes but I chose 60 bytes to safety. Also unconditionally set IP checksum offload bit in Tx descriptor if any TCP or UDP checksum offload is requested. This is the same way as Linux does but it's not mentioned in data sheet. Obtained from: NetBSD Tested by: remko, danger
2008-03-28 01:21:21 +00:00
* have garbled payload. To work around this, if TX IP checksum
Another small update to the re(4) driver: - Change the workaround for the autopad/checksum offload bug so that instead of lying about the map size, we actually create a properly padded mbuf and map it as usual. The other trick works, but is ugly. This approach also gives us a chance to zero the pad space to avoid possibly leaking data. - With the PCIe devices, it looks issuing a TX command while there's already a transmission in progress doesn't have any effect. In other words, if you send two packets in rapid succession, the second one may end up sitting in the TX DMA ring until another transmit command is issued later in the future. Basically, if re_txeof() sees that there are still descriptors outstanding, it needs to manually resume the TX DMA channel by issuing another TX command to make sure all transmissions are flushed out. (The PCI devices seem to keep the TX channel moving until all descriptors have been consumed. I'm not sure why the PCIe devices behave differently.) (You can see this issue if you do the following test: plug an re(4) interface into another host via crossover cable, and from the other host do 'ping -c 2 <host with re(4) NIC>' to prime the ARP cache, then do 'ping -c 1 -s 1473 <host with re(4) NIC>'. You're supposed to see two packets sent in response, but you may only see one. If you do 'ping -c 1 -s 1473 <host with re(4) NIC>' again, you'll see two packets, but one will be the missing fragment from the last ping, followed by one of the fragments from this ping.) - Add the PCI ID for the US Robotics 997902 NIC, which is based on the RTL8169S. - Add a tsleep() of 1 second in re_detach() after the interrupt handler is disconnected. This should allow any tasks queued up by the ISR to drain. Now, I know you're supposed to use taskqueue_drain() for this, but something about the way taskqueue_drain() works with taskqueue_fast queues doesn't seem quite right, and I refuse to be tricked into fixing it.
2006-08-01 17:18:25 +00:00
* offload is enabled, we always manually pad short frames out
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
* to the minimum ethernet frame size.
Another small update to the re(4) driver: - Change the workaround for the autopad/checksum offload bug so that instead of lying about the map size, we actually create a properly padded mbuf and map it as usual. The other trick works, but is ugly. This approach also gives us a chance to zero the pad space to avoid possibly leaking data. - With the PCIe devices, it looks issuing a TX command while there's already a transmission in progress doesn't have any effect. In other words, if you send two packets in rapid succession, the second one may end up sitting in the TX DMA ring until another transmit command is issued later in the future. Basically, if re_txeof() sees that there are still descriptors outstanding, it needs to manually resume the TX DMA channel by issuing another TX command to make sure all transmissions are flushed out. (The PCI devices seem to keep the TX channel moving until all descriptors have been consumed. I'm not sure why the PCIe devices behave differently.) (You can see this issue if you do the following test: plug an re(4) interface into another host via crossover cable, and from the other host do 'ping -c 2 <host with re(4) NIC>' to prime the ARP cache, then do 'ping -c 1 -s 1473 <host with re(4) NIC>'. You're supposed to see two packets sent in response, but you may only see one. If you do 'ping -c 1 -s 1473 <host with re(4) NIC>' again, you'll see two packets, but one will be the missing fragment from the last ping, followed by one of the fragments from this ping.) - Add the PCI ID for the US Robotics 997902 NIC, which is based on the RTL8169S. - Add a tsleep() of 1 second in re_detach() after the interrupt handler is disconnected. This should allow any tasks queued up by the ISR to drain. Now, I know you're supposed to use taskqueue_drain() for this, but something about the way taskqueue_drain() works with taskqueue_fast queues doesn't seem quite right, and I refuse to be tricked into fixing it.
2006-08-01 17:18:25 +00:00
*/
For RTL8139C+ controllers, have controller handle padding short checksum offload frames. Software workaround used for broken controllers(RTL8169, RTL8168, RTL8168B) seem to cause watchdog timeouts on RTL8139C+. Introduce a new flag RL_FLAG_AUTOPAD to mark automatic padding feature of controller and set it for RTL8139C+ and controllers that use new descriptor format. This fixes watchdog timeouts seen on RTL8139C+. Reported by: Dimitri Rodis < DimitriR <> integritasystems dot com > Tested by: Dimitri Rodis < DimitriR <> integritasystems dot com >
2009-04-20 07:13:04 +00:00
if ((sc->rl_flags & RL_FLAG_AUTOPAD) == 0 &&
Add basic support for RTL8168C, RTL8168CP, RTL8111C and RTL8111CP. ATM Tx/Rx checksum offload is supported but TSO and jumbo frame is not yet supported. Because these newer controllers use different descriptor formats, a flag RL_FLAG_DESCV2 flag was introduced to handle that case in Tx/Rx handler. Also newer controllers seems to require to not touch 'enable Tx/Rx bit' in RL_CPLUS_CMD register so don't blindly try to set that bits. Note, it seems that there is still power-saving related issue where driver fails to attach PHY. Rebooting seems to fix that issue but number of required reboots varys. Many thanks to users that helped during developement. I really appreciate their patient and test/feedbacks.
2008-07-02 07:54:53 +00:00
(*m_head)->m_pkthdr.len < RL_IP4CSUMTX_PADLEN &&
In revision 1.70, 1.71 and 1.84 re(4) tried to workaround checksum offload bugs by manual padding for short IP/UDP frames. Unfortunately it seems that these workaround does not work reliably on newer PCIe variants of RealTek chips. To workaround the hardware bug, always pad short frames if Tx IP checksum offload is requested. It seems that the hardware has a bug in IP checksum offload handling. NetBSD manually pads short frames only when the length of IP frame is less than 28 bytes but I chose 60 bytes to safety. Also unconditionally set IP checksum offload bit in Tx descriptor if any TCP or UDP checksum offload is requested. This is the same way as Linux does but it's not mentioned in data sheet. Obtained from: NetBSD Tested by: remko, danger
2008-03-28 01:21:21 +00:00
((*m_head)->m_pkthdr.csum_flags & CSUM_IP) != 0) {
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
padlen = RL_MIN_FRAMELEN - (*m_head)->m_pkthdr.len;
if (M_WRITABLE(*m_head) == 0) {
/* Get a writable copy. */
Mechanically substitute flags from historic mbuf allocator with malloc(9) flags in sys/dev.
2012-12-04 09:32:43 +00:00
m_new = m_dup(*m_head, M_NOWAIT);
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
m_freem(*m_head);
if (m_new == NULL) {
*m_head = NULL;
return (ENOBUFS);
}
*m_head = m_new;
}
if ((*m_head)->m_next != NULL ||
M_TRAILINGSPACE(*m_head) < padlen) {
Mechanically substitute flags from historic mbuf allocator with malloc(9) flags in sys/dev.
2012-12-04 09:32:43 +00:00
m_new = m_defrag(*m_head, M_NOWAIT);
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
if (m_new == NULL) {
m_freem(*m_head);
*m_head = NULL;
return (ENOBUFS);
}
} else
m_new = *m_head;
The TCP checksum offload handling in the 8111B/8168B and 8101E PCIe can apparently be confused by short TCP segments that have been manually padded to the minimum ethernet frame size. The driver does short frame padding in software as a workaround for a bug in the 8169 PCI devices that causes short IP fragments to be corrupted due to an apparent conflict between the hardware autopadding and hardware IP checksumming. To fix this, we avoid software padding for short TCP segments, since the hardware seems to autopad and checksum these correctly (even the older 8169 NICs get these right). Short UDP packets appear to be handled correctly in all cases. This should work around the IP header checksum bug in the 8169 while not tripping the TCP checksum bug in the 8111B/8168B and 8101E.
2007-01-25 17:30:30 +00:00
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
/*
* Manually pad short frames, and zero the pad space
* to avoid leaking data.
*/
bzero(mtod(m_new, char *) + m_new->m_pkthdr.len, padlen);
m_new->m_pkthdr.len += padlen;
m_new->m_len = m_new->m_pkthdr.len;
*m_head = m_new;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
prod = sc->rl_ldata.rl_tx_prodidx;
txd = &sc->rl_ldata.rl_tx_desc[prod];
error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
*m_head, segs, &nsegs, BUS_DMA_NOWAIT);
if (error == EFBIG) {
Mechanically substitute flags from historic mbuf allocator with malloc(9) flags in sys/dev.
2012-12-04 09:32:43 +00:00
m_new = m_collapse(*m_head, M_NOWAIT, RL_NTXSEGS);
o In re_newbuf() and re_encap() if re_dma_map_desc() aborts the mapping operation as it ran out of free descriptors or if there are too many segments in the first place, call bus_dmamap_unload() in order to unload the already loaded segments. For trying to map the defragmented mbuf (chain) in re_encap() this introduces re_dma_map_desc() setting arg.rl_maxsegs to 0 as a new failure mode. Previously we just ignored this case, corrupting our view of the TX ring. o In re_txeof(): - Don't clear IFF_DRV_OACTIVE unless there are at least 4 free TX descriptors. Further down the road re_encap() will bail if there aren't at least 4 free TX descriptors, causing re_start() to abort and prepend the dequeued mbuf again so it makes no sense to pretend we could process mbufs again when in fact we won't. While at it replace this magic 4 with a macro RL_TX_DESC_THLD throughout this driver. - Don't cancel the watchdog timeout as soon as there's at least one free TX descriptor but instead only if all descriptors have been handled. It's perfectly normal, especially in the DEVICE_POLLING case, that re_txeof() is called when only a part of the enqueued TX descriptors have been handled, causing the watchdog to be disarmed prematurely. o In re_encap(): - If m_defrag() fails just drop the packet like other NIC drivers do. This should only happen when there's a mbuf shortage, in which case it was possible to end up with an IFQ full of packets which couldn't be processed as they couldn't be defragmented as they were taking up all the mbufs themselves. This includes adjusting re_start() to not trying to prepend the mbuf (chain) if re_encap() has freed it. - Remove dupe initialization of members of struct rl_dmaload_arg to values that didn't change since trying to process the fragmented mbuf chain. While at it remove an unused member from struct rl_dmaload_arg. o In re_start() remove a abandoned, banal comment. The corresponding code was moved to re_attach() some time ago. With these changes re(4) now survives one day (until stopped) of hammering out packets here. Reviewed by: yongari MFC after: 2 weeks
2007-01-16 20:35:23 +00:00
if (m_new == NULL) {
m_freem(*m_head);
*m_head = NULL;
pass in pointer to m_head to re_encap because m_defrag could free the original mbuf causing a free'd mbuf passed to bpf later and panic'ing the system.. This should only effect jumbo frames. MFC after: 5 days
2004-09-18 18:08:28 +00:00
return (ENOBUFS);
o In re_newbuf() and re_encap() if re_dma_map_desc() aborts the mapping operation as it ran out of free descriptors or if there are too many segments in the first place, call bus_dmamap_unload() in order to unload the already loaded segments. For trying to map the defragmented mbuf (chain) in re_encap() this introduces re_dma_map_desc() setting arg.rl_maxsegs to 0 as a new failure mode. Previously we just ignored this case, corrupting our view of the TX ring. o In re_txeof(): - Don't clear IFF_DRV_OACTIVE unless there are at least 4 free TX descriptors. Further down the road re_encap() will bail if there aren't at least 4 free TX descriptors, causing re_start() to abort and prepend the dequeued mbuf again so it makes no sense to pretend we could process mbufs again when in fact we won't. While at it replace this magic 4 with a macro RL_TX_DESC_THLD throughout this driver. - Don't cancel the watchdog timeout as soon as there's at least one free TX descriptor but instead only if all descriptors have been handled. It's perfectly normal, especially in the DEVICE_POLLING case, that re_txeof() is called when only a part of the enqueued TX descriptors have been handled, causing the watchdog to be disarmed prematurely. o In re_encap(): - If m_defrag() fails just drop the packet like other NIC drivers do. This should only happen when there's a mbuf shortage, in which case it was possible to end up with an IFQ full of packets which couldn't be processed as they couldn't be defragmented as they were taking up all the mbufs themselves. This includes adjusting re_start() to not trying to prepend the mbuf (chain) if re_encap() has freed it. - Remove dupe initialization of members of struct rl_dmaload_arg to values that didn't change since trying to process the fragmented mbuf chain. While at it remove an unused member from struct rl_dmaload_arg. o In re_start() remove a abandoned, banal comment. The corresponding code was moved to re_attach() some time ago. With these changes re(4) now survives one day (until stopped) of hammering out packets here. Reviewed by: yongari MFC after: 2 weeks
2007-01-16 20:35:23 +00:00
}
*m_head = m_new;
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag,
txd->tx_dmamap, *m_head, segs, &nsegs, BUS_DMA_NOWAIT);
if (error != 0) {
m_freem(*m_head);
o In re_newbuf() and re_encap() if re_dma_map_desc() aborts the mapping operation as it ran out of free descriptors or if there are too many segments in the first place, call bus_dmamap_unload() in order to unload the already loaded segments. For trying to map the defragmented mbuf (chain) in re_encap() this introduces re_dma_map_desc() setting arg.rl_maxsegs to 0 as a new failure mode. Previously we just ignored this case, corrupting our view of the TX ring. o In re_txeof(): - Don't clear IFF_DRV_OACTIVE unless there are at least 4 free TX descriptors. Further down the road re_encap() will bail if there aren't at least 4 free TX descriptors, causing re_start() to abort and prepend the dequeued mbuf again so it makes no sense to pretend we could process mbufs again when in fact we won't. While at it replace this magic 4 with a macro RL_TX_DESC_THLD throughout this driver. - Don't cancel the watchdog timeout as soon as there's at least one free TX descriptor but instead only if all descriptors have been handled. It's perfectly normal, especially in the DEVICE_POLLING case, that re_txeof() is called when only a part of the enqueued TX descriptors have been handled, causing the watchdog to be disarmed prematurely. o In re_encap(): - If m_defrag() fails just drop the packet like other NIC drivers do. This should only happen when there's a mbuf shortage, in which case it was possible to end up with an IFQ full of packets which couldn't be processed as they couldn't be defragmented as they were taking up all the mbufs themselves. This includes adjusting re_start() to not trying to prepend the mbuf (chain) if re_encap() has freed it. - Remove dupe initialization of members of struct rl_dmaload_arg to values that didn't change since trying to process the fragmented mbuf chain. While at it remove an unused member from struct rl_dmaload_arg. o In re_start() remove a abandoned, banal comment. The corresponding code was moved to re_attach() some time ago. With these changes re(4) now survives one day (until stopped) of hammering out packets here. Reviewed by: yongari MFC after: 2 weeks
2007-01-16 20:35:23 +00:00
*m_head = NULL;
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
return (error);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
} else if (error != 0)
return (error);
if (nsegs == 0) {
m_freem(*m_head);
*m_head = NULL;
return (EIO);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
/* Check for number of available descriptors. */
if (sc->rl_ldata.rl_tx_free - nsegs <= 1) {
bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap);
return (ENOBUFS);
}
bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap,
BUS_DMASYNC_PREWRITE);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/*
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
* Set up checksum offload. Note: checksum offload bits must
* appear in all descriptors of a multi-descriptor transmit
* attempt. This is according to testing done with an 8169
* chip. This is a requirement.
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
*/
Add basic support for RTL8168C, RTL8168CP, RTL8111C and RTL8111CP. ATM Tx/Rx checksum offload is supported but TSO and jumbo frame is not yet supported. Because these newer controllers use different descriptor formats, a flag RL_FLAG_DESCV2 flag was introduced to handle that case in Tx/Rx handler. Also newer controllers seems to require to not touch 'enable Tx/Rx bit' in RL_CPLUS_CMD register so don't blindly try to set that bits. Note, it seems that there is still power-saving related issue where driver fails to attach PHY. Rebooting seems to fix that issue but number of required reboots varys. Many thanks to users that helped during developement. I really appreciate their patient and test/feedbacks.
2008-07-02 07:54:53 +00:00
vlanctl = 0;
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
csum_flags = 0;
Implement TSO on RealTek RTL8168/8111 C or later controllers. RealTek changed TX descriptor format for later controllers so these controllers require MSS configuration in different location of TX descriptor. TSO is enabled by default for controllers that use new descriptor format. For old controllers, TSO is still disabled by default due to broken frames under certain conditions but users can enable it. Special thanks to Hayes Wang at RealTek. MFC after: 2 weeks
2011-01-10 23:28:46 +00:00
if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
if ((sc->rl_flags & RL_FLAG_DESCV2) != 0) {
csum_flags |= RL_TDESC_CMD_LGSEND;
vlanctl |= ((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
RL_TDESC_CMD_MSSVALV2_SHIFT);
} else {
csum_flags |= RL_TDESC_CMD_LGSEND |
((uint32_t)(*m_head)->m_pkthdr.tso_segsz <<
RL_TDESC_CMD_MSSVAL_SHIFT);
}
} else {
In revision 1.70, 1.71 and 1.84 re(4) tried to workaround checksum offload bugs by manual padding for short IP/UDP frames. Unfortunately it seems that these workaround does not work reliably on newer PCIe variants of RealTek chips. To workaround the hardware bug, always pad short frames if Tx IP checksum offload is requested. It seems that the hardware has a bug in IP checksum offload handling. NetBSD manually pads short frames only when the length of IP frame is less than 28 bytes but I chose 60 bytes to safety. Also unconditionally set IP checksum offload bit in Tx descriptor if any TCP or UDP checksum offload is requested. This is the same way as Linux does but it's not mentioned in data sheet. Obtained from: NetBSD Tested by: remko, danger
2008-03-28 01:21:21 +00:00
/*
* Unconditionally enable IP checksum if TCP or UDP
* checksum is required. Otherwise, TCP/UDP checksum
Correct comment typos.
2013-06-28 14:41:06 +00:00
* doesn't make effects.
In revision 1.70, 1.71 and 1.84 re(4) tried to workaround checksum offload bugs by manual padding for short IP/UDP frames. Unfortunately it seems that these workaround does not work reliably on newer PCIe variants of RealTek chips. To workaround the hardware bug, always pad short frames if Tx IP checksum offload is requested. It seems that the hardware has a bug in IP checksum offload handling. NetBSD manually pads short frames only when the length of IP frame is less than 28 bytes but I chose 60 bytes to safety. Also unconditionally set IP checksum offload bit in Tx descriptor if any TCP or UDP checksum offload is requested. This is the same way as Linux does but it's not mentioned in data sheet. Obtained from: NetBSD Tested by: remko, danger
2008-03-28 01:21:21 +00:00
*/
if (((*m_head)->m_pkthdr.csum_flags & RE_CSUM_FEATURES) != 0) {
Add basic support for RTL8168C, RTL8168CP, RTL8111C and RTL8111CP. ATM Tx/Rx checksum offload is supported but TSO and jumbo frame is not yet supported. Because these newer controllers use different descriptor formats, a flag RL_FLAG_DESCV2 flag was introduced to handle that case in Tx/Rx handler. Also newer controllers seems to require to not touch 'enable Tx/Rx bit' in RL_CPLUS_CMD register so don't blindly try to set that bits. Note, it seems that there is still power-saving related issue where driver fails to attach PHY. Rebooting seems to fix that issue but number of required reboots varys. Many thanks to users that helped during developement. I really appreciate their patient and test/feedbacks.
2008-07-02 07:54:53 +00:00
if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) {
csum_flags |= RL_TDESC_CMD_IPCSUM;
if (((*m_head)->m_pkthdr.csum_flags &
CSUM_TCP) != 0)
csum_flags |= RL_TDESC_CMD_TCPCSUM;
if (((*m_head)->m_pkthdr.csum_flags &
CSUM_UDP) != 0)
csum_flags |= RL_TDESC_CMD_UDPCSUM;
} else {
vlanctl |= RL_TDESC_CMD_IPCSUMV2;
if (((*m_head)->m_pkthdr.csum_flags &
CSUM_TCP) != 0)
vlanctl |= RL_TDESC_CMD_TCPCSUMV2;
if (((*m_head)->m_pkthdr.csum_flags &
CSUM_UDP) != 0)
vlanctl |= RL_TDESC_CMD_UDPCSUMV2;
}
In revision 1.70, 1.71 and 1.84 re(4) tried to workaround checksum offload bugs by manual padding for short IP/UDP frames. Unfortunately it seems that these workaround does not work reliably on newer PCIe variants of RealTek chips. To workaround the hardware bug, always pad short frames if Tx IP checksum offload is requested. It seems that the hardware has a bug in IP checksum offload handling. NetBSD manually pads short frames only when the length of IP frame is less than 28 bytes but I chose 60 bytes to safety. Also unconditionally set IP checksum offload bit in Tx descriptor if any TCP or UDP checksum offload is requested. This is the same way as Linux does but it's not mentioned in data sheet. Obtained from: NetBSD Tested by: remko, danger
2008-03-28 01:21:21 +00:00
}
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
VLAN hardware tag information should be set for all desciptors of a multi-descriptor transmission attempt. Datasheet said nothing about this requirements. This should fix a long-standing VLAN hardware tagging issues with re(4). Reported by: Giulio Ferro ( auryn AT zirakzigil DOT org ) Tested by: Giulio Ferro ( auryn AT zirakzigil DOT org )
2008-03-23 05:06:16 +00:00
/*
* Set up hardware VLAN tagging. Note: vlan tag info must
* appear in all descriptors of a multi-descriptor
* transmission attempt.
*/
if ((*m_head)->m_flags & M_VLANTAG)
Fix VLAN hardware tag insertion/stripping on big-endian architectures. Reported by: naddy Tested on: sparc64 MFC after: 1 week
2008-08-13 03:40:08 +00:00
vlanctl |= bswap16((*m_head)->m_pkthdr.ether_vtag) |
Add basic support for RTL8168C, RTL8168CP, RTL8111C and RTL8111CP. ATM Tx/Rx checksum offload is supported but TSO and jumbo frame is not yet supported. Because these newer controllers use different descriptor formats, a flag RL_FLAG_DESCV2 flag was introduced to handle that case in Tx/Rx handler. Also newer controllers seems to require to not touch 'enable Tx/Rx bit' in RL_CPLUS_CMD register so don't blindly try to set that bits. Note, it seems that there is still power-saving related issue where driver fails to attach PHY. Rebooting seems to fix that issue but number of required reboots varys. Many thanks to users that helped during developement. I really appreciate their patient and test/feedbacks.
2008-07-02 07:54:53 +00:00
RL_TDESC_VLANCTL_TAG;
VLAN hardware tag information should be set for all desciptors of a multi-descriptor transmission attempt. Datasheet said nothing about this requirements. This should fix a long-standing VLAN hardware tagging issues with re(4). Reported by: Giulio Ferro ( auryn AT zirakzigil DOT org ) Tested by: Giulio Ferro ( auryn AT zirakzigil DOT org )
2008-03-23 05:06:16 +00:00
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
si = prod;
for (i = 0; i < nsegs; i++, prod = RL_TX_DESC_NXT(sc, prod)) {
desc = &sc->rl_ldata.rl_tx_list[prod];
Add basic support for RTL8168C, RTL8168CP, RTL8111C and RTL8111CP. ATM Tx/Rx checksum offload is supported but TSO and jumbo frame is not yet supported. Because these newer controllers use different descriptor formats, a flag RL_FLAG_DESCV2 flag was introduced to handle that case in Tx/Rx handler. Also newer controllers seems to require to not touch 'enable Tx/Rx bit' in RL_CPLUS_CMD register so don't blindly try to set that bits. Note, it seems that there is still power-saving related issue where driver fails to attach PHY. Rebooting seems to fix that issue but number of required reboots varys. Many thanks to users that helped during developement. I really appreciate their patient and test/feedbacks.
2008-07-02 07:54:53 +00:00
desc->rl_vlanctl = htole32(vlanctl);
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr));
desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr));
cmdstat = segs[i].ds_len;
if (i != 0)
cmdstat |= RL_TDESC_CMD_OWN;
if (prod == sc->rl_ldata.rl_tx_desc_cnt - 1)
cmdstat |= RL_TDESC_CMD_EOR;
desc->rl_cmdstat = htole32(cmdstat | csum_flags);
sc->rl_ldata.rl_tx_free--;
}
/* Update producer index. */
sc->rl_ldata.rl_tx_prodidx = prod;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
/* Set EOF on the last descriptor. */
ei = RL_TX_DESC_PRV(sc, prod);
desc = &sc->rl_ldata.rl_tx_list[ei];
desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF);
desc = &sc->rl_ldata.rl_tx_list[si];
/* Set SOF and transfer ownership of packet to the chip. */
desc->rl_cmdstat |= htole32(RL_TDESC_CMD_OWN | RL_TDESC_CMD_SOF);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
/*
* Insure that the map for this transmission
* is placed at the array index of the last descriptor
* in this chain. (Swap last and first dmamaps.)
*/
txd_last = &sc->rl_ldata.rl_tx_desc[ei];
map = txd->tx_dmamap;
txd->tx_dmamap = txd_last->tx_dmamap;
txd_last->tx_dmamap = map;
txd_last->tx_m = *m_head;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Whitespace pass
2004-07-06 02:46:53 +00:00
return (0);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
static void
Remove TX taskqueue and directly invoke re_start in interrupt task.
2011-01-25 23:27:28 +00:00
re_start(struct ifnet *ifp)
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
{
Remove TX taskqueue and directly invoke re_start in interrupt task.
2011-01-25 23:27:28 +00:00
struct rl_softc *sc;
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
Remove TX taskqueue and directly invoke re_start in interrupt task.
2011-01-25 23:27:28 +00:00
sc = ifp->if_softc;
RL_LOCK(sc);
re_start_locked(ifp);
RL_UNLOCK(sc);
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/*
* Main transmit routine for C+ and gigE NICs.
*/
static void
Remove TX taskqueue and directly invoke re_start in interrupt task.
2011-01-25 23:27:28 +00:00
re_start_locked(struct ifnet *ifp)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
struct rl_softc *sc;
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
struct mbuf *m_head;
int queued;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
sc = ifp->if_softc;
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#ifdef DEV_NETMAP
/* XXX is this necessary ? */
if (ifp->if_capenable & IFCAP_NETMAP) {
struct netmap_kring *kring = &NA(ifp)->tx_rings[0];
if (sc->rl_ldata.rl_tx_prodidx != kring->nr_hwcur) {
/* kick the tx unit */
CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
#ifdef RE_TX_MODERATION
CSR_WRITE_4(sc, RL_TIMERCNT, 1);
#endif
sc->rl_watchdog_timer = 5;
}
return;
}
#endif /* DEV_NETMAP */
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
Remove TX taskqueue and directly invoke re_start in interrupt task.
2011-01-25 23:27:28 +00:00
IFF_DRV_RUNNING || (sc->rl_flags & RL_FLAG_LINK) == 0)
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
return;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
for (queued = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) &&
sc->rl_ldata.rl_tx_free > 1;) {
ALTQ support for re(4). Submitted by: Chris Dionissopoulos, Theo Schlossnagle PR: kern/78681 MFC after: 2 weeks
2005-03-12 17:35:37 +00:00
IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
if (m_head == NULL)
break;
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
if (re_encap(sc, &m_head) != 0) {
o In re_newbuf() and re_encap() if re_dma_map_desc() aborts the mapping operation as it ran out of free descriptors or if there are too many segments in the first place, call bus_dmamap_unload() in order to unload the already loaded segments. For trying to map the defragmented mbuf (chain) in re_encap() this introduces re_dma_map_desc() setting arg.rl_maxsegs to 0 as a new failure mode. Previously we just ignored this case, corrupting our view of the TX ring. o In re_txeof(): - Don't clear IFF_DRV_OACTIVE unless there are at least 4 free TX descriptors. Further down the road re_encap() will bail if there aren't at least 4 free TX descriptors, causing re_start() to abort and prepend the dequeued mbuf again so it makes no sense to pretend we could process mbufs again when in fact we won't. While at it replace this magic 4 with a macro RL_TX_DESC_THLD throughout this driver. - Don't cancel the watchdog timeout as soon as there's at least one free TX descriptor but instead only if all descriptors have been handled. It's perfectly normal, especially in the DEVICE_POLLING case, that re_txeof() is called when only a part of the enqueued TX descriptors have been handled, causing the watchdog to be disarmed prematurely. o In re_encap(): - If m_defrag() fails just drop the packet like other NIC drivers do. This should only happen when there's a mbuf shortage, in which case it was possible to end up with an IFQ full of packets which couldn't be processed as they couldn't be defragmented as they were taking up all the mbufs themselves. This includes adjusting re_start() to not trying to prepend the mbuf (chain) if re_encap() has freed it. - Remove dupe initialization of members of struct rl_dmaload_arg to values that didn't change since trying to process the fragmented mbuf chain. While at it remove an unused member from struct rl_dmaload_arg. o In re_start() remove a abandoned, banal comment. The corresponding code was moved to re_attach() some time ago. With these changes re(4) now survives one day (until stopped) of hammering out packets here. Reviewed by: yongari MFC after: 2 weeks
2007-01-16 20:35:23 +00:00
if (m_head == NULL)
break;
ALTQ support for re(4). Submitted by: Chris Dionissopoulos, Theo Schlossnagle PR: kern/78681 MFC after: 2 weeks
2005-03-12 17:35:37 +00:00
IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
Propagate rename of IFF_OACTIVE and IFF_RUNNING to IFF_DRV_OACTIVE and IFF_DRV_RUNNING, as well as the move from ifnet.if_flags to ifnet.if_drv_flags. Device drivers are now responsible for synchronizing access to these flags, as they are in if_drv_flags. This helps prevent races between the network stack and device driver in maintaining the interface flags field. Many __FreeBSD__ and __FreeBSD_version checks maintained and continued; some less so. Reviewed by: pjd, bz MFC after: 7 days
2005-08-09 10:20:02 +00:00
ifp->if_drv_flags |= IFF_DRV_OACTIVE;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
break;
}
/*
* If there's a BPF listener, bounce a copy of this frame
* to him.
*/
Catch up the rest of the drivers with the ether_vlan_mtap modifications. If these drivers are setting M_VLANTAG because they are stripping the layer 2 802.1Q headers, then they need to be re-inserting them so any bpf(4) peers can properly decode them. It should be noted that this is compiled tested only. MFC after: 3 weeks
2007-03-04 03:38:08 +00:00
ETHER_BPF_MTAP(ifp, m_head);
ALTQ support for re(4). Submitted by: Chris Dionissopoulos, Theo Schlossnagle PR: kern/78681 MFC after: 2 weeks
2005-03-12 17:35:37 +00:00
queued++;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
if (queued == 0) {
#ifdef RE_TX_MODERATION
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt)
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
CSR_WRITE_4(sc, RL_TIMERCNT, 1);
#endif
ALTQ support for re(4). Submitted by: Chris Dionissopoulos, Theo Schlossnagle PR: kern/78681 MFC after: 2 weeks
2005-03-12 17:35:37 +00:00
return;
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
}
ALTQ support for re(4). Submitted by: Chris Dionissopoulos, Theo Schlossnagle PR: kern/78681 MFC after: 2 weeks
2005-03-12 17:35:37 +00:00
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/* Flush the TX descriptors */
bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag,
sc->rl_ldata.rl_tx_list_map,
BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD);
Another small update to the re(4) driver: - Change the workaround for the autopad/checksum offload bug so that instead of lying about the map size, we actually create a properly padded mbuf and map it as usual. The other trick works, but is ugly. This approach also gives us a chance to zero the pad space to avoid possibly leaking data. - With the PCIe devices, it looks issuing a TX command while there's already a transmission in progress doesn't have any effect. In other words, if you send two packets in rapid succession, the second one may end up sitting in the TX DMA ring until another transmit command is issued later in the future. Basically, if re_txeof() sees that there are still descriptors outstanding, it needs to manually resume the TX DMA channel by issuing another TX command to make sure all transmissions are flushed out. (The PCI devices seem to keep the TX channel moving until all descriptors have been consumed. I'm not sure why the PCIe devices behave differently.) (You can see this issue if you do the following test: plug an re(4) interface into another host via crossover cable, and from the other host do 'ping -c 2 <host with re(4) NIC>' to prime the ARP cache, then do 'ping -c 1 -s 1473 <host with re(4) NIC>'. You're supposed to see two packets sent in response, but you may only see one. If you do 'ping -c 1 -s 1473 <host with re(4) NIC>' again, you'll see two packets, but one will be the missing fragment from the last ping, followed by one of the fragments from this ping.) - Add the PCI ID for the US Robotics 997902 NIC, which is based on the RTL8169S. - Add a tsleep() of 1 second in re_detach() after the interrupt handler is disconnected. This should allow any tasks queued up by the ISR to drain. Now, I know you're supposed to use taskqueue_drain() for this, but something about the way taskqueue_drain() works with taskqueue_fast queues doesn't seem quite right, and I refuse to be tricked into fixing it.
2006-08-01 17:18:25 +00:00
CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
#ifdef RE_TX_MODERATION
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/*
* Use the countdown timer for interrupt moderation.
* 'TX done' interrupts are disabled. Instead, we reset the
* countdown timer, which will begin counting until it hits
* the value in the TIMERINT register, and then trigger an
* interrupt. Each time we write to the TIMERCNT register,
* the timer count is reset to 0.
*/
CSR_WRITE_4(sc, RL_TIMERCNT, 1);
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
#endif
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/*
* Set a timeout in case the chip goes out to lunch.
*/
- Use the re_tick() callout instead of if_slowtimo() for driving re_watchdog() in order to avoid races accessing if_timer. - Use bus_get_dma_tag() so re(4) works on platforms requiring it. - Remove invalid BUS_DMA_ALLOCNOW when creating the parent DMA tag and the tags that are used for static memory allocations. - Don't bother to set if_mtu to ETHERMTU, ether_ifattach() does that. - Remove an unused variable in re_intr().
2006-12-20 02:13:59 +00:00
sc->rl_watchdog_timer = 5;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
static void
re_set_jumbo(struct rl_softc *sc, int jumbo)
{
if (sc->rl_hwrev->rl_rev == RL_HWREV_8168E_VL) {
pci_set_max_read_req(sc->rl_dev, 4096);
return;
}
CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
if (jumbo != 0) {
Use correct Config registers for RTL8139 family. Unlike RTL8168 and RTL810x family , RTL8139 has different register map for Config registers. While here, follow the lead of re(4) in WOL configuration. - Disable WOL_UCAST and WOL_MCAST capabilities by default. - Config5 register write does not need to unlock EEPROM access on RTL8139 family but unlocking EEPROM access does not affect its operation and make it consistent with re(4). Reported by: Matt Renzelmann mjr <> cs dot wisc dot edu
2012-02-25 04:54:51 +00:00
CSR_WRITE_1(sc, sc->rl_cfg3, CSR_READ_1(sc, sc->rl_cfg3) |
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
RL_CFG3_JUMBO_EN0);
switch (sc->rl_hwrev->rl_rev) {
case RL_HWREV_8168DP:
break;
case RL_HWREV_8168E:
Use correct Config registers for RTL8139 family. Unlike RTL8168 and RTL810x family , RTL8139 has different register map for Config registers. While here, follow the lead of re(4) in WOL configuration. - Disable WOL_UCAST and WOL_MCAST capabilities by default. - Config5 register write does not need to unlock EEPROM access on RTL8139 family but unlocking EEPROM access does not affect its operation and make it consistent with re(4). Reported by: Matt Renzelmann mjr <> cs dot wisc dot edu
2012-02-25 04:54:51 +00:00
CSR_WRITE_1(sc, sc->rl_cfg4,
CSR_READ_1(sc, sc->rl_cfg4) | 0x01);
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
break;
default:
Use correct Config registers for RTL8139 family. Unlike RTL8168 and RTL810x family , RTL8139 has different register map for Config registers. While here, follow the lead of re(4) in WOL configuration. - Disable WOL_UCAST and WOL_MCAST capabilities by default. - Config5 register write does not need to unlock EEPROM access on RTL8139 family but unlocking EEPROM access does not affect its operation and make it consistent with re(4). Reported by: Matt Renzelmann mjr <> cs dot wisc dot edu
2012-02-25 04:54:51 +00:00
CSR_WRITE_1(sc, sc->rl_cfg4,
CSR_READ_1(sc, sc->rl_cfg4) | RL_CFG4_JUMBO_EN1);
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
}
} else {
Use correct Config registers for RTL8139 family. Unlike RTL8168 and RTL810x family , RTL8139 has different register map for Config registers. While here, follow the lead of re(4) in WOL configuration. - Disable WOL_UCAST and WOL_MCAST capabilities by default. - Config5 register write does not need to unlock EEPROM access on RTL8139 family but unlocking EEPROM access does not affect its operation and make it consistent with re(4). Reported by: Matt Renzelmann mjr <> cs dot wisc dot edu
2012-02-25 04:54:51 +00:00
CSR_WRITE_1(sc, sc->rl_cfg3, CSR_READ_1(sc, sc->rl_cfg3) &
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
~RL_CFG3_JUMBO_EN0);
switch (sc->rl_hwrev->rl_rev) {
case RL_HWREV_8168DP:
break;
case RL_HWREV_8168E:
Use correct Config registers for RTL8139 family. Unlike RTL8168 and RTL810x family , RTL8139 has different register map for Config registers. While here, follow the lead of re(4) in WOL configuration. - Disable WOL_UCAST and WOL_MCAST capabilities by default. - Config5 register write does not need to unlock EEPROM access on RTL8139 family but unlocking EEPROM access does not affect its operation and make it consistent with re(4). Reported by: Matt Renzelmann mjr <> cs dot wisc dot edu
2012-02-25 04:54:51 +00:00
CSR_WRITE_1(sc, sc->rl_cfg4,
CSR_READ_1(sc, sc->rl_cfg4) & ~0x01);
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
break;
default:
Use correct Config registers for RTL8139 family. Unlike RTL8168 and RTL810x family , RTL8139 has different register map for Config registers. While here, follow the lead of re(4) in WOL configuration. - Disable WOL_UCAST and WOL_MCAST capabilities by default. - Config5 register write does not need to unlock EEPROM access on RTL8139 family but unlocking EEPROM access does not affect its operation and make it consistent with re(4). Reported by: Matt Renzelmann mjr <> cs dot wisc dot edu
2012-02-25 04:54:51 +00:00
CSR_WRITE_1(sc, sc->rl_cfg4,
CSR_READ_1(sc, sc->rl_cfg4) & ~RL_CFG4_JUMBO_EN1);
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
}
}
CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
switch (sc->rl_hwrev->rl_rev) {
case RL_HWREV_8168DP:
pci_set_max_read_req(sc->rl_dev, 4096);
break;
default:
if (jumbo != 0)
pci_set_max_read_req(sc->rl_dev, 512);
else
pci_set_max_read_req(sc->rl_dev, 4096);
}
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
static void
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_init(void *xsc)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
struct rl_softc *sc = xsc;
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
RL_LOCK(sc);
re_init_locked(sc);
RL_UNLOCK(sc);
}
static void
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_init_locked(struct rl_softc *sc)
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
{
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
struct ifnet *ifp = sc->rl_ifp;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
struct mii_data *mii;
- Add support for 8110SCe part. Some magic registers were taken from Linux driver. - Swap hardware revisions for 8110S and 8169S as Linux driver claims. Reviewed by: yongari (early version)
2009-01-20 20:22:28 +00:00
uint32_t reg;
Always honor configured VLAN/checksum offload capabilities. Previously re(4) used to blindly enable VLAN hardware tag stripping and Rx checksum offload regardless of enabled optional features of interface.
2008-03-23 04:59:13 +00:00
uint16_t cfg;
Copy MAC address to fix alignment problems. Tested on alpha (AS600) with RTL8169S Suggested by: jhb
2006-02-20 10:29:40 +00:00
union {
uint32_t align_dummy;
u_char eaddr[ETHER_ADDR_LEN];
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
} eaddr;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
RL_LOCK_ASSERT(sc);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
mii = device_get_softc(sc->rl_miibus);
Minimize interface reinitialization by checking IFF_DRV_RUNNING flag. This fixes unnecessary interface UP/DOWNs during getting an IP address via DHCP. Tested by: Warren Block ( wblock<> wonkity dot com )
2009-11-19 23:12:44 +00:00
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
return;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/*
* Cancel pending I/O and free all RX/TX buffers.
*/
re_stop(sc);
Always put controller into known state before device intialization. While here remove re_reset calls invoked in system error case as controller reset is always done in device initialization.
2008-12-11 00:46:07 +00:00
/* Put controller into known state. */
re_reset(sc);
If driver is not able to allocate RX buffer, do not start driver. While I'm here move RX buffer allocation and descriptor initialization up to not touch hardware registers in case of RX buffer allocation failure.
2011-01-13 23:15:09 +00:00
/*
* For C+ mode, initialize the RX descriptors and mbufs.
*/
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
if (ifp->if_mtu > RL_MTU) {
if (re_jrx_list_init(sc) != 0) {
device_printf(sc->rl_dev,
"no memory for jumbo RX buffers\n");
re_stop(sc);
return;
}
/* Disable checksum offloading for jumbo frames. */
ifp->if_capenable &= ~(IFCAP_HWCSUM | IFCAP_TSO4);
ifp->if_hwassist &= ~(RE_CSUM_FEATURES | CSUM_TSO);
} else {
if (re_rx_list_init(sc) != 0) {
device_printf(sc->rl_dev,
"no memory for RX buffers\n");
re_stop(sc);
return;
}
}
re_set_jumbo(sc, ifp->if_mtu > RL_MTU);
} else {
if (re_rx_list_init(sc) != 0) {
device_printf(sc->rl_dev, "no memory for RX buffers\n");
re_stop(sc);
return;
}
if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
pci_get_device(sc->rl_dev) != RT_DEVICEID_8101E) {
if (ifp->if_mtu > RL_MTU)
pci_set_max_read_req(sc->rl_dev, 512);
else
pci_set_max_read_req(sc->rl_dev, 4096);
}
If driver is not able to allocate RX buffer, do not start driver. While I'm here move RX buffer allocation and descriptor initialization up to not touch hardware registers in case of RX buffer allocation failure.
2011-01-13 23:15:09 +00:00
}
re_tx_list_init(sc);
Teach the re(4) driver about the CFG2 register, which tells us whether we're on a 32-bit/64-bit bus or not. Use this to decide if we should set the PCI dual-address cycle enable bit in the C+ command register. (Enabling DAC on a 32-bit bus seems to do bad things.) Also, initialize the C+ command register early in the re_init() routine. The documentation says this register should be configured first.
2003-09-13 23:51:35 +00:00
/*
* Enable C+ RX and TX mode, as well as VLAN stripping and
Remove the dual-address cycle stuff. DAC is used to allow a bus master device to access 64-bit addresses from a 32-bit PCI bus. While the RealTek manual says you can set this bit and the chip will perform DAC only if you give it a DMA address with any of the upper 32 bits set, this appears not to be the case. If I turn on the DAC bit, the chip sets the 'system error' bit in the status register when I to do a DMA on my Athlon test box with 32-bit PCI bus (VIA chipset) even though I only have 128MB of physical memory, and thus can never give the chip a 64-bit address. Obviously, I can't just set it and forget it, so until I figure out the right rule for when it's safe/necessary to enable it, keep it turned off.
2003-09-20 21:18:27 +00:00
* RX checksum offload. We must configure the C+ register
Teach the re(4) driver about the CFG2 register, which tells us whether we're on a 32-bit/64-bit bus or not. Use this to decide if we should set the PCI dual-address cycle enable bit in the C+ command register. (Enabling DAC on a 32-bit bus seems to do bad things.) Also, initialize the C+ command register early in the re_init() routine. The documentation says this register should be configured first.
2003-09-13 23:51:35 +00:00
* before all others.
*/
Always honor configured VLAN/checksum offload capabilities. Previously re(4) used to blindly enable VLAN hardware tag stripping and Rx checksum offload regardless of enabled optional features of interface.
2008-03-23 04:59:13 +00:00
cfg = RL_CPLUSCMD_PCI_MRW;
if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
cfg |= RL_CPLUSCMD_RXCSUM_ENB;
if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
cfg |= RL_CPLUSCMD_VLANSTRIP;
Add basic support for RTL8168C, RTL8168CP, RTL8111C and RTL8111CP. ATM Tx/Rx checksum offload is supported but TSO and jumbo frame is not yet supported. Because these newer controllers use different descriptor formats, a flag RL_FLAG_DESCV2 flag was introduced to handle that case in Tx/Rx handler. Also newer controllers seems to require to not touch 'enable Tx/Rx bit' in RL_CPLUS_CMD register so don't blindly try to set that bits. Note, it seems that there is still power-saving related issue where driver fails to attach PHY. Rebooting seems to fix that issue but number of required reboots varys. Many thanks to users that helped during developement. I really appreciate their patient and test/feedbacks.
2008-07-02 07:54:53 +00:00
if ((sc->rl_flags & RL_FLAG_MACSTAT) != 0) {
cfg |= RL_CPLUSCMD_MACSTAT_DIS;
/* XXX magic. */
cfg |= 0x0001;
} else
cfg |= RL_CPLUSCMD_RXENB | RL_CPLUSCMD_TXENB;
CSR_WRITE_2(sc, RL_CPLUS_CMD, cfg);
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SC ||
sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE) {
- Add support for 8110SCe part. Some magic registers were taken from Linux driver. - Swap hardware revisions for 8110S and 8169S as Linux driver claims. Reviewed by: yongari (early version)
2009-01-20 20:22:28 +00:00
reg = 0x000fff00;
Use correct Config registers for RTL8139 family. Unlike RTL8168 and RTL810x family , RTL8139 has different register map for Config registers. While here, follow the lead of re(4) in WOL configuration. - Disable WOL_UCAST and WOL_MCAST capabilities by default. - Config5 register write does not need to unlock EEPROM access on RTL8139 family but unlocking EEPROM access does not affect its operation and make it consistent with re(4). Reported by: Matt Renzelmann mjr <> cs dot wisc dot edu
2012-02-25 04:54:51 +00:00
if ((CSR_READ_1(sc, sc->rl_cfg2) & RL_CFG2_PCI66MHZ) != 0)
- Add support for 8110SCe part. Some magic registers were taken from Linux driver. - Swap hardware revisions for 8110S and 8169S as Linux driver claims. Reviewed by: yongari (early version)
2009-01-20 20:22:28 +00:00
reg |= 0x000000ff;
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE)
- Add support for 8110SCe part. Some magic registers were taken from Linux driver. - Swap hardware revisions for 8110S and 8169S as Linux driver claims. Reviewed by: yongari (early version)
2009-01-20 20:22:28 +00:00
reg |= 0x00f00000;
CSR_WRITE_4(sc, 0x7c, reg);
/* Disable interrupt mitigation. */
CSR_WRITE_2(sc, 0xe2, 0);
}
The number of bits reserved for MSS in RealTek controllers is 11bits. This limits the maximum interface MTU size in TSO case as upper stack should not generate TCP segments with MSS greater than the limit. Armed with this information, disable TSO if interface MTU is greater than the limit.
2008-08-04 02:05:09 +00:00
/*
* Disable TSO if interface MTU size is greater than MSS
* allowed in controller.
*/
if (ifp->if_mtu > RL_TSO_MTU && (ifp->if_capenable & IFCAP_TSO4) != 0) {
ifp->if_capenable &= ~IFCAP_TSO4;
ifp->if_hwassist &= ~CSUM_TSO;
}
Teach the re(4) driver about the CFG2 register, which tells us whether we're on a 32-bit/64-bit bus or not. Use this to decide if we should set the PCI dual-address cycle enable bit in the C+ command register. (Enabling DAC on a 32-bit bus seems to do bad things.) Also, initialize the C+ command register early in the re_init() routine. The documentation says this register should be configured first.
2003-09-13 23:51:35 +00:00
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/*
* Init our MAC address. Even though the chipset
* documentation doesn't mention it, we need to enter "Config
* register write enable" mode to modify the ID registers.
*/
Copy MAC address to fix alignment problems. Tested on alpha (AS600) with RTL8169S Suggested by: jhb
2006-02-20 10:29:40 +00:00
/* Copy MAC address on stack to align. */
bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG);
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
CSR_WRITE_4(sc, RL_IDR0,
htole32(*(u_int32_t *)(&eaddr.eaddr[0])));
CSR_WRITE_4(sc, RL_IDR4,
htole32(*(u_int32_t *)(&eaddr.eaddr[4])));
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
It seems that enabling Tx and Rx before setting descriptor DMA addresses shall access invalid descriptor DMA addresses on PCIe hardwares and then panicked the system. To fix it set descriptor DMA addresses before enabling Tx and Rx such that hardware can see valid descriptor DMA addresses. Also set RL_EARLY_TX_THRESH before starting Tx and Rx. Reported by: steve.tell AT crashmail DOT de Tested by: steve.tell AT crashmail DOT de Obtained from: NetBSD MFC after: 1 week
2007-01-23 00:44:12 +00:00
/*
* Load the addresses of the RX and TX lists into the chip.
*/
CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI,
RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr));
CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO,
RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr));
CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI,
RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr));
CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO,
RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr));
- Implement the RX EARLYOFF and RXDV GATED bits as done by RealTek's Linux driver as version 8.037.00 for RTL8168{E-VL,EP,F,G,GU} and RTL8111B. This makes reception of packets work with the RTL8168G (HW rev. 0x4c000000) in my Shuttle DS47. - Consistently use RL_MSI_MESSAGES. In joint forces with: yongari MFC after: 5 days
2014-02-05 23:13:40 +00:00
if ((sc->rl_flags & RL_FLAG_RXDV_GATED) != 0)
CSR_WRITE_4(sc, RL_MISC, CSR_READ_4(sc, RL_MISC) &
~0x00080000);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/*
- Do not read and write RX configuration register multiple times. - Always program RX configuration register from scratch instead of doing read/modify/write. - Rename re_setmulti() to re_set_rxmode() to be reflect reality. - Simplify hash filter logic a little while I am here. Reviewed by: yongari (early version)
2009-01-20 19:58:01 +00:00
* Set the initial TX configuration.
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
*/
- For the 8169 chips, read the station address by forcing an EEPROM autoload and then copying the contends of the station address registers. For some reason, reading the EEPROM on the 8169S doesn't work right. This gets around the problem, and allows us to read the station address correctly on the 8169S. - Insert a delay after initiating packet transmition in re_diag() to allow lots of time for the frame to echo back to the host, and wait for both the 'RX complete' and 'timeout expired' bits in the ISR register to be set. - Deal more intelligently with the fact that the frame length field in the RX descriptor is a different width on the 8139C+ than it is on the 8169/8169S/8110S - For the 8169, you have to set bit 17 in the TX config register to enter digital loopback mode, but for the 8139C+, you have to set both bits 17 and 18. Take this into account so that re_diag() works properly for both types of chips.
2003-09-11 06:56:46 +00:00
if (sc->rl_testmode) {
if (sc->rl_type == RL_8169)
CSR_WRITE_4(sc, RL_TXCFG,
RL_TXCFG_CONFIG|RL_LOOPTEST_ON);
else
CSR_WRITE_4(sc, RL_TXCFG,
RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS);
} else
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG);
It seems that enabling Tx and Rx before setting descriptor DMA addresses shall access invalid descriptor DMA addresses on PCIe hardwares and then panicked the system. To fix it set descriptor DMA addresses before enabling Tx and Rx such that hardware can see valid descriptor DMA addresses. Also set RL_EARLY_TX_THRESH before starting Tx and Rx. Reported by: steve.tell AT crashmail DOT de Tested by: steve.tell AT crashmail DOT de Obtained from: NetBSD MFC after: 1 week
2007-01-23 00:44:12 +00:00
CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/*
- Do not read and write RX configuration register multiple times. - Always program RX configuration register from scratch instead of doing read/modify/write. - Rename re_setmulti() to re_set_rxmode() to be reflect reality. - Simplify hash filter logic a little while I am here. Reviewed by: yongari (early version)
2009-01-20 19:58:01 +00:00
* Set the initial RX configuration.
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
*/
- Do not read and write RX configuration register multiple times. - Always program RX configuration register from scratch instead of doing read/modify/write. - Rename re_setmulti() to re_set_rxmode() to be reflect reality. - Simplify hash filter logic a little while I am here. Reviewed by: yongari (early version)
2009-01-20 19:58:01 +00:00
re_set_rxmode(sc);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Follow the lead of vendor's interrupt moderation mechanism. It seems RTL8169/RTL8168/RTL810xE has a kind of interrupt moderation mechanism but it is not documented at all. The magic value dramatically reduced number of interrupts without noticeable performance drops so apply it to all RTL8169/RTL8169 controllers. Vendor's FreeBSD driver also applies it to RTL810xE controllers but their Linux driver explicitly cleared the register, so do not enable interrupt moderation for RTL810xE controllers. While I'm here sort 8169 specific registers. Obtained from: RealTek FreeBSD driver
2010-11-08 21:50:50 +00:00
/* Configure interrupt moderation. */
if (sc->rl_type == RL_8169) {
Apply TX interrupt moderation to all RTL810xE PCIe Fast Ethernet controllers. Experimentation with RTL8102E, RTL8103E and RTL8105E showed dramatic decrement of TX completion interrupts under high TX load(e.g. from 147k interrupts/second to 10k interrupts/second) With this change, TX interrupt moderation is applied to all controllers except RTL8139C+.
2011-01-24 00:01:06 +00:00
/* Magic from vendor. */
CSR_WRITE_2(sc, RL_INTRMOD, 0x5100);
Follow the lead of vendor's interrupt moderation mechanism. It seems RTL8169/RTL8168/RTL810xE has a kind of interrupt moderation mechanism but it is not documented at all. The magic value dramatically reduced number of interrupts without noticeable performance drops so apply it to all RTL8169/RTL8169 controllers. Vendor's FreeBSD driver also applies it to RTL810xE controllers but their Linux driver explicitly cleared the register, so do not enable interrupt moderation for RTL810xE controllers. While I'm here sort 8169 specific registers. Obtained from: RealTek FreeBSD driver
2010-11-08 21:50:50 +00:00
}
Don't enable RX and TX before their initial configuration is done, i. e. after setting up interrupt moderation but before turning interrupts on. This matches what Realtek's r8168 Linux driver does as of version 8.039.00 and fixes problems with certain incarnations of certain MAC revisions like the interface requiring an extra up/down-cycle after boot to start working or DMA configuration not being adhered to. PR: 193743, 197535 MFC after: 1 week
2015-04-09 21:35:44 +00:00
/*
* Enable transmit and receive.
*/
CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB | RL_CMD_RX_ENB);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
#ifdef DEVICE_POLLING
/*
* Disable interrupts if we are polling.
*/
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
if (ifp->if_capenable & IFCAP_POLLING)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
CSR_WRITE_2(sc, RL_IMR, 0);
else /* otherwise ... */
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
#endif
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/*
* Enable interrupts.
*/
if (sc->rl_testmode)
CSR_WRITE_2(sc, RL_IMR, 0);
else
CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/* Set initial TX threshold */
sc->rl_txthresh = RL_TX_THRESH_INIT;
/* Start RX/TX process. */
CSR_WRITE_4(sc, RL_MISSEDPKT, 0);
/*
* Initialize the timer interrupt register so that
* a timer interrupt will be generated once the timer
* reaches a certain number of ticks. The timer is
Do not use interrupt taskqueue on controllers with MSI/MSI-X capability. One of reason using interrupt taskqueue in re(4) was to reduce number of TX/RX interrupts under load because re(4) controllers have no good TX/RX interrupt moderation mechanism. Basic TX interrupt moderation is done by hardware for most controllers but RX interrupt moderation through undocumented register showed poor RX performance so it was disabled in r215025. Using taskqueue to handle RX interrupt greatly reduced number of interrupts but re(4) consumed all available CPU cycles to run the taskqueue under high TX/RX network load. This can happen even with RTL810x fast ethernet controller and I believe this is not acceptable for most systems. To mitigate the issue, use one-shot timer register to moderate RX interrupts. The timer register provides programmable one-shot timer and can be used to suppress interrupt generation. The timer runs at 125MHZ on PCIe controllers so the minimum time allowed for the timer is 8ns. Data sheet says the register is 32 bits but experimentation shows only lower 13 bits are valid so maximum time that can be programmed is 65.528us. This yields theoretical maximum number of RX interrupts that could be generated per second is about 15260. Combined with TX completion interrupts re(4) shall generate less than 20k interrupts. This number is still slightly high compared to other intelligent ethernet controllers but system is very responsive even under high network load. Introduce sysctl variable dev.re.%d.int_rx_mod that controls amount of time to delay RX interrupt processing in units of us. Value 0 completely disables RX interrupt moderation. To provide old behavior for controllers that have MSI/MSI-X capability, introduce a new tunable hw.re.intr_filter. If the tunable is set to non-zero value, driver will use interrupt taskqueue. The default value of the tunable is 0. This tunable has no effect on controllers that has no MSI/MSI-X capability or if MSI/MSI-X is explicitly disabled by administrator. While I'm here cleanup interrupt setup/teardown since re(4) uses single MSI/MSI-X message at this moment.
2011-01-26 20:25:40 +00:00
* reloaded on each transmit.
*/
#ifdef RE_TX_MODERATION
/*
* Use timer interrupt register to moderate TX interrupt
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
* moderation, which dramatically improves TX frame rate.
*/
if (sc->rl_type == RL_8169)
CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800);
else
CSR_WRITE_4(sc, RL_TIMERINT, 0x400);
Do not use interrupt taskqueue on controllers with MSI/MSI-X capability. One of reason using interrupt taskqueue in re(4) was to reduce number of TX/RX interrupts under load because re(4) controllers have no good TX/RX interrupt moderation mechanism. Basic TX interrupt moderation is done by hardware for most controllers but RX interrupt moderation through undocumented register showed poor RX performance so it was disabled in r215025. Using taskqueue to handle RX interrupt greatly reduced number of interrupts but re(4) consumed all available CPU cycles to run the taskqueue under high TX/RX network load. This can happen even with RTL810x fast ethernet controller and I believe this is not acceptable for most systems. To mitigate the issue, use one-shot timer register to moderate RX interrupts. The timer register provides programmable one-shot timer and can be used to suppress interrupt generation. The timer runs at 125MHZ on PCIe controllers so the minimum time allowed for the timer is 8ns. Data sheet says the register is 32 bits but experimentation shows only lower 13 bits are valid so maximum time that can be programmed is 65.528us. This yields theoretical maximum number of RX interrupts that could be generated per second is about 15260. Combined with TX completion interrupts re(4) shall generate less than 20k interrupts. This number is still slightly high compared to other intelligent ethernet controllers but system is very responsive even under high network load. Introduce sysctl variable dev.re.%d.int_rx_mod that controls amount of time to delay RX interrupt processing in units of us. Value 0 completely disables RX interrupt moderation. To provide old behavior for controllers that have MSI/MSI-X capability, introduce a new tunable hw.re.intr_filter. If the tunable is set to non-zero value, driver will use interrupt taskqueue. The default value of the tunable is 0. This tunable has no effect on controllers that has no MSI/MSI-X capability or if MSI/MSI-X is explicitly disabled by administrator. While I'm here cleanup interrupt setup/teardown since re(4) uses single MSI/MSI-X message at this moment.
2011-01-26 20:25:40 +00:00
#else
/*
* Use timer interrupt register to moderate RX interrupt
* moderation.
*/
if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 &&
intr_filter == 0) {
if (sc->rl_type == RL_8169)
CSR_WRITE_4(sc, RL_TIMERINT_8169,
RL_USECS(sc->rl_int_rx_mod));
} else {
if (sc->rl_type == RL_8169)
CSR_WRITE_4(sc, RL_TIMERINT_8169, RL_USECS(0));
}
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
#endif
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
/*
* For 8169 gigE NICs, set the max allowed RX packet
* size so we can receive jumbo frames.
*/
For re(4) controllers that uses new jumbo frame scheme(RTL8168C/D/E), limit maximum RX buffer size to RE_RX_DESC_BUFLEN instead of blindly configuring it to 16KB. Due to lack of documentation, re(4) didn't allow jumbo frame on these controllers. However it seems controller is confused with jumbo frame such that it can DMA the received frame to wrong address instead of splitting it into multiple RX buffers. Of course, this caused panic. Since re(4) does not support jumbo frames on these controllers, make controller drop frame that is longer than RE_RX_DESC_BUFLEN sized frame. Fortunately RTL810x controllers, which do not support jumbo frame, have no such issues but this change also limited maximum RX buffer size allowed to RTL810x controllers. Allowing 16KB RX buffer for controllers that have no such capability is meaningless. MFC after: 3 days
2011-01-12 03:43:47 +00:00
if (sc->rl_type == RL_8169) {
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
/*
* For controllers that use new jumbo frame scheme,
Correct comment typos.
2013-06-28 14:41:06 +00:00
* set maximum size of jumbo frame depending on
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
* controller revisions.
*/
if (ifp->if_mtu > RL_MTU)
CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
sc->rl_hwrev->rl_max_mtu +
ETHER_VLAN_ENCAP_LEN + ETHER_HDR_LEN +
ETHER_CRC_LEN);
else
CSR_WRITE_2(sc, RL_MAXRXPKTLEN,
RE_RX_DESC_BUFLEN);
} else if ((sc->rl_flags & RL_FLAG_PCIE) != 0 &&
sc->rl_hwrev->rl_max_mtu == RL_MTU) {
/* RTL810x has no jumbo frame support. */
For re(4) controllers that uses new jumbo frame scheme(RTL8168C/D/E), limit maximum RX buffer size to RE_RX_DESC_BUFLEN instead of blindly configuring it to 16KB. Due to lack of documentation, re(4) didn't allow jumbo frame on these controllers. However it seems controller is confused with jumbo frame such that it can DMA the received frame to wrong address instead of splitting it into multiple RX buffers. Of course, this caused panic. Since re(4) does not support jumbo frames on these controllers, make controller drop frame that is longer than RE_RX_DESC_BUFLEN sized frame. Fortunately RTL810x controllers, which do not support jumbo frame, have no such issues but this change also limited maximum RX buffer size allowed to RTL810x controllers. Allowing 16KB RX buffer for controllers that have no such capability is meaningless. MFC after: 3 days
2011-01-12 03:43:47 +00:00
CSR_WRITE_2(sc, RL_MAXRXPKTLEN, RE_RX_DESC_BUFLEN);
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
} else
For re(4) controllers that uses new jumbo frame scheme(RTL8168C/D/E), limit maximum RX buffer size to RE_RX_DESC_BUFLEN instead of blindly configuring it to 16KB. Due to lack of documentation, re(4) didn't allow jumbo frame on these controllers. However it seems controller is confused with jumbo frame such that it can DMA the received frame to wrong address instead of splitting it into multiple RX buffers. Of course, this caused panic. Since re(4) does not support jumbo frames on these controllers, make controller drop frame that is longer than RE_RX_DESC_BUFLEN sized frame. Fortunately RTL810x controllers, which do not support jumbo frame, have no such issues but this change also limited maximum RX buffer size allowed to RTL810x controllers. Allowing 16KB RX buffer for controllers that have no such capability is meaningless. MFC after: 3 days
2011-01-12 03:43:47 +00:00
CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383);
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
if (sc->rl_testmode)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
return;
Use correct Config registers for RTL8139 family. Unlike RTL8168 and RTL810x family , RTL8139 has different register map for Config registers. While here, follow the lead of re(4) in WOL configuration. - Disable WOL_UCAST and WOL_MCAST capabilities by default. - Config5 register write does not need to unlock EEPROM access on RTL8139 family but unlocking EEPROM access does not affect its operation and make it consistent with re(4). Reported by: Matt Renzelmann mjr <> cs dot wisc dot edu
2012-02-25 04:54:51 +00:00
CSR_WRITE_1(sc, sc->rl_cfg1, CSR_READ_1(sc, sc->rl_cfg1) |
RL_CFG1_DRVLOAD);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Propagate rename of IFF_OACTIVE and IFF_RUNNING to IFF_DRV_OACTIVE and IFF_DRV_RUNNING, as well as the move from ifnet.if_flags to ifnet.if_drv_flags. Device drivers are now responsible for synchronizing access to these flags, as they are in if_drv_flags. This helps prevent races between the network stack and device driver in maintaining the interface flags field. Many __FreeBSD__ and __FreeBSD_version checks maintained and continued; some less so. Reviewed by: pjd, bz MFC after: 7 days
2005-08-09 10:20:02 +00:00
ifp->if_drv_flags |= IFF_DRV_RUNNING;
ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Instead of allocating variables for each events/hardware flags, use a dedicated flag that represents controller capabilities/events. This will simplify many part of code that requires different workaround for each controller revisions and will enhance readability. While I'm here move PHY wakeup code up before mii_phy_probe() which seems to help to wake PHY in some cases.
2008-07-02 06:29:43 +00:00
sc->rl_flags &= ~RL_FLAG_LINK;
Perform media change after setting IFF_DRV_RUNNING flag. Without it, driver would ignore the first link state update if controller already established a link such that it would have to take additional link state handling in re_tick().
2011-11-22 23:27:59 +00:00
mii_mediachg(mii);
- Use the re_tick() callout instead of if_slowtimo() for driving re_watchdog() in order to avoid races accessing if_timer. - Use bus_get_dma_tag() so re(4) works on platforms requiring it. - Remove invalid BUS_DMA_ALLOCNOW when creating the parent DMA tag and the tags that are used for static memory allocations. - Don't bother to set if_mtu to ETHERMTU, ether_ifattach() does that. - Remove an unused variable in re_intr().
2006-12-20 02:13:59 +00:00
sc->rl_watchdog_timer = 0;
- Use if_printf() and device_printf() in re(4) and remove rl_unit from the softc. - Use callout_init_mtx() and rather than timeout/untimeout in both rl(4) and re(4). - Fix locking for ifmedia by locking the driver in the ifmedia handlers rather than in the miibus functions. (re(4) didn't lock the mii stuff at all!) - Fix some locking in re_ioctl(). Note: the two drivers share the same softc declared in if_rlreg.h, so they had to be change simultaneously. MFC after: 1 week Tested by: several on rl(4), none on re(4)
2005-09-29 16:52:19 +00:00
callout_reset(&sc->rl_stat_callout, hz, re_tick, sc);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
/*
* Set media options.
*/
static int
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_ifmedia_upd(struct ifnet *ifp)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
struct rl_softc *sc;
struct mii_data *mii;
Make sure to return the result of meida change request. Previously it used to return success regardless of the result.
2008-12-08 01:44:18 +00:00
int error;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
sc = ifp->if_softc;
mii = device_get_softc(sc->rl_miibus);
- Use if_printf() and device_printf() in re(4) and remove rl_unit from the softc. - Use callout_init_mtx() and rather than timeout/untimeout in both rl(4) and re(4). - Fix locking for ifmedia by locking the driver in the ifmedia handlers rather than in the miibus functions. (re(4) didn't lock the mii stuff at all!) - Fix some locking in re_ioctl(). Note: the two drivers share the same softc declared in if_rlreg.h, so they had to be change simultaneously. MFC after: 1 week Tested by: several on rl(4), none on re(4)
2005-09-29 16:52:19 +00:00
RL_LOCK(sc);
Make sure to return the result of meida change request. Previously it used to return success regardless of the result.
2008-12-08 01:44:18 +00:00
error = mii_mediachg(mii);
- Use if_printf() and device_printf() in re(4) and remove rl_unit from the softc. - Use callout_init_mtx() and rather than timeout/untimeout in both rl(4) and re(4). - Fix locking for ifmedia by locking the driver in the ifmedia handlers rather than in the miibus functions. (re(4) didn't lock the mii stuff at all!) - Fix some locking in re_ioctl(). Note: the two drivers share the same softc declared in if_rlreg.h, so they had to be change simultaneously. MFC after: 1 week Tested by: several on rl(4), none on re(4)
2005-09-29 16:52:19 +00:00
RL_UNLOCK(sc);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Make sure to return the result of meida change request. Previously it used to return success regardless of the result.
2008-12-08 01:44:18 +00:00
return (error);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
/*
* Report current media status.
*/
static void
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
struct rl_softc *sc;
struct mii_data *mii;
sc = ifp->if_softc;
mii = device_get_softc(sc->rl_miibus);
- Use if_printf() and device_printf() in re(4) and remove rl_unit from the softc. - Use callout_init_mtx() and rather than timeout/untimeout in both rl(4) and re(4). - Fix locking for ifmedia by locking the driver in the ifmedia handlers rather than in the miibus functions. (re(4) didn't lock the mii stuff at all!) - Fix some locking in re_ioctl(). Note: the two drivers share the same softc declared in if_rlreg.h, so they had to be change simultaneously. MFC after: 1 week Tested by: several on rl(4), none on re(4)
2005-09-29 16:52:19 +00:00
RL_LOCK(sc);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
mii_pollstat(mii);
ifmr->ifm_active = mii->mii_media_active;
ifmr->ifm_status = mii->mii_media_status;
Close a race where SIOCGIFMEDIA ioctl get inconsistent link status. Because driver is accessing a common MII structure in mii_pollstat(), updating user supplied structure should be done before dropping a driver lock. Reported by: Karim (fodillemlinkarimi <> gmail dot com)
2011-10-17 19:49:00 +00:00
RL_UNLOCK(sc);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
static int
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
struct rl_softc *sc = ifp->if_softc;
struct ifreq *ifr = (struct ifreq *) data;
struct mii_data *mii;
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
int error = 0;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Whitespace pass
2004-07-06 02:46:53 +00:00
switch (command) {
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
case SIOCSIFMTU:
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
if (ifr->ifr_mtu < ETHERMIN ||
Add preliminary support for RTL8168G, RTL8168GU and RTL8411B. RTL8168GU has two variants(GMII and MII) but it uses the same chip revision id. Driver checks PCI device id of controller and sets internal capability flag(i.e. jumbo frame and link speed down in WOL). H/W donated by: RealTek Semiconductor Corp.
2013-10-29 05:30:21 +00:00
ifr->ifr_mtu > sc->rl_hwrev->rl_max_mtu ||
((sc->rl_flags & RL_FLAG_FASTETHER) != 0 &&
ifr->ifr_mtu > RL_MTU)) {
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
error = EINVAL;
Don't allow jumbo frame on 8139C+ controller. While I'm here add a check for minimal MTU length.
2008-03-03 03:41:06 +00:00
break;
}
RL_LOCK(sc);
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
if (ifp->if_mtu != ifr->ifr_mtu) {
Don't allow jumbo frame on 8139C+ controller. While I'm here add a check for minimal MTU length.
2008-03-03 03:41:06 +00:00
ifp->if_mtu = ifr->ifr_mtu;
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
(ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
re_init_locked(sc);
}
if (ifp->if_mtu > RL_TSO_MTU &&
(ifp->if_capenable & IFCAP_TSO4) != 0) {
ifp->if_capenable &= ~(IFCAP_TSO4 |
IFCAP_VLAN_HWTSO);
ifp->if_hwassist &= ~CSUM_TSO;
}
Add TSO on VLANs. Because re(4) has a TSO limitation for jumbo frame, make sure to update VLAN capabilities whenever jumbo frame is configured. While I'm here rearrange interface capabilities configuration. The controller requires VLAN hardware tagging to make TSO work on VLANs so explicitly check this requirement.
2010-02-22 18:51:14 +00:00
VLAN_CAPABILITIES(ifp);
The number of bits reserved for MSS in RealTek controllers is 11bits. This limits the maximum interface MTU size in TSO case as upper stack should not generate TCP segments with MSS greater than the limit. Armed with this information, disable TSO if interface MTU is greater than the limit.
2008-08-04 02:05:09 +00:00
}
- Use if_printf() and device_printf() in re(4) and remove rl_unit from the softc. - Use callout_init_mtx() and rather than timeout/untimeout in both rl(4) and re(4). - Fix locking for ifmedia by locking the driver in the ifmedia handlers rather than in the miibus functions. (re(4) didn't lock the mii stuff at all!) - Fix some locking in re_ioctl(). Note: the two drivers share the same softc declared in if_rlreg.h, so they had to be change simultaneously. MFC after: 1 week Tested by: several on rl(4), none on re(4)
2005-09-29 16:52:19 +00:00
RL_UNLOCK(sc);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
break;
case SIOCSIFFLAGS:
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
RL_LOCK(sc);
Don't reinitialize the hardware if only PROMISC flag was changed. Previously whenever PROMISC mode turned on/off link renegotiation occurs and it could resulted in network unavailability for serveral seconds.(Depending on switch STP settings it could last several tens seconds.) Reported by: Prokofiev S.P. < proks AT logos DOT uptel DOT net > Tested by: Prokofiev S.P. < proks AT logos DOT uptel DOT net >
2007-04-18 00:40:43 +00:00
if ((ifp->if_flags & IFF_UP) != 0) {
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
if (((ifp->if_flags ^ sc->rl_if_flags)
Toggling IFF_ALLMULTI flag does not require full interface reintialization.
2008-07-02 06:34:02 +00:00
& (IFF_PROMISC | IFF_ALLMULTI)) != 0)
- Do not read and write RX configuration register multiple times. - Always program RX configuration register from scratch instead of doing read/modify/write. - Rename re_setmulti() to re_set_rxmode() to be reflect reality. - Simplify hash filter logic a little while I am here. Reviewed by: yongari (early version)
2009-01-20 19:58:01 +00:00
re_set_rxmode(sc);
Don't reinitialize the hardware if only PROMISC flag was changed. Previously whenever PROMISC mode turned on/off link renegotiation occurs and it could resulted in network unavailability for serveral seconds.(Depending on switch STP settings it could last several tens seconds.) Reported by: Prokofiev S.P. < proks AT logos DOT uptel DOT net > Tested by: Prokofiev S.P. < proks AT logos DOT uptel DOT net >
2007-04-18 00:40:43 +00:00
} else
re_init_locked(sc);
} else {
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
re_stop(sc);
}
sc->rl_if_flags = ifp->if_flags;
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
RL_UNLOCK(sc);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
break;
case SIOCADDMULTI:
case SIOCDELMULTI:
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
RL_LOCK(sc);
Minimize interface reinitialization by checking IFF_DRV_RUNNING flag. This fixes unnecessary interface UP/DOWNs during getting an IP address via DHCP. Tested by: Warren Block ( wblock<> wonkity dot com )
2009-11-19 23:12:44 +00:00
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
re_set_rxmode(sc);
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
RL_UNLOCK(sc);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
break;
case SIOCGIFMEDIA:
case SIOCSIFMEDIA:
mii = device_get_softc(sc->rl_miibus);
error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
break;
case SIOCSIFCAP:
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
{
Correctly switch IFCAP_VLAN_HWTAGGING on and off. Reported & tested by: Peter Blok <pblok bsd4all.org>
2006-03-03 12:32:51 +00:00
int mask, reinit;
mask = ifr->ifr_reqcap ^ ifp->if_capenable;
reinit = 0;
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
#ifdef DEVICE_POLLING
if (mask & IFCAP_POLLING) {
if (ifr->ifr_reqcap & IFCAP_POLLING) {
error = ether_poll_register(re_poll, ifp);
if (error)
style(9).
2010-11-05 18:24:50 +00:00
return (error);
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
RL_LOCK(sc);
/* Disable interrupts */
CSR_WRITE_2(sc, RL_IMR, 0x0000);
ifp->if_capenable |= IFCAP_POLLING;
RL_UNLOCK(sc);
} else {
error = ether_poll_deregister(ifp);
/* Enable interrupts. */
RL_LOCK(sc);
CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS);
ifp->if_capenable &= ~IFCAP_POLLING;
RL_UNLOCK(sc);
}
}
#endif /* DEVICE_POLLING */
Add missing driver lock in SIOCSIFCAP handler.
2011-11-16 22:09:14 +00:00
RL_LOCK(sc);
Allow TX/RX checksum offloading to be configured independently.
2011-01-13 22:49:10 +00:00
if ((mask & IFCAP_TXCSUM) != 0 &&
(ifp->if_capabilities & IFCAP_TXCSUM) != 0) {
ifp->if_capenable ^= IFCAP_TXCSUM;
Disable TX IP/TCP/UDP checksum offloading for RTL8168C/RTL8168CP. Previously only TX IP checksum offloading was disabled but it's reported that TX checksum offloading for UDP datagrams with IP options also generates corrupted frames. Reporter's controller is RTL8168CP but I guess RTL8168C also have the same issue since it shall share the same core. Reported and tested by: tuexen
2014-05-13 05:19:29 +00:00
if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
ifp->if_hwassist |= RE_CSUM_FEATURES;
else
Fix typo.
2006-11-21 05:41:11 +00:00
ifp->if_hwassist &= ~RE_CSUM_FEATURES;
Correctly switch IFCAP_VLAN_HWTAGGING on and off. Reported & tested by: Peter Blok <pblok bsd4all.org>
2006-03-03 12:32:51 +00:00
reinit = 1;
}
Allow TX/RX checksum offloading to be configured independently.
2011-01-13 22:49:10 +00:00
if ((mask & IFCAP_RXCSUM) != 0 &&
(ifp->if_capabilities & IFCAP_RXCSUM) != 0) {
ifp->if_capenable ^= IFCAP_RXCSUM;
reinit = 1;
}
Add TSO on VLANs. Because re(4) has a TSO limitation for jumbo frame, make sure to update VLAN capabilities whenever jumbo frame is configured. While I'm here rearrange interface capabilities configuration. The controller requires VLAN hardware tagging to make TSO work on VLANs so explicitly check this requirement.
2010-02-22 18:51:14 +00:00
if ((mask & IFCAP_TSO4) != 0 &&
Do not toggle IFCAP_TSO4 if we would also do TSO6. Given the driver does not currently announce/support TSO6 that cannot happen. Clean it up anyway for consistency. Reviewed by: yongari MFC after: 1 week
2012-04-24 08:30:55 +00:00
(ifp->if_capabilities & IFCAP_TSO4) != 0) {
Add TSO support. Tested by: wilko, Pieter de Goeje < pieter AT degoeje DOT nl >
2006-11-21 04:40:30 +00:00
ifp->if_capenable ^= IFCAP_TSO4;
Add TSO on VLANs. Because re(4) has a TSO limitation for jumbo frame, make sure to update VLAN capabilities whenever jumbo frame is configured. While I'm here rearrange interface capabilities configuration. The controller requires VLAN hardware tagging to make TSO work on VLANs so explicitly check this requirement.
2010-02-22 18:51:14 +00:00
if ((IFCAP_TSO4 & ifp->if_capenable) != 0)
Add TSO support. Tested by: wilko, Pieter de Goeje < pieter AT degoeje DOT nl >
2006-11-21 04:40:30 +00:00
ifp->if_hwassist |= CSUM_TSO;
else
ifp->if_hwassist &= ~CSUM_TSO;
The number of bits reserved for MSS in RealTek controllers is 11bits. This limits the maximum interface MTU size in TSO case as upper stack should not generate TCP segments with MSS greater than the limit. Armed with this information, disable TSO if interface MTU is greater than the limit.
2008-08-04 02:05:09 +00:00
if (ifp->if_mtu > RL_TSO_MTU &&
(ifp->if_capenable & IFCAP_TSO4) != 0) {
ifp->if_capenable &= ~IFCAP_TSO4;
ifp->if_hwassist &= ~CSUM_TSO;
}
Add TSO support. Tested by: wilko, Pieter de Goeje < pieter AT degoeje DOT nl >
2006-11-21 04:40:30 +00:00
}
Add TSO on VLANs. Because re(4) has a TSO limitation for jumbo frame, make sure to update VLAN capabilities whenever jumbo frame is configured. While I'm here rearrange interface capabilities configuration. The controller requires VLAN hardware tagging to make TSO work on VLANs so explicitly check this requirement.
2010-02-22 18:51:14 +00:00
if ((mask & IFCAP_VLAN_HWTSO) != 0 &&
(ifp->if_capabilities & IFCAP_VLAN_HWTSO) != 0)
ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
(ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) {
ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
/* TSO over VLAN requires VLAN hardware tagging. */
if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
ifp->if_capenable &= ~IFCAP_VLAN_HWTSO;
reinit = 1;
}
Implement initial jumbo frame support for RTL8168/8111 C/D/E PCIe GbE controllers. It seems these controllers no longer support multi-fragmented RX buffers such that driver have to allocate physically contiguous buffers. o Retire RL_FLAG_NOJUMBO flag and introduce RL_FLAG_JUMBOV2 to mark controllers that use new jumbo frame scheme. o Configure PCIe max read request size to 4096 for standard frames and reduce it to 512 for jumbo frames. o TSO/checksum offloading is not supported for jumbo frames on these controllers. Reflect it to ioctl handler and driver initialization. o Remove unused rl_stats_no_timeout in softc. o Embed a pointer to structure rl_hwrev into softc to keep track of controller MTU limitation and remove rl_hwrev in softc since that information is available through a pointer to structure rl_hwrev. Special thanks to Realtek for donating sample hardwares which made this possible. H/W donated by: Realtek Semiconductor Corp.
2011-01-17 03:24:33 +00:00
if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
(mask & (IFCAP_HWCSUM | IFCAP_TSO4 |
IFCAP_VLAN_HWTSO)) != 0)
reinit = 1;
Implement WOL. Tested by: Fabian Keil ( freebsd-listen AT fabienkeli DOT de )
2008-03-03 03:33:58 +00:00
if ((mask & IFCAP_WOL) != 0 &&
(ifp->if_capabilities & IFCAP_WOL) != 0) {
if ((mask & IFCAP_WOL_UCAST) != 0)
ifp->if_capenable ^= IFCAP_WOL_UCAST;
if ((mask & IFCAP_WOL_MCAST) != 0)
ifp->if_capenable ^= IFCAP_WOL_MCAST;
if ((mask & IFCAP_WOL_MAGIC) != 0)
ifp->if_capenable ^= IFCAP_WOL_MAGIC;
}
Minimize interface reinitialization by checking IFF_DRV_RUNNING flag. This fixes unnecessary interface UP/DOWNs during getting an IP address via DHCP. Tested by: Warren Block ( wblock<> wonkity dot com )
2009-11-19 23:12:44 +00:00
if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING) {
ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
Add missing driver lock in SIOCSIFCAP handler.
2011-11-16 22:09:14 +00:00
re_init_locked(sc);
Minimize interface reinitialization by checking IFF_DRV_RUNNING flag. This fixes unnecessary interface UP/DOWNs during getting an IP address via DHCP. Tested by: Warren Block ( wblock<> wonkity dot com )
2009-11-19 23:12:44 +00:00
}
Add missing driver lock in SIOCSIFCAP handler.
2011-11-16 22:09:14 +00:00
RL_UNLOCK(sc);
o Correctly set IFCAP_VLAN_HWCSUM as re(4) can do VLAN tagging/checksum offloading in hardware. o Correctly set media header length for VLAN.
2006-11-21 04:23:52 +00:00
VLAN_CAPABILITIES(ifp);
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
break;
default:
error = ether_ioctl(ifp, command, data);
break;
}
Whitespace pass
2004-07-06 02:46:53 +00:00
return (error);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
static void
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_watchdog(struct rl_softc *sc)
- Use the re_tick() callout instead of if_slowtimo() for driving re_watchdog() in order to avoid races accessing if_timer. - Use bus_get_dma_tag() so re(4) works on platforms requiring it. - Remove invalid BUS_DMA_ALLOCNOW when creating the parent DMA tag and the tags that are used for static memory allocations. - Don't bother to set if_mtu to ETHERMTU, ether_ifattach() does that. - Remove an unused variable in re_intr().
2006-12-20 02:13:59 +00:00
{
o Implemented miibus_statchg handler. It detects whether re(4) established a valid link or not. In miibus_statchg handler add a check for established link is valid one for the controller(e.g. 1000baseT is not a valid link for fastethernet controllers.) o Added a flag RE_FLAG_FASTETHER to mark fastethernet controllers. o Added additional check to know whether we've really encountered watchdog timeouts or missed Tx completion interrupts. This change may help to track down the cause of watchdog timeouts. o In interrupt handler, removed a check for link state change interrupt. Not all controllers have the bit and re(4) did not rely on the event for a long time. In addition, re(4) didn't request the interrupt in RL_IMR register. Tested by: rpaulo
2008-12-08 02:34:13 +00:00
struct ifnet *ifp;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
- Use the re_tick() callout instead of if_slowtimo() for driving re_watchdog() in order to avoid races accessing if_timer. - Use bus_get_dma_tag() so re(4) works on platforms requiring it. - Remove invalid BUS_DMA_ALLOCNOW when creating the parent DMA tag and the tags that are used for static memory allocations. - Don't bother to set if_mtu to ETHERMTU, ether_ifattach() does that. - Remove an unused variable in re_intr().
2006-12-20 02:13:59 +00:00
RL_LOCK_ASSERT(sc);
if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
return;
o Implemented miibus_statchg handler. It detects whether re(4) established a valid link or not. In miibus_statchg handler add a check for established link is valid one for the controller(e.g. 1000baseT is not a valid link for fastethernet controllers.) o Added a flag RE_FLAG_FASTETHER to mark fastethernet controllers. o Added additional check to know whether we've really encountered watchdog timeouts or missed Tx completion interrupts. This change may help to track down the cause of watchdog timeouts. o In interrupt handler, removed a check for link state change interrupt. Not all controllers have the bit and re(4) did not rely on the event for a long time. In addition, re(4) didn't request the interrupt in RL_IMR register. Tested by: rpaulo
2008-12-08 02:34:13 +00:00
ifp = sc->rl_ifp;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
re_txeof(sc);
o Implemented miibus_statchg handler. It detects whether re(4) established a valid link or not. In miibus_statchg handler add a check for established link is valid one for the controller(e.g. 1000baseT is not a valid link for fastethernet controllers.) o Added a flag RE_FLAG_FASTETHER to mark fastethernet controllers. o Added additional check to know whether we've really encountered watchdog timeouts or missed Tx completion interrupts. This change may help to track down the cause of watchdog timeouts. o In interrupt handler, removed a check for link state change interrupt. Not all controllers have the bit and re(4) did not rely on the event for a long time. In addition, re(4) didn't request the interrupt in RL_IMR register. Tested by: rpaulo
2008-12-08 02:34:13 +00:00
if (sc->rl_ldata.rl_tx_free == sc->rl_ldata.rl_tx_desc_cnt) {
if_printf(ifp, "watchdog timeout (missed Tx interrupts) "
"-- recovering\n");
if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
Remove TX taskqueue and directly invoke re_start in interrupt task.
2011-01-25 23:27:28 +00:00
re_start_locked(ifp);
o Implemented miibus_statchg handler. It detects whether re(4) established a valid link or not. In miibus_statchg handler add a check for established link is valid one for the controller(e.g. 1000baseT is not a valid link for fastethernet controllers.) o Added a flag RE_FLAG_FASTETHER to mark fastethernet controllers. o Added additional check to know whether we've really encountered watchdog timeouts or missed Tx completion interrupts. This change may help to track down the cause of watchdog timeouts. o In interrupt handler, removed a check for link state change interrupt. Not all controllers have the bit and re(4) did not rely on the event for a long time. In addition, re(4) didn't request the interrupt in RL_IMR register. Tested by: rpaulo
2008-12-08 02:34:13 +00:00
return;
}
if_printf(ifp, "watchdog timeout\n");
Mechanically convert to if_inc_counter().
2014-09-19 03:51:26 +00:00
if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
o Implemented miibus_statchg handler. It detects whether re(4) established a valid link or not. In miibus_statchg handler add a check for established link is valid one for the controller(e.g. 1000baseT is not a valid link for fastethernet controllers.) o Added a flag RE_FLAG_FASTETHER to mark fastethernet controllers. o Added additional check to know whether we've really encountered watchdog timeouts or missed Tx completion interrupts. This change may help to track down the cause of watchdog timeouts. o In interrupt handler, removed a check for link state change interrupt. Not all controllers have the bit and re(4) did not rely on the event for a long time. In addition, re(4) didn't request the interrupt in RL_IMR register. Tested by: rpaulo
2008-12-08 02:34:13 +00:00
When user_frac in the polling subsystem is low it is going to busy the CPU for too long period than necessary. Additively, interfaces are kept polled (in the tick) even if no more packets are available. In order to avoid such situations a new generic mechanism can be implemented in proactive way, keeping track of the time spent on any packet and fragmenting the time for any tick, stopping the processing as soon as possible. In order to implement such mechanism, the polling handler needs to change, returning the number of packets processed. While the intended logic is not part of this patch, the polling KPI is broken by this commit, adding an int return value and the new flag IFCAP_POLLING_NOCOUNT (which will signal that the return value is meaningless for the installed handler and checking should be skipped). Bump __FreeBSD_version in order to signal such situation. Reviewed by: emaste Sponsored by: Sandvine Incorporated
2009-05-30 15:14:44 +00:00
re_rxeof(sc, NULL);
Minimize interface reinitialization by checking IFF_DRV_RUNNING flag. This fixes unnecessary interface UP/DOWNs during getting an IP address via DHCP. Tested by: Warren Block ( wblock<> wonkity dot com )
2009-11-19 23:12:44 +00:00
ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
re_init_locked(sc);
o Implemented miibus_statchg handler. It detects whether re(4) established a valid link or not. In miibus_statchg handler add a check for established link is valid one for the controller(e.g. 1000baseT is not a valid link for fastethernet controllers.) o Added a flag RE_FLAG_FASTETHER to mark fastethernet controllers. o Added additional check to know whether we've really encountered watchdog timeouts or missed Tx completion interrupts. This change may help to track down the cause of watchdog timeouts. o In interrupt handler, removed a check for link state change interrupt. Not all controllers have the bit and re(4) did not rely on the event for a long time. In addition, re(4) didn't request the interrupt in RL_IMR register. Tested by: rpaulo
2008-12-08 02:34:13 +00:00
if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
Remove TX taskqueue and directly invoke re_start in interrupt task.
2011-01-25 23:27:28 +00:00
re_start_locked(ifp);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
/*
* Stop the adapter and free any mbufs allocated to the
* RX and TX lists.
*/
static void
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_stop(struct rl_softc *sc)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
Remove register keyword.
2008-08-04 03:47:29 +00:00
int i;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
struct ifnet *ifp;
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
struct rl_txdesc *txd;
struct rl_rxdesc *rxd;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
RL_LOCK_ASSERT(sc);
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
ifp = sc->rl_ifp;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
- Use the re_tick() callout instead of if_slowtimo() for driving re_watchdog() in order to avoid races accessing if_timer. - Use bus_get_dma_tag() so re(4) works on platforms requiring it. - Remove invalid BUS_DMA_ALLOCNOW when creating the parent DMA tag and the tags that are used for static memory allocations. - Don't bother to set if_mtu to ETHERMTU, ether_ifattach() does that. - Remove an unused variable in re_intr().
2006-12-20 02:13:59 +00:00
sc->rl_watchdog_timer = 0;
- Use if_printf() and device_printf() in re(4) and remove rl_unit from the softc. - Use callout_init_mtx() and rather than timeout/untimeout in both rl(4) and re(4). - Fix locking for ifmedia by locking the driver in the ifmedia handlers rather than in the miibus functions. (re(4) didn't lock the mii stuff at all!) - Fix some locking in re_ioctl(). Note: the two drivers share the same softc declared in if_rlreg.h, so they had to be change simultaneously. MFC after: 1 week Tested by: several on rl(4), none on re(4)
2005-09-29 16:52:19 +00:00
callout_stop(&sc->rl_stat_callout);
Propagate rename of IFF_OACTIVE and IFF_RUNNING to IFF_DRV_OACTIVE and IFF_DRV_RUNNING, as well as the move from ifnet.if_flags to ifnet.if_drv_flags. Device drivers are now responsible for synchronizing access to these flags, as they are in if_drv_flags. This helps prevent races between the network stack and device driver in maintaining the interface flags field. Many __FreeBSD__ and __FreeBSD_version checks maintained and continued; some less so. Reviewed by: pjd, bz MFC after: 7 days
2005-08-09 10:20:02 +00:00
ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Disable accepting frames in re_stop() to put RX MAC into idle state. Because there is no reliable way to know whether RX MAC is in stopped state, rejecting all frames would be the only way to minimize possible races. Otherwise it's possible to receive frames while stop command execution is in progress and controller can DMA the frame to freed RX buffer during that period. This was observed on recent PCIe controllers(i.e. RTL8111F). While this change may not be required on old controllers it wouldn't make negative effects on old controllers. One side effect of this change is disabling receive so driver reprograms RL_RXCFG to receive WOL frames when it is put into suspend or shutdown. This should address occasional 'memory modified free' errors seen on recent RealTek controllers.
2011-11-23 02:08:05 +00:00
/*
* Disable accepting frames to put RX MAC into idle state.
* Otherwise it's possible to get frames while stop command
* execution is in progress and controller can DMA the frame
* to already freed RX buffer during that period.
*/
CSR_WRITE_4(sc, RL_RXCFG, CSR_READ_4(sc, RL_RXCFG) &
~(RL_RXCFG_RX_ALLPHYS | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_MULTI |
RL_RXCFG_RX_BROAD));
Make sure to stop TX MAC before freeing queued TX frames. For RTL8111DP, check if the TX MAC is active by reading RL_GTXSTART register. For RTL8402/8168E-VL/8168F/8411, wait until TX queue is empty.
2011-11-23 22:07:13 +00:00
if ((sc->rl_flags & RL_FLAG_WAIT_TXPOLL) != 0) {
for (i = RL_TIMEOUT; i > 0; i--) {
if ((CSR_READ_1(sc, sc->rl_txstart) &
RL_TXSTART_START) == 0)
break;
DELAY(20);
}
if (i == 0)
device_printf(sc->rl_dev,
"stopping TX poll timed out!\n");
CSR_WRITE_1(sc, RL_COMMAND, 0x00);
} else if ((sc->rl_flags & RL_FLAG_CMDSTOP) != 0) {
Newer RealTek controllers requires setting stop request bit to terminate active Tx/Rx operation.
2008-12-11 01:26:18 +00:00
CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_STOPREQ | RL_CMD_TX_ENB |
RL_CMD_RX_ENB);
Make sure to stop TX MAC before freeing queued TX frames. For RTL8111DP, check if the TX MAC is active by reading RL_GTXSTART register. For RTL8402/8168E-VL/8168F/8411, wait until TX queue is empty.
2011-11-23 22:07:13 +00:00
if ((sc->rl_flags & RL_FLAG_CMDSTOP_WAIT_TXQ) != 0) {
for (i = RL_TIMEOUT; i > 0; i--) {
if ((CSR_READ_4(sc, RL_TXCFG) &
RL_TXCFG_QUEUE_EMPTY) != 0)
break;
DELAY(100);
}
if (i == 0)
device_printf(sc->rl_dev,
"stopping TXQ timed out!\n");
}
} else
Newer RealTek controllers requires setting stop request bit to terminate active Tx/Rx operation.
2008-12-11 01:26:18 +00:00
CSR_WRITE_1(sc, RL_COMMAND, 0x00);
DELAY(1000);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
CSR_WRITE_2(sc, RL_IMR, 0x0000);
Add support for the RealTek 8169SC/8110SC and RTL8101E devices. The latter is a PCIe 10/100 chip. Finally fix the EEPROM reading code so that we can access the EEPROMs on all devices. In order to access the EEPROM, we must select 'EEPROM programming' mode, and then set the EEPROM chip select bit. Previously, we were setting both bits simultaneously, which doesn't work: they must be set in the right sequence. Always obtain the station address from the EEPROM, now that EEPROM reading works correctly. Make the TX interrupt moderation code based on the internal timer optional and turned off by default. Make the re_diag() routine conditional and off by default. When it is on, only use it for the original 8169, which was the only device that that really needed it. Modify interrupt handling to use a fast interrupt handler and fast taskqeueue. Correct the rgephy driver so that it only applies the DSP fixup for PHY revs 0 and 1. Later chips are fixed and don't need the fixup. Make the rgephy driver advertise both 1000_FD and 1000_HD bits in autoneg mode. A couple of the devices don't autoneg correctly unless configured this way.
2006-06-26 20:31:32 +00:00
CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
if (sc->rl_head != NULL) {
m_freem(sc->rl_head);
sc->rl_head = sc->rl_tail = NULL;
}
/* Free the TX list buffers. */
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) {
txd = &sc->rl_ldata.rl_tx_desc[i];
if (txd->tx_m != NULL) {
bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag,
txd->tx_dmamap);
m_freem(txd->tx_m);
txd->tx_m = NULL;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
}
/* Free the RX list buffers. */
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
rxd = &sc->rl_ldata.rl_rx_desc[i];
if (rxd->rx_m != NULL) {
Use a RX DMA tag to free loaded RX DMA maps. Previously it used a TX DMA tag.
2012-01-17 19:31:03 +00:00
bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag,
Overhaul re(4). o Increased number of Rx/Tx descriptors to 256 for 8169 GigEs because it's hard to push the hardware to the limit with default 64 descriptors. TSO requires large number of Tx descriptors to pass a full sized TCP segment(65535 bytes IP packet) to hardware. Previously it consumed 32 Tx descriptors, assuming MCLBYTES DMA segment size, to send the TCP segment which means re(4) couldn't queue more than two full sized IP packets. For 8139C+ it still uses 64 Rx/Tx descriptors due to its hardware limitations. With this changes there are (very) small waste of memory for 8139C+ users but I don't think it would affect 8139C+ users for most cases. o Various bus_dma(9) fixes. - The hardware supports DAC so allow 64bit DMA operations. - Removed BUS_DMA_ALLOC_NOW flag. - Increased DMA segment size to 4096 from MCLBYTES because TSO consumes too many descriptors with MCLBYTES DMA segment size. - Tx/Rx side bus_dmamap_load_mbuf_sg(9) support. With these changes the code is more readable than previous one and got a (slightly) better performance as it doesn't need to pass/ decode arguments to/from callback function. - Removed unnecessary callback function re_dmamap_desc() and nuked rl_dmaload_arg structure which was used in the callback. - Additional protection for DMA map load failure. In case of failure reuse current map instead of returning a bogus DMA map. - Deferred DMA map unloading/sync operation for maximum performance until we really need to load new DMA map. If we happen to reuse current map(e.g. input error) there is no need to sync/unload/load again. - The number of allowable Tx DMA segments for a mbuf chains are now 32 instead of magic nseg value. If the number of available Tx descriptors are short enough to send highly fragmented mbuf chains an optimized re_defrag() is called to collapse mbuf chains which is supposed to be much faster than m_defrag(9). re_defrag() was borrowed from ath(4). - Separated Rx/Tx DMA tag from a common DMA tag such that Rx DMA tag correctly uses DMA maps that were created with DMA alignment restriction(8bytes alignments). Tx DMA tag does not have such alignment limitation. - Added additional sanity checks for DMA ring map load failure. - Added additional spare Rx DMA map for graceful handling of Rx DMA map load failure. - Fixed misused bus_dmamap_sync(9) and added missing bus_dmamap_sync(9) in re_encap()/re_txeof()/re_rxeof(). o Enabled TSO again as re(4) have reasonable number of Tx descriptors. o Don't touch DMA address of a Tx descriptor in re_txeof(). It's not needed. o Fix incorrect update of if_ierrors counter. For Rx buffer shortage it should update if_qdrops as the buffer is reused. o Added checks for unsupported H/W revisions and return ENXIO for these hardwares. This is required to remove resource allocation code in re_probe as other drivers do in device probe routine. o Modified descriptor index manipulation macros as it's now possible to have different number of descriptors for Rx/Tx. o In re_start, to save a lock operation, use IFQ_DRV_IS_EMPTY before trying to invoke IFQ_DRV_DEQUEUE. Also don't blindly call re_encap since we already know the number of available Tx descriptors in advance. o Removed RL_TX_DESC_THLD which was used to reserve RL_TX_DESC_THLD descriptors in Tx path. There is no such a limitation mentioned in 8139C+/8169/8110/8168/8101/8111 datasheet and it seems to work ok without reserving RL_TX_DESC_THLD descriptors. o Fix a comment for RL_GTXSTART. The register is 8bits register. o Added comments for 8169/8139C+ hardware restrictions on descriptors. o Removed forward declaration for "struct rl_softc", it's not needed. o Added a new structure rl_txdesc for Tx descriptor managements and a structure rl_rxdesc for Rx descriptor managements. o Removed unused member variable rl_intlock in driver softc. There are still several unused member variables which are supposed to be used to access hardware statistics counters. But it seems that accessing hardware counters were not implemented yet.
2008-01-15 01:10:31 +00:00
rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag,
rxd->rx_dmamap);
m_freem(rxd->rx_m);
rxd->rx_m = NULL;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
}
Free allocated jumbo buffers when controller is stopped.
2012-01-17 19:36:53 +00:00
if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) {
rxd = &sc->rl_ldata.rl_jrx_desc[i];
if (rxd->rx_m != NULL) {
bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag,
rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag,
rxd->rx_dmamap);
m_freem(rxd->rx_m);
rxd->rx_m = NULL;
}
}
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
/*
* Device suspend routine. Stop the interface and save some PCI
* settings in case the BIOS doesn't restore them properly on
* resume.
*/
static int
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_suspend(device_t dev)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
struct rl_softc *sc;
sc = device_get_softc(dev);
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
RL_LOCK(sc);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
re_stop(sc);
Implement WOL. Tested by: Fabian Keil ( freebsd-listen AT fabienkeli DOT de )
2008-03-03 03:33:58 +00:00
re_setwol(sc);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
sc->suspended = 1;
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
RL_UNLOCK(sc);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
return (0);
}
/*
* Device resume routine. Restore some PCI settings in case the BIOS
* doesn't, re-enable busmastering, and restart the interface if
* appropriate.
*/
static int
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_resume(device_t dev)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
struct rl_softc *sc;
struct ifnet *ifp;
sc = device_get_softc(dev);
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
RL_LOCK(sc);
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
ifp = sc->rl_ifp;
For RTL8168C SPIN2 controllers, make sure to take the controller out of sleep mode prior to accessing to PHY. This should fix device attach failure seen on these controllers. Also enable the sleep mode when device is put into sleep state. PR: kern/123123, kern/123053
2008-12-17 06:01:03 +00:00
/* Take controller out of sleep mode. */
if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
CSR_WRITE_1(sc, RL_GPIO,
CSR_READ_1(sc, RL_GPIO) | 0x01);
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
Implement WOL. Tested by: Fabian Keil ( freebsd-listen AT fabienkeli DOT de )
2008-03-03 03:33:58 +00:00
/*
* Clear WOL matching such that normal Rx filtering
* wouldn't interfere with WOL patterns.
*/
re_clrwol(sc);
Move interface reinitialization down after disabling WOL in resume path.
2009-11-19 23:14:40 +00:00
/* reinitialize interface if necessary */
if (ifp->if_flags & IFF_UP)
re_init_locked(sc);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
sc->suspended = 0;
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
RL_UNLOCK(sc);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
return (0);
}
/*
* Stop all chip I/O so that the kernel's probe routines don't
* get confused by errant DMAs when rebooting.
*/
Fix function prototype for device_shutdown method.
2007-11-22 02:45:00 +00:00
static int
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_shutdown(device_t dev)
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
{
struct rl_softc *sc;
sc = device_get_softc(dev);
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
RL_LOCK(sc);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
re_stop(sc);
In re_shutdown() mark interface as down since otherwise we will panic if interrupt comes in later on, which can happen in some uncommon cases. Another possible fix is to call re_detach() instead of re_stop(), like ve(4) does, but I am not sure if the latter is really RTTD, so that stick with this one-liner for now. PR: kern/80005 Approved by: silence on -arch, no reply from selected network gurus
2005-08-18 14:29:01 +00:00
/*
* Mark interface as down since otherwise we will panic if
* interrupt comes in later on, which can happen in some
Don't recommend re_detach() (like in vr(4)) as another possible method to prevent panicing in interrupt handler after re_shutdown(), sometimes seen on SMP systems. This would work here only because re_detach() clears IFF_UP (to prevent another race) and it was demonstrated that it's not enough to call vr_detach() in vr_shutdown() to prevent a panic.
2005-09-15 19:07:10 +00:00
* cases.
In re_shutdown() mark interface as down since otherwise we will panic if interrupt comes in later on, which can happen in some uncommon cases. Another possible fix is to call re_detach() instead of re_stop(), like ve(4) does, but I am not sure if the latter is really RTTD, so that stick with this one-liner for now. PR: kern/80005 Approved by: silence on -arch, no reply from selected network gurus
2005-08-18 14:29:01 +00:00
*/
sc->rl_ifp->if_flags &= ~IFF_UP;
Implement WOL. Tested by: Fabian Keil ( freebsd-listen AT fabienkeli DOT de )
2008-03-03 03:33:58 +00:00
re_setwol(sc);
merge in if_rl locking because if_re was originally based upon if_rl. This essentially merges revs 1.143-1.1445 of sys/pci/if_rl.c. This now marks the interrupt MPSAFE along with making the mutex non-recursive. Looked over by: bms
2004-09-03 16:41:41 +00:00
RL_UNLOCK(sc);
Fix function prototype for device_shutdown method.
2007-11-22 02:45:00 +00:00
return (0);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
}
Implement WOL. Tested by: Fabian Keil ( freebsd-listen AT fabienkeli DOT de )
2008-03-03 03:33:58 +00:00
To save more power, switch to 10/100Mbps link when controller is put into suspend/shutdown. Old PCI controllers performed that operation in firmware but for RTL8111C or newer controllers, it's responsibility of driver. It's not clear whether the firmware of RTL8111B still downgrades its speed to 10/100Mbps so leave it as it was.
2011-11-23 23:29:18 +00:00
static void
re_set_linkspeed(struct rl_softc *sc)
{
struct mii_softc *miisc;
struct mii_data *mii;
int aneg, i, phyno;
RL_LOCK_ASSERT(sc);
mii = device_get_softc(sc->rl_miibus);
mii_pollstat(mii);
aneg = 0;
if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
(IFM_ACTIVE | IFM_AVALID)) {
switch IFM_SUBTYPE(mii->mii_media_active) {
case IFM_10_T:
case IFM_100_TX:
return;
case IFM_1000_T:
aneg++;
break;
default:
break;
}
}
miisc = LIST_FIRST(&mii->mii_phys);
phyno = miisc->mii_phy;
LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
PHY_RESET(miisc);
re_miibus_writereg(sc->rl_dev, phyno, MII_100T2CR, 0);
re_miibus_writereg(sc->rl_dev, phyno,
MII_ANAR, ANAR_TX_FD | ANAR_TX | ANAR_10_FD | ANAR_10 | ANAR_CSMA);
re_miibus_writereg(sc->rl_dev, phyno,
MII_BMCR, BMCR_AUTOEN | BMCR_STARTNEG);
DELAY(1000);
if (aneg != 0) {
/*
* Poll link state until re(4) get a 10/100Mbps link.
*/
for (i = 0; i < MII_ANEGTICKS_GIGE; i++) {
mii_pollstat(mii);
if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID))
== (IFM_ACTIVE | IFM_AVALID)) {
switch (IFM_SUBTYPE(mii->mii_media_active)) {
case IFM_10_T:
case IFM_100_TX:
return;
default:
break;
}
}
RL_UNLOCK(sc);
pause("relnk", hz);
RL_LOCK(sc);
}
if (i == MII_ANEGTICKS_GIGE)
device_printf(sc->rl_dev,
"establishing a link failed, WOL may not work!");
}
/*
* No link, force MAC to have 100Mbps, full-duplex link.
* MAC does not require reprogramming on resolved speed/duplex,
* so this is just for completeness.
*/
mii->mii_media_status = IFM_AVALID | IFM_ACTIVE;
mii->mii_media_active = IFM_ETHER | IFM_100_TX | IFM_FDX;
}
Implement WOL. Tested by: Fabian Keil ( freebsd-listen AT fabienkeli DOT de )
2008-03-03 03:33:58 +00:00
static void
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_setwol(struct rl_softc *sc)
Implement WOL. Tested by: Fabian Keil ( freebsd-listen AT fabienkeli DOT de )
2008-03-03 03:33:58 +00:00
{
struct ifnet *ifp;
int pmc;
uint16_t pmstat;
uint8_t v;
RL_LOCK_ASSERT(sc);
Do a sweep of the tree replacing calls to pci_find_extcap() with calls to pci_find_cap() instead.
2011-03-23 13:10:15 +00:00
if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
Implement WOL. Tested by: Fabian Keil ( freebsd-listen AT fabienkeli DOT de )
2008-03-03 03:33:58 +00:00
return;
ifp = sc->rl_ifp;
For RTL8168C SPIN2 controllers, make sure to take the controller out of sleep mode prior to accessing to PHY. This should fix device attach failure seen on these controllers. Also enable the sleep mode when device is put into sleep state. PR: kern/123123, kern/123053
2008-12-17 06:01:03 +00:00
/* Put controller into sleep mode. */
if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) {
if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80)
CSR_WRITE_1(sc, RL_GPIO,
CSR_READ_1(sc, RL_GPIO) & ~0x01);
}
Disable accepting frames in re_stop() to put RX MAC into idle state. Because there is no reliable way to know whether RX MAC is in stopped state, rejecting all frames would be the only way to minimize possible races. Otherwise it's possible to receive frames while stop command execution is in progress and controller can DMA the frame to freed RX buffer during that period. This was observed on recent PCIe controllers(i.e. RTL8111F). While this change may not be required on old controllers it wouldn't make negative effects on old controllers. One side effect of this change is disabling receive so driver reprograms RL_RXCFG to receive WOL frames when it is put into suspend or shutdown. This should address occasional 'memory modified free' errors seen on recent RealTek controllers.
2011-11-23 02:08:05 +00:00
if ((ifp->if_capenable & IFCAP_WOL) != 0) {
re_set_rxmode(sc);
To save more power, switch to 10/100Mbps link when controller is put into suspend/shutdown. Old PCI controllers performed that operation in firmware but for RTL8111C or newer controllers, it's responsibility of driver. It's not clear whether the firmware of RTL8111B still downgrades its speed to 10/100Mbps so leave it as it was.
2011-11-23 23:29:18 +00:00
if ((sc->rl_flags & RL_FLAG_WOL_MANLINK) != 0)
re_set_linkspeed(sc);
Disable accepting frames in re_stop() to put RX MAC into idle state. Because there is no reliable way to know whether RX MAC is in stopped state, rejecting all frames would be the only way to minimize possible races. Otherwise it's possible to receive frames while stop command execution is in progress and controller can DMA the frame to freed RX buffer during that period. This was observed on recent PCIe controllers(i.e. RTL8111F). While this change may not be required on old controllers it wouldn't make negative effects on old controllers. One side effect of this change is disabling receive so driver reprograms RL_RXCFG to receive WOL frames when it is put into suspend or shutdown. This should address occasional 'memory modified free' errors seen on recent RealTek controllers.
2011-11-23 02:08:05 +00:00
if ((sc->rl_flags & RL_FLAG_WOLRXENB) != 0)
CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RX_ENB);
}
Implement WOL. Tested by: Fabian Keil ( freebsd-listen AT fabienkeli DOT de )
2008-03-03 03:33:58 +00:00
/* Enable config register write. */
CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
/* Enable PME. */
Use correct Config registers for RTL8139 family. Unlike RTL8168 and RTL810x family , RTL8139 has different register map for Config registers. While here, follow the lead of re(4) in WOL configuration. - Disable WOL_UCAST and WOL_MCAST capabilities by default. - Config5 register write does not need to unlock EEPROM access on RTL8139 family but unlocking EEPROM access does not affect its operation and make it consistent with re(4). Reported by: Matt Renzelmann mjr <> cs dot wisc dot edu
2012-02-25 04:54:51 +00:00
v = CSR_READ_1(sc, sc->rl_cfg1);
Implement WOL. Tested by: Fabian Keil ( freebsd-listen AT fabienkeli DOT de )
2008-03-03 03:33:58 +00:00
v &= ~RL_CFG1_PME;
if ((ifp->if_capenable & IFCAP_WOL) != 0)
v |= RL_CFG1_PME;
Use correct Config registers for RTL8139 family. Unlike RTL8168 and RTL810x family , RTL8139 has different register map for Config registers. While here, follow the lead of re(4) in WOL configuration. - Disable WOL_UCAST and WOL_MCAST capabilities by default. - Config5 register write does not need to unlock EEPROM access on RTL8139 family but unlocking EEPROM access does not affect its operation and make it consistent with re(4). Reported by: Matt Renzelmann mjr <> cs dot wisc dot edu
2012-02-25 04:54:51 +00:00
CSR_WRITE_1(sc, sc->rl_cfg1, v);
Implement WOL. Tested by: Fabian Keil ( freebsd-listen AT fabienkeli DOT de )
2008-03-03 03:33:58 +00:00
Use correct Config registers for RTL8139 family. Unlike RTL8168 and RTL810x family , RTL8139 has different register map for Config registers. While here, follow the lead of re(4) in WOL configuration. - Disable WOL_UCAST and WOL_MCAST capabilities by default. - Config5 register write does not need to unlock EEPROM access on RTL8139 family but unlocking EEPROM access does not affect its operation and make it consistent with re(4). Reported by: Matt Renzelmann mjr <> cs dot wisc dot edu
2012-02-25 04:54:51 +00:00
v = CSR_READ_1(sc, sc->rl_cfg3);
Implement WOL. Tested by: Fabian Keil ( freebsd-listen AT fabienkeli DOT de )
2008-03-03 03:33:58 +00:00
v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
v |= RL_CFG3_WOL_MAGIC;
Use correct Config registers for RTL8139 family. Unlike RTL8168 and RTL810x family , RTL8139 has different register map for Config registers. While here, follow the lead of re(4) in WOL configuration. - Disable WOL_UCAST and WOL_MCAST capabilities by default. - Config5 register write does not need to unlock EEPROM access on RTL8139 family but unlocking EEPROM access does not affect its operation and make it consistent with re(4). Reported by: Matt Renzelmann mjr <> cs dot wisc dot edu
2012-02-25 04:54:51 +00:00
CSR_WRITE_1(sc, sc->rl_cfg3, v);
Implement WOL. Tested by: Fabian Keil ( freebsd-listen AT fabienkeli DOT de )
2008-03-03 03:33:58 +00:00
Use correct Config registers for RTL8139 family. Unlike RTL8168 and RTL810x family , RTL8139 has different register map for Config registers. While here, follow the lead of re(4) in WOL configuration. - Disable WOL_UCAST and WOL_MCAST capabilities by default. - Config5 register write does not need to unlock EEPROM access on RTL8139 family but unlocking EEPROM access does not affect its operation and make it consistent with re(4). Reported by: Matt Renzelmann mjr <> cs dot wisc dot edu
2012-02-25 04:54:51 +00:00
v = CSR_READ_1(sc, sc->rl_cfg5);
Writing access to RL_CFG5 register also requires EEPROM write access. While I'm here, enable WOL through magic packet but disable waking up system via unicast, multicast and broadcast frames. Otherwise, multicast or unicast frame(e.g. ICMP echo request) can wake up system which is not probably wanted behavior on most environments. This was not known as problem because RL_CFG5 register access had not effect until this change. The capability to wake up system with unicast/multicast frames are still set in driver, default off, so users who need that feature can still activate it with ifconfig(8).
2011-11-22 23:19:49 +00:00
v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST |
RL_CFG5_WOL_LANWAKE);
Implement WOL. Tested by: Fabian Keil ( freebsd-listen AT fabienkeli DOT de )
2008-03-03 03:33:58 +00:00
if ((ifp->if_capenable & IFCAP_WOL_UCAST) != 0)
v |= RL_CFG5_WOL_UCAST;
if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
v |= RL_CFG5_WOL_MCAST | RL_CFG5_WOL_BCAST;
if ((ifp->if_capenable & IFCAP_WOL) != 0)
v |= RL_CFG5_WOL_LANWAKE;
Use correct Config registers for RTL8139 family. Unlike RTL8168 and RTL810x family , RTL8139 has different register map for Config registers. While here, follow the lead of re(4) in WOL configuration. - Disable WOL_UCAST and WOL_MCAST capabilities by default. - Config5 register write does not need to unlock EEPROM access on RTL8139 family but unlocking EEPROM access does not affect its operation and make it consistent with re(4). Reported by: Matt Renzelmann mjr <> cs dot wisc dot edu
2012-02-25 04:54:51 +00:00
CSR_WRITE_1(sc, sc->rl_cfg5, v);
Implement WOL. Tested by: Fabian Keil ( freebsd-listen AT fabienkeli DOT de )
2008-03-03 03:33:58 +00:00
Writing access to RL_CFG5 register also requires EEPROM write access. While I'm here, enable WOL through magic packet but disable waking up system via unicast, multicast and broadcast frames. Otherwise, multicast or unicast frame(e.g. ICMP echo request) can wake up system which is not probably wanted behavior on most environments. This was not known as problem because RL_CFG5 register access had not effect until this change. The capability to wake up system with unicast/multicast frames are still set in driver, default off, so users who need that feature can still activate it with ifconfig(8).
2011-11-22 23:19:49 +00:00
/* Config register write done. */
CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
Fix a logic error which resulted in putting PHY into sleep when WOL is active. If WOL is active driver should not put PHY into sleep. This change makes WOL work on RTL8168E.
2012-01-19 20:13:16 +00:00
if ((ifp->if_capenable & IFCAP_WOL) == 0 &&
Add preliminary support for 8168E/8111E PCIe controller. While I'm here simplify device description string. Tested by: Michael Beckmann < michael <> apfel dot de > MFC after: 5 days
2010-04-09 22:50:28 +00:00
(sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0)
CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) & ~0x80);
Implement WOL. Tested by: Fabian Keil ( freebsd-listen AT fabienkeli DOT de )
2008-03-03 03:33:58 +00:00
/*
* It seems that hardware resets its link speed to 100Mbps in
* power down mode so switching to 100Mbps in driver is not
* needed.
*/
/* Request PME if WOL is requested. */
pmstat = pci_read_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, 2);
pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
if ((ifp->if_capenable & IFCAP_WOL) != 0)
pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
pci_write_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
}
static void
Use ANSI C declarations for all functions.
2008-08-04 03:45:07 +00:00
re_clrwol(struct rl_softc *sc)
Implement WOL. Tested by: Fabian Keil ( freebsd-listen AT fabienkeli DOT de )
2008-03-03 03:33:58 +00:00
{
int pmc;
uint8_t v;
RL_LOCK_ASSERT(sc);
Do a sweep of the tree replacing calls to pci_find_extcap() with calls to pci_find_cap() instead.
2011-03-23 13:10:15 +00:00
if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
Implement WOL. Tested by: Fabian Keil ( freebsd-listen AT fabienkeli DOT de )
2008-03-03 03:33:58 +00:00
return;
/* Enable config register write. */
CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
Use correct Config registers for RTL8139 family. Unlike RTL8168 and RTL810x family , RTL8139 has different register map for Config registers. While here, follow the lead of re(4) in WOL configuration. - Disable WOL_UCAST and WOL_MCAST capabilities by default. - Config5 register write does not need to unlock EEPROM access on RTL8139 family but unlocking EEPROM access does not affect its operation and make it consistent with re(4). Reported by: Matt Renzelmann mjr <> cs dot wisc dot edu
2012-02-25 04:54:51 +00:00
v = CSR_READ_1(sc, sc->rl_cfg3);
Implement WOL. Tested by: Fabian Keil ( freebsd-listen AT fabienkeli DOT de )
2008-03-03 03:33:58 +00:00
v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
Use correct Config registers for RTL8139 family. Unlike RTL8168 and RTL810x family , RTL8139 has different register map for Config registers. While here, follow the lead of re(4) in WOL configuration. - Disable WOL_UCAST and WOL_MCAST capabilities by default. - Config5 register write does not need to unlock EEPROM access on RTL8139 family but unlocking EEPROM access does not affect its operation and make it consistent with re(4). Reported by: Matt Renzelmann mjr <> cs dot wisc dot edu
2012-02-25 04:54:51 +00:00
CSR_WRITE_1(sc, sc->rl_cfg3, v);
Implement WOL. Tested by: Fabian Keil ( freebsd-listen AT fabienkeli DOT de )
2008-03-03 03:33:58 +00:00
/* Config register write done. */
While accessing EEPROM command register use pre-defined constant instead of hardcoded value.
2008-07-02 05:01:19 +00:00
CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
Implement WOL. Tested by: Fabian Keil ( freebsd-listen AT fabienkeli DOT de )
2008-03-03 03:33:58 +00:00
Use correct Config registers for RTL8139 family. Unlike RTL8168 and RTL810x family , RTL8139 has different register map for Config registers. While here, follow the lead of re(4) in WOL configuration. - Disable WOL_UCAST and WOL_MCAST capabilities by default. - Config5 register write does not need to unlock EEPROM access on RTL8139 family but unlocking EEPROM access does not affect its operation and make it consistent with re(4). Reported by: Matt Renzelmann mjr <> cs dot wisc dot edu
2012-02-25 04:54:51 +00:00
v = CSR_READ_1(sc, sc->rl_cfg5);
Implement WOL. Tested by: Fabian Keil ( freebsd-listen AT fabienkeli DOT de )
2008-03-03 03:33:58 +00:00
v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
v &= ~RL_CFG5_WOL_LANWAKE;
Use correct Config registers for RTL8139 family. Unlike RTL8168 and RTL810x family , RTL8139 has different register map for Config registers. While here, follow the lead of re(4) in WOL configuration. - Disable WOL_UCAST and WOL_MCAST capabilities by default. - Config5 register write does not need to unlock EEPROM access on RTL8139 family but unlocking EEPROM access does not affect its operation and make it consistent with re(4). Reported by: Matt Renzelmann mjr <> cs dot wisc dot edu
2012-02-25 04:54:51 +00:00
CSR_WRITE_1(sc, sc->rl_cfg5, v);
Implement WOL. Tested by: Fabian Keil ( freebsd-listen AT fabienkeli DOT de )
2008-03-03 03:33:58 +00:00
}
Add simple MAC statistics counter reading support. Unfortunately useful counters like rl_missed_pkts is 16 bits quantity which is too small to hold meaningful information happened in a second. This means driver should frequently read these counters in order not to lose accuracy and that approach is too inefficient in driver's view. Moreover it seems there is no way to trigger an interrupt to detect counter near-full or wraparound event as well as lacking clearing the MAC counters. Another limitation of reading the counters from RealTek controllers is lack of interrupt firing at the end of DMA cycle of MAC counter read request such that driver have to poll the end of the DMA which is a time consuming process as well as inefficient. The more severe issue of the MAC counter read request is it takes too long to complete the DMA. All these limitation made maintaining MAC counters in driver impractical. For now, just provide simple sysctl interface to trigger reading the MAC counters. These counters could be used to track down driver issues. Users can read MAC counters maintained in controller with the following command. #sysctl dev.re.0.stats=1 While I'm here add check for validity of dma map and allocated memory before unloading/freeing them. Tested by: rmacklem
2010-11-05 19:28:00 +00:00
static void
re_add_sysctls(struct rl_softc *sc)
{
struct sysctl_ctx_list *ctx;
struct sysctl_oid_list *children;
Do not use interrupt taskqueue on controllers with MSI/MSI-X capability. One of reason using interrupt taskqueue in re(4) was to reduce number of TX/RX interrupts under load because re(4) controllers have no good TX/RX interrupt moderation mechanism. Basic TX interrupt moderation is done by hardware for most controllers but RX interrupt moderation through undocumented register showed poor RX performance so it was disabled in r215025. Using taskqueue to handle RX interrupt greatly reduced number of interrupts but re(4) consumed all available CPU cycles to run the taskqueue under high TX/RX network load. This can happen even with RTL810x fast ethernet controller and I believe this is not acceptable for most systems. To mitigate the issue, use one-shot timer register to moderate RX interrupts. The timer register provides programmable one-shot timer and can be used to suppress interrupt generation. The timer runs at 125MHZ on PCIe controllers so the minimum time allowed for the timer is 8ns. Data sheet says the register is 32 bits but experimentation shows only lower 13 bits are valid so maximum time that can be programmed is 65.528us. This yields theoretical maximum number of RX interrupts that could be generated per second is about 15260. Combined with TX completion interrupts re(4) shall generate less than 20k interrupts. This number is still slightly high compared to other intelligent ethernet controllers but system is very responsive even under high network load. Introduce sysctl variable dev.re.%d.int_rx_mod that controls amount of time to delay RX interrupt processing in units of us. Value 0 completely disables RX interrupt moderation. To provide old behavior for controllers that have MSI/MSI-X capability, introduce a new tunable hw.re.intr_filter. If the tunable is set to non-zero value, driver will use interrupt taskqueue. The default value of the tunable is 0. This tunable has no effect on controllers that has no MSI/MSI-X capability or if MSI/MSI-X is explicitly disabled by administrator. While I'm here cleanup interrupt setup/teardown since re(4) uses single MSI/MSI-X message at this moment.
2011-01-26 20:25:40 +00:00
int error;
Add simple MAC statistics counter reading support. Unfortunately useful counters like rl_missed_pkts is 16 bits quantity which is too small to hold meaningful information happened in a second. This means driver should frequently read these counters in order not to lose accuracy and that approach is too inefficient in driver's view. Moreover it seems there is no way to trigger an interrupt to detect counter near-full or wraparound event as well as lacking clearing the MAC counters. Another limitation of reading the counters from RealTek controllers is lack of interrupt firing at the end of DMA cycle of MAC counter read request such that driver have to poll the end of the DMA which is a time consuming process as well as inefficient. The more severe issue of the MAC counter read request is it takes too long to complete the DMA. All these limitation made maintaining MAC counters in driver impractical. For now, just provide simple sysctl interface to trigger reading the MAC counters. These counters could be used to track down driver issues. Users can read MAC counters maintained in controller with the following command. #sysctl dev.re.0.stats=1 While I'm here add check for validity of dma map and allocated memory before unloading/freeing them. Tested by: rmacklem
2010-11-05 19:28:00 +00:00
ctx = device_get_sysctl_ctx(sc->rl_dev);
children = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->rl_dev));
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "stats",
CTLTYPE_INT | CTLFLAG_RW, sc, 0, re_sysctl_stats, "I",
"Statistics Information");
Do not use interrupt taskqueue on controllers with MSI/MSI-X capability. One of reason using interrupt taskqueue in re(4) was to reduce number of TX/RX interrupts under load because re(4) controllers have no good TX/RX interrupt moderation mechanism. Basic TX interrupt moderation is done by hardware for most controllers but RX interrupt moderation through undocumented register showed poor RX performance so it was disabled in r215025. Using taskqueue to handle RX interrupt greatly reduced number of interrupts but re(4) consumed all available CPU cycles to run the taskqueue under high TX/RX network load. This can happen even with RTL810x fast ethernet controller and I believe this is not acceptable for most systems. To mitigate the issue, use one-shot timer register to moderate RX interrupts. The timer register provides programmable one-shot timer and can be used to suppress interrupt generation. The timer runs at 125MHZ on PCIe controllers so the minimum time allowed for the timer is 8ns. Data sheet says the register is 32 bits but experimentation shows only lower 13 bits are valid so maximum time that can be programmed is 65.528us. This yields theoretical maximum number of RX interrupts that could be generated per second is about 15260. Combined with TX completion interrupts re(4) shall generate less than 20k interrupts. This number is still slightly high compared to other intelligent ethernet controllers but system is very responsive even under high network load. Introduce sysctl variable dev.re.%d.int_rx_mod that controls amount of time to delay RX interrupt processing in units of us. Value 0 completely disables RX interrupt moderation. To provide old behavior for controllers that have MSI/MSI-X capability, introduce a new tunable hw.re.intr_filter. If the tunable is set to non-zero value, driver will use interrupt taskqueue. The default value of the tunable is 0. This tunable has no effect on controllers that has no MSI/MSI-X capability or if MSI/MSI-X is explicitly disabled by administrator. While I'm here cleanup interrupt setup/teardown since re(4) uses single MSI/MSI-X message at this moment.
2011-01-26 20:25:40 +00:00
if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0)
return;
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "int_rx_mod",
CTLTYPE_INT | CTLFLAG_RW, &sc->rl_int_rx_mod, 0,
sysctl_hw_re_int_mod, "I", "re RX interrupt moderation");
/* Pull in device tunables. */
sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
error = resource_int_value(device_get_name(sc->rl_dev),
device_get_unit(sc->rl_dev), "int_rx_mod", &sc->rl_int_rx_mod);
if (error == 0) {
if (sc->rl_int_rx_mod < RL_TIMER_MIN ||
sc->rl_int_rx_mod > RL_TIMER_MAX) {
device_printf(sc->rl_dev, "int_rx_mod value out of "
"range; using default: %d\n",
RL_TIMER_DEFAULT);
sc->rl_int_rx_mod = RL_TIMER_DEFAULT;
}
}
Add simple MAC statistics counter reading support. Unfortunately useful counters like rl_missed_pkts is 16 bits quantity which is too small to hold meaningful information happened in a second. This means driver should frequently read these counters in order not to lose accuracy and that approach is too inefficient in driver's view. Moreover it seems there is no way to trigger an interrupt to detect counter near-full or wraparound event as well as lacking clearing the MAC counters. Another limitation of reading the counters from RealTek controllers is lack of interrupt firing at the end of DMA cycle of MAC counter read request such that driver have to poll the end of the DMA which is a time consuming process as well as inefficient. The more severe issue of the MAC counter read request is it takes too long to complete the DMA. All these limitation made maintaining MAC counters in driver impractical. For now, just provide simple sysctl interface to trigger reading the MAC counters. These counters could be used to track down driver issues. Users can read MAC counters maintained in controller with the following command. #sysctl dev.re.0.stats=1 While I'm here add check for validity of dma map and allocated memory before unloading/freeing them. Tested by: rmacklem
2010-11-05 19:28:00 +00:00
}
static int
re_sysctl_stats(SYSCTL_HANDLER_ARGS)
{
struct rl_softc *sc;
struct rl_stats *stats;
int error, i, result;
result = -1;
error = sysctl_handle_int(oidp, &result, 0, req);
if (error || req->newptr == NULL)
return (error);
if (result == 1) {
sc = (struct rl_softc *)arg1;
RL_LOCK(sc);
When driver is not running, do not send DUMP command to controller and just show old (cached) values. Controller will not respond to the command unless MAC is enabled so DUMP request for down interface caused request timeout.
2011-01-10 23:47:11 +00:00
if ((sc->rl_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
RL_UNLOCK(sc);
goto done;
}
Add simple MAC statistics counter reading support. Unfortunately useful counters like rl_missed_pkts is 16 bits quantity which is too small to hold meaningful information happened in a second. This means driver should frequently read these counters in order not to lose accuracy and that approach is too inefficient in driver's view. Moreover it seems there is no way to trigger an interrupt to detect counter near-full or wraparound event as well as lacking clearing the MAC counters. Another limitation of reading the counters from RealTek controllers is lack of interrupt firing at the end of DMA cycle of MAC counter read request such that driver have to poll the end of the DMA which is a time consuming process as well as inefficient. The more severe issue of the MAC counter read request is it takes too long to complete the DMA. All these limitation made maintaining MAC counters in driver impractical. For now, just provide simple sysctl interface to trigger reading the MAC counters. These counters could be used to track down driver issues. Users can read MAC counters maintained in controller with the following command. #sysctl dev.re.0.stats=1 While I'm here add check for validity of dma map and allocated memory before unloading/freeing them. Tested by: rmacklem
2010-11-05 19:28:00 +00:00
bus_dmamap_sync(sc->rl_ldata.rl_stag,
sc->rl_ldata.rl_smap, BUS_DMASYNC_PREREAD);
CSR_WRITE_4(sc, RL_DUMPSTATS_HI,
RL_ADDR_HI(sc->rl_ldata.rl_stats_addr));
CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
RL_ADDR_LO(sc->rl_ldata.rl_stats_addr));
CSR_WRITE_4(sc, RL_DUMPSTATS_LO,
RL_ADDR_LO(sc->rl_ldata.rl_stats_addr |
RL_DUMPSTATS_START));
for (i = RL_TIMEOUT; i > 0; i--) {
if ((CSR_READ_4(sc, RL_DUMPSTATS_LO) &
RL_DUMPSTATS_START) == 0)
break;
DELAY(1000);
}
bus_dmamap_sync(sc->rl_ldata.rl_stag,
sc->rl_ldata.rl_smap, BUS_DMASYNC_POSTREAD);
RL_UNLOCK(sc);
if (i == 0) {
device_printf(sc->rl_dev,
Correct comment typos.
2013-06-28 14:41:06 +00:00
"DUMP statistics request timed out\n");
Add simple MAC statistics counter reading support. Unfortunately useful counters like rl_missed_pkts is 16 bits quantity which is too small to hold meaningful information happened in a second. This means driver should frequently read these counters in order not to lose accuracy and that approach is too inefficient in driver's view. Moreover it seems there is no way to trigger an interrupt to detect counter near-full or wraparound event as well as lacking clearing the MAC counters. Another limitation of reading the counters from RealTek controllers is lack of interrupt firing at the end of DMA cycle of MAC counter read request such that driver have to poll the end of the DMA which is a time consuming process as well as inefficient. The more severe issue of the MAC counter read request is it takes too long to complete the DMA. All these limitation made maintaining MAC counters in driver impractical. For now, just provide simple sysctl interface to trigger reading the MAC counters. These counters could be used to track down driver issues. Users can read MAC counters maintained in controller with the following command. #sysctl dev.re.0.stats=1 While I'm here add check for validity of dma map and allocated memory before unloading/freeing them. Tested by: rmacklem
2010-11-05 19:28:00 +00:00
return (ETIMEDOUT);
}
When driver is not running, do not send DUMP command to controller and just show old (cached) values. Controller will not respond to the command unless MAC is enabled so DUMP request for down interface caused request timeout.
2011-01-10 23:47:11 +00:00
done:
Add simple MAC statistics counter reading support. Unfortunately useful counters like rl_missed_pkts is 16 bits quantity which is too small to hold meaningful information happened in a second. This means driver should frequently read these counters in order not to lose accuracy and that approach is too inefficient in driver's view. Moreover it seems there is no way to trigger an interrupt to detect counter near-full or wraparound event as well as lacking clearing the MAC counters. Another limitation of reading the counters from RealTek controllers is lack of interrupt firing at the end of DMA cycle of MAC counter read request such that driver have to poll the end of the DMA which is a time consuming process as well as inefficient. The more severe issue of the MAC counter read request is it takes too long to complete the DMA. All these limitation made maintaining MAC counters in driver impractical. For now, just provide simple sysctl interface to trigger reading the MAC counters. These counters could be used to track down driver issues. Users can read MAC counters maintained in controller with the following command. #sysctl dev.re.0.stats=1 While I'm here add check for validity of dma map and allocated memory before unloading/freeing them. Tested by: rmacklem
2010-11-05 19:28:00 +00:00
stats = sc->rl_ldata.rl_stats;
printf("%s statistics:\n", device_get_nameunit(sc->rl_dev));
printf("Tx frames : %ju\n",
(uintmax_t)le64toh(stats->rl_tx_pkts));
printf("Rx frames : %ju\n",
(uintmax_t)le64toh(stats->rl_rx_pkts));
printf("Tx errors : %ju\n",
(uintmax_t)le64toh(stats->rl_tx_errs));
printf("Rx errors : %u\n",
le32toh(stats->rl_rx_errs));
printf("Rx missed frames : %u\n",
(uint32_t)le16toh(stats->rl_missed_pkts));
printf("Rx frame alignment errs : %u\n",
(uint32_t)le16toh(stats->rl_rx_framealign_errs));
printf("Tx single collisions : %u\n",
le32toh(stats->rl_tx_onecoll));
printf("Tx multiple collisions : %u\n",
le32toh(stats->rl_tx_multicolls));
printf("Rx unicast frames : %ju\n",
(uintmax_t)le64toh(stats->rl_rx_ucasts));
printf("Rx broadcast frames : %ju\n",
(uintmax_t)le64toh(stats->rl_rx_bcasts));
printf("Rx multicast frames : %u\n",
le32toh(stats->rl_rx_mcasts));
printf("Tx aborts : %u\n",
(uint32_t)le16toh(stats->rl_tx_aborts));
printf("Tx underruns : %u\n",
(uint32_t)le16toh(stats->rl_rx_underruns));
}
return (error);
}
Do not use interrupt taskqueue on controllers with MSI/MSI-X capability. One of reason using interrupt taskqueue in re(4) was to reduce number of TX/RX interrupts under load because re(4) controllers have no good TX/RX interrupt moderation mechanism. Basic TX interrupt moderation is done by hardware for most controllers but RX interrupt moderation through undocumented register showed poor RX performance so it was disabled in r215025. Using taskqueue to handle RX interrupt greatly reduced number of interrupts but re(4) consumed all available CPU cycles to run the taskqueue under high TX/RX network load. This can happen even with RTL810x fast ethernet controller and I believe this is not acceptable for most systems. To mitigate the issue, use one-shot timer register to moderate RX interrupts. The timer register provides programmable one-shot timer and can be used to suppress interrupt generation. The timer runs at 125MHZ on PCIe controllers so the minimum time allowed for the timer is 8ns. Data sheet says the register is 32 bits but experimentation shows only lower 13 bits are valid so maximum time that can be programmed is 65.528us. This yields theoretical maximum number of RX interrupts that could be generated per second is about 15260. Combined with TX completion interrupts re(4) shall generate less than 20k interrupts. This number is still slightly high compared to other intelligent ethernet controllers but system is very responsive even under high network load. Introduce sysctl variable dev.re.%d.int_rx_mod that controls amount of time to delay RX interrupt processing in units of us. Value 0 completely disables RX interrupt moderation. To provide old behavior for controllers that have MSI/MSI-X capability, introduce a new tunable hw.re.intr_filter. If the tunable is set to non-zero value, driver will use interrupt taskqueue. The default value of the tunable is 0. This tunable has no effect on controllers that has no MSI/MSI-X capability or if MSI/MSI-X is explicitly disabled by administrator. While I'm here cleanup interrupt setup/teardown since re(4) uses single MSI/MSI-X message at this moment.
2011-01-26 20:25:40 +00:00
static int
sysctl_int_range(SYSCTL_HANDLER_ARGS, int low, int high)
{
int error, value;
if (arg1 == NULL)
return (EINVAL);
value = *(int *)arg1;
error = sysctl_handle_int(oidp, &value, 0, req);
if (error || req->newptr == NULL)
return (error);
if (value < low || value > high)
return (EINVAL);
*(int *)arg1 = value;
return (0);
}
static int
sysctl_hw_re_int_mod(SYSCTL_HANDLER_ARGS)
{
return (sysctl_int_range(oidp, arg1, arg2, req, RL_TIMER_MIN,
RL_TIMER_MAX));
}
Reference in New Issue Copy Permalink