freebsd-dev/sys/dev/re/if_re.c

3791 lines
100 KiB
C
Raw Normal View History

/*-
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy 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.
*/
#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
/*
* 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
* 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
*
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
*
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.
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
* 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
*/
#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>
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>
#include <sys/lock.h>
#include <sys/mutex.h>
#include <sys/sysctl.h>
#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>
#include <net/if_arp.h>
#include <net/ethernet.h>
#include <net/if_dl.h>
#include <net/if_media.h>
#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>
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
#include <pci/if_rlreg.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
MODULE_DEPEND(re, pci, 1, 1, 1);
MODULE_DEPEND(re, ether, 1, 1, 1);
MODULE_DEPEND(re, miibus, 1, 1, 1);
/* "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"
/* 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);
static int msi_disable = 0;
TUNABLE_INT("hw.re.msi_disable", &msi_disable);
static int msix_disable = 0;
TUNABLE_INT("hw.re.msix_disable", &msix_disable);
static int prefer_iomap = 0;
TUNABLE_INT("hw.re.prefer_iomap", &prefer_iomap);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy 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.
*/
static struct rl_type re_devs[] = {
{ DLINK_VENDORID, DLINK_DEVICEID_528T, 0,
"D-Link DGE-528(T) Gigabit Ethernet Adapter" },
{ RT_VENDORID, RT_DEVICEID_8139, 0,
"RealTek 8139C+ 10/100BaseTX" },
{ RT_VENDORID, RT_DEVICEID_8101E, 0,
"RealTek 810xE PCIe 10/100baseTX" },
{ RT_VENDORID, RT_DEVICEID_8168, 0,
"RealTek 8168/8111 B/C/CP/D/DP/E PCIe Gigabit Ethernet" },
{ RT_VENDORID, RT_DEVICEID_8169, 0,
"RealTek 8169/8169S/8169SB(L)/8110S/8110SB(L) Gigabit Ethernet" },
{ 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
};
static struct rl_hwrev re_hwrevs[] = {
{ RL_HWREV_8139, RL_8139, "", RL_MTU },
{ 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 },
{ RL_HWREV_8168B_SPIN1, RL_8169, "8168", RL_JUMBO_MTU },
{ 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 },
{ RL_HWREV_8401E, RL_8169, "8401E", RL_MTU },
{ RL_HWREV_8105E, RL_8169, "8105E", RL_MTU },
{ RL_HWREV_8168B_SPIN2, RL_8169, "8168", RL_JUMBO_MTU },
{ RL_HWREV_8168B_SPIN3, RL_8169, "8168", RL_JUMBO_MTU },
{ 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},
{ 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);
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 *);
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 *);
#ifdef RE_FIXUP_RX
static __inline void re_fixup_rx
(struct mbuf *);
#endif
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 *);
#ifdef DEVICE_POLLING
static int re_poll (struct ifnet *, enum poll_cmd, int);
static int re_poll_locked (struct ifnet *, enum poll_cmd, int);
#endif
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 *);
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 *);
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 *);
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 *);
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);
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 *);
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);
static void re_set_jumbo (struct rl_softc *, int);
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 *);
static void re_setwol (struct rl_softc *);
static void re_clrwol (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
#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 *);
#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 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);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy 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),
/* bus interface */
DEVMETHOD(bus_print_child, bus_generic_print_child),
DEVMETHOD(bus_driver_added, bus_generic_driver_added),
/* MII interface */
DEVMETHOD(miibus_readreg, re_miibus_readreg),
DEVMETHOD(miibus_writereg, re_miibus_writereg),
DEVMETHOD(miibus_statchg, re_miibus_statchg),
{ 0, 0 }
};
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
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
{
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
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.
*/
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
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
{
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
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;
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++) {
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);
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));
*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
}
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
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);
/* Let the rgephy driver read the GMEDIASTAT register */
if (reg == RL_GMEDIASTAT) {
rval = CSR_READ_1(sc, RL_GMEDIASTAT);
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
CSR_WRITE_4(sc, RL_PHYAR, reg << 16);
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;
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
}
if (i == RL_PHY_TIMEOUT) {
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);
}
/*
* 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
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) |
(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
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;
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
}
if (i == RL_PHY_TIMEOUT) {
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);
}
/*
* 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
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);
}
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:
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);
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:
device_printf(sc->rl_dev, "bad phy register\n");
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);
if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) {
/* 8139C+ has different bit layout. */
rval &= ~(BMCR_LOOP | BMCR_ISO);
}
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
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);
}
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;
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:
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:
device_printf(sc->rl_dev, "bad phy register\n");
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);
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
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
{
struct rl_softc *sc;
struct ifnet *ifp;
struct mii_data *mii;
2004-07-06 02:48:29 +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
}
/*
* 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
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;
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
RL_LOCK_ASSERT(sc);
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
rxfilt = RL_RXCFG_CONFIG | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_BROAD;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy 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 (ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) {
if (ifp->if_flags & IFF_PROMISC)
rxfilt |= RL_RXCFG_RX_ALLPHYS;
/*
* Unlike other hardwares, we have to explicitly set
* RL_RXCFG_RX_MULTI to receive multicast frames in
* promiscuous mode.
*/
rxfilt |= RL_RXCFG_RX_MULTI;
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
}
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;
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));
}
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
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.
*/
if ((sc->rl_flags & RL_FLAG_PCIE) != 0) {
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;
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +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
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
{
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
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)
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
if ((sc->rl_flags & RL_FLAG_MACRESET) != 0)
CSR_WRITE_1(sc, 0x82, 1);
if (sc->rl_hwrev->rl_rev == RL_HWREV_8169S)
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
}
#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
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
{
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;
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 */
MGETHDR(m0, M_DONTWAIT, MT_DATA);
if (m0 == NULL)
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
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;
ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
re_init_locked(sc);
sc->rl_flags |= RL_FLAG_LINK;
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);
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;
/*
* 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
CSR_WRITE_2(sc, RL_ISR, 0xFFFF);
RL_UNLOCK(sc);
/* XXX: re_diag must not be called when in ALTQ mode */
IF_HANDOFF(&ifp->if_snd, m0, ifp);
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 */
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);
CSR_WRITE_2(sc, RL_ISR, status);
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) {
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) {
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) {
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);
device_printf(sc->rl_dev, "received RX 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
eh->ether_dhost, ":", eh->ether_shost, ":",
ntohs(eh->ether_type));
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;
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);
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);
}
#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
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
{
struct rl_type *t;
uint16_t devid, vendor;
uint16_t revid, sdevid;
int i;
2010-11-05 18:23:43 +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 attach to rev. 3 of the Linksys EG1032 adapter.
* Rev. 2 is supported by sk(4).
*/
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
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
}
}
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
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
{
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
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
{
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.
* 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
* 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
*/
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,
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
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 */
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,
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 */
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,
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 */
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
}
}
/* 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);
}
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
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];
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;
struct rl_hwrev *hw_rev;
int hwrev;
u_int16_t devid, re_did = 0;
int error = 0, i, phy, rid;
int msic, msixc, reg;
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);
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,
MTX_DEF);
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);
devid = pci_get_device(dev);
/*
* Prefer memory space register mapping over IO space.
* Because RTL8169SC does not seem to work when memory mapping
2010-11-05 18:23:43 +00:00
* is used always activate io mapping.
*/
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;
}
sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type,
&sc->rl_res_id, RF_ACTIVE);
if (sc->rl_res == NULL && prefer_iomap == 0) {
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);
}
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);
msic = pci_msi_count(dev);
msixc = pci_msix_count(dev);
if (pci_find_cap(dev, PCIY_EXPRESS, &reg) == 0)
sc->rl_flags |= RL_FLAG_PCIE;
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) {
msixc = 1;
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) {
if (msixc == 1) {
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;
}
}
/* Prefer MSI to INTx. */
if (msixc == 0 && msic > 0) {
msic = 1;
if (pci_alloc_msi(dev, &msic) == 0) {
if (msic == RL_MSI_MESSAGES) {
device_printf(dev, "Using %d MSI message\n",
msic);
sc->rl_flags |= RL_FLAG_MSI;
/* 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);
CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
} else
pci_release_msi(dev);
}
if ((sc->rl_flags & RL_FLAG_MSI) == 0)
msic = 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
/* Allocate interrupt */
if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0) {
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,
"couldn't llocate IRQ resources for "
"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 ((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);
}
hw_rev = re_hwrevs;
hwrev = CSR_READ_4(sc, RL_TXCFG);
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);
hwrev &= RL_TXCFG_HWREV;
break;
}
device_printf(dev, "MAC rev. 0x%08x\n", hwrev & 0x00700000);
while (hw_rev->rl_desc != NULL) {
if (hw_rev->rl_rev == hwrev) {
sc->rl_type = hw_rev->rl_type;
sc->rl_hwrev = hw_rev;
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) {
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
switch (hw_rev->rl_rev) {
case RL_HWREV_8139CPLUS:
sc->rl_flags |= RL_FLAG_FASTETHER | RL_FLAG_AUTOPAD;
break;
case RL_HWREV_8100E:
case RL_HWREV_8101E:
sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_FASTETHER;
break;
case RL_HWREV_8102E:
case RL_HWREV_8102EL:
case RL_HWREV_8102EL_SPIN1:
sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 |
RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP |
RL_FLAG_AUTOPAD;
break;
case RL_HWREV_8103E:
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;
break;
case RL_HWREV_8401E:
case RL_HWREV_8105E:
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;
case RL_HWREV_8168B_SPIN1:
case RL_HWREV_8168B_SPIN2:
sc->rl_flags |= RL_FLAG_WOLRXENB;
/* FALLTHROUGH */
case RL_HWREV_8168B_SPIN3:
sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_MACSTAT;
break;
case RL_HWREV_8168C_SPIN2:
sc->rl_flags |= RL_FLAG_MACSLEEP;
/* FALLTHROUGH */
case RL_HWREV_8168C:
if ((hwrev & 0x00700000) == 0x00200000)
sc->rl_flags |= RL_FLAG_MACSLEEP;
/* FALLTHROUGH */
case RL_HWREV_8168CP:
case RL_HWREV_8168D:
case RL_HWREV_8168DP:
sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2;
break;
case RL_HWREV_8168E:
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;
break;
case RL_HWREV_8168E_VL:
sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR |
RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP |
RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2;
break;
case RL_HWREV_8169_8110SB:
case RL_HWREV_8169_8110SBL:
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;
break;
default:
break;
}
/* Reset the adapter. */
RL_LOCK(sc);
re_reset(sc);
RL_UNLOCK(sc);
/* Enable PME. */
CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
cfg = CSR_READ_1(sc, RL_CFG1);
cfg |= RL_CFG1_PME;
CSR_WRITE_1(sc, RL_CFG1, cfg);
cfg = CSR_READ_1(sc, RL_CFG5);
cfg &= RL_CFG5_PME_STS;
CSR_WRITE_1(sc, RL_CFG5, cfg);
CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
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]);
bcopy(as, eaddr, sizeof(eaddr));
}
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. */
sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN;
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;
} 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. */
sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN;
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;
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy 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;
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
ifp = sc->rl_ifp = if_alloc(IFT_ETHER);
if (ifp == NULL) {
device_printf(dev, "can not if_alloc()\n");
error = ENOSPC;
goto fail;
}
/* 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);
}
/* Take PHY out of power down mode. */
if ((sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0) {
CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) | 0x80);
if (hw_rev->rl_rev == RL_HWREV_8401E)
CSR_WRITE_1(sc, 0xD1, CSR_READ_1(sc, 0xD1) & ~0x08);
}
if ((sc->rl_flags & RL_FLAG_PHYWAKE) != 0) {
re_gmii_writereg(dev, 1, 0x1f, 0);
re_gmii_writereg(dev, 1, 0x0e, 0);
}
#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");
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy driver to handle the on-board PHY in the 8169S/8110S (it needs special DSP initialization).
2003-09-08 02:11:25 +00:00
goto fail;
}
ifp->if_softc = sc;
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;
/*
* RTL8168/8111C generates wrong IP checksummed frame if the
* packet has IP options so disable TX IP checksum offloading.
*/
if (sc->rl_hwrev->rl_rev == RL_HWREV_8168C ||
sc->rl_hwrev->rl_rev == RL_HWREV_8168C_SPIN2)
ifp->if_hwassist = CSUM_TCP | CSUM_UDP;
else
ifp->if_hwassist = CSUM_IP | CSUM_TCP | CSUM_UDP;
ifp->if_hwassist |= CSUM_TSO;
ifp->if_capabilities = IFCAP_HWCSUM | IFCAP_TSO4;
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;
IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN);
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
TASK_INIT(&sc->rl_inttask, 0, re_int_task, 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
/*
* Call MI attach routine.
*/
ether_ifattach(ifp, eaddr);
/* VLAN capability setup */
ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING;
if (ifp->if_capabilities & IFCAP_HWCSUM)
ifp->if_capabilities |= IFCAP_VLAN_HWCSUM;
/* Enable WOL if PM is supported. */
if (pci_find_cap(sc->rl_dev, PCIY_PMG, &reg) == 0)
ifp->if_capabilities |= IFCAP_WOL;
ifp->if_capenable = ifp->if_capabilities;
/*
* Don't enable TSO by default. It is known to generate
* corrupted TCP segments(bad TCP options) under certain
* circumtances.
*/
ifp->if_hwassist &= ~CSUM_TSO;
ifp->if_capenable &= ~(IFCAP_TSO4 | IFCAP_VLAN_HWTSO);
#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.
*/
ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
#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
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
}
#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 {
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) {
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:
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy 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
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;
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);
ifp = sc->rl_ifp;
KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized"));
/* These should only be active if attach succeeded */
if (device_is_attached(dev)) {
#ifdef DEVICE_POLLING
if (ifp->if_capenable & IFCAP_POLLING)
ether_poll_deregister(ifp);
#endif
RL_LOCK(sc);
#if 0
sc->suspended = 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
re_stop(sc);
RL_UNLOCK(sc);
callout_drain(&sc->rl_stat_callout);
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
* 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;
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);
/*
* 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;
}
if (ifp != NULL)
if_free(ifp);
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;
}
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);
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)
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) {
if (sc->rl_ldata.rl_rx_list_map)
bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag,
sc->rl_ldata.rl_rx_list_map);
if (sc->rl_ldata.rl_rx_list_map && sc->rl_ldata.rl_rx_list)
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) {
if (sc->rl_ldata.rl_tx_list_map)
bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag,
sc->rl_ldata.rl_tx_list_map);
if (sc->rl_ldata.rl_tx_list_map && sc->rl_ldata.rl_tx_list)
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) {
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) {
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
}
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) {
if (sc->rl_ldata.rl_smap)
bus_dmamap_unload(sc->rl_ldata.rl_stag,
sc->rl_ldata.rl_smap);
if (sc->rl_ldata.rl_smap && sc->rl_ldata.rl_stats)
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);
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
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;
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
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
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 = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
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
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
/*
* 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
*/
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);
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
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;
m = m_getjcl(M_DONTWAIT, MT_DATA, M_PKTHDR, MJUM9BYTES);
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);
}
#ifdef RE_FIXUP_RX
static __inline void
re_fixup_rx(struct mbuf *m)
{
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
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;
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;
/* 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
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
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
}
/* 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
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
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;
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.
*/
static int
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;
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;
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
RL_LOCK_ASSERT(sc);
ifp = sc->rl_ifp;
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)) {
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);
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
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;
}
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;
/*
* 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.
*/
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) {
ifp->if_ierrors++;
/*
* 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.
*/
if (jumbo != 0)
rxerr = re_jumbo_newbuf(sc, i);
else
rxerr = re_newbuf(sc, i);
if (rxerr != 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
ifp->if_iqdrops++;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy 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) {
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;
}
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);
#ifdef RE_FIXUP_RX
re_fixup_rx(m);
#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
ifp->if_ipackets++;
m->m_pkthdr.rcvif = ifp;
/* Do RX checksumming if enabled */
if (ifp->if_capenable & IFCAP_RXCSUM) {
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
}
}
maxpkt--;
if (rxvlan & RL_RDESC_VLANCTL_TAG) {
m->m_pkthdr.ether_vtag =
bswap16((rxvlan & RL_RDESC_VLANCTL_DATA));
m->m_flags |= M_VLANTAG;
}
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);
RL_LOCK(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
}
/* 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;
if (rx_npktsp != NULL)
*rx_npktsp = rx_npkts;
if (maxpkt)
2010-11-05 18:24:50 +00:00
return (EAGAIN);
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
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;
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy 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))
ifp->if_collisions++;
if (txstat & RL_TDESC_STAT_TXERRSUM)
ifp->if_oerrors++;
else
ifp->if_opackets++;
}
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) {
#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.
*/
2004-07-06 02:46:53 +00:00
CSR_WRITE_4(sc, RL_TIMERCNT, 1);
#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
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;
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;
RL_LOCK_ASSERT(sc);
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);
if ((sc->rl_flags & RL_FLAG_LINK) == 0)
re_miibus_statchg(sc->rl_dev);
/*
* 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
2010-11-05 18:23:43 +00:00
* lost on PCIe based controllers under certain situations.
*/
re_txeof(sc);
re_watchdog(sc);
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
static int
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;
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)
rx_npkts = re_poll_locked(ifp, cmd, count);
RL_UNLOCK(sc);
return (rx_npkts);
}
static int
re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count)
{
struct rl_softc *sc = ifp->if_softc;
int rx_npkts;
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
sc->rxcycles = count;
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);
if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
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)
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);
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.
*/
if (status & RL_ISR_SYSTEM_ERR) {
ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
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
}
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 */
static int
re_intr(void *arg)
{
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)
return (FILTER_STRAY);
CSR_WRITE_2(sc, RL_IMR, 0);
taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask);
return (FILTER_HANDLED);
}
static void
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;
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;
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);
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) {
RL_UNLOCK(sc);
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
#ifdef DEVICE_POLLING
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
if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW))
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
/*
* 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 & (
#ifdef RE_TX_MODERATION
RL_ISR_TIMEOUT_EXPIRED|
#else
RL_ISR_TX_OK|
#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
if (status & RL_ISR_SYSTEM_ERR) {
ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
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
if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
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
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
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
}
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
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;
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);
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
* 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
*/
if ((sc->rl_flags & RL_FLAG_AUTOPAD) == 0 &&
(*m_head)->m_pkthdr.len < RL_IP4CSUMTX_PADLEN &&
((*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. */
m_new = m_dup(*m_head, M_DONTWAIT);
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) {
m_new = m_defrag(*m_head, M_DONTWAIT);
if (m_new == NULL) {
m_freem(*m_head);
*m_head = NULL;
return (ENOBUFS);
}
} else
m_new = *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
/*
* 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) {
m_new = m_collapse(*m_head, M_DONTWAIT, 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;
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
*/
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;
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 {
/*
* Unconditionally enable IP checksum if TCP or UDP
* checksum is required. Otherwise, TCP/UDP checksum
* does't make effects.
*/
if (((*m_head)->m_pkthdr.csum_flags & RE_CSUM_FEATURES) != 0) {
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;
}
}
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
/*
* 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)
vlanctl |= bswap16((*m_head)->m_pkthdr.ether_vtag) |
RL_TDESC_VLANCTL_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
si = prod;
for (i = 0; i < nsegs; i++, prod = RL_TX_DESC_NXT(sc, prod)) {
desc = &sc->rl_ldata.rl_tx_list[prod];
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
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
re_start(struct ifnet *ifp)
{
struct rl_softc *sc;
sc = ifp->if_softc;
RL_LOCK(sc);
re_start_locked(ifp);
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
/*
* Main transmit routine for C+ and gigE NICs.
*/
static void
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;
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)) !=
IFF_DRV_RUNNING || (sc->rl_flags & RL_FLAG_LINK) == 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
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;) {
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;
IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
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.
*/
ETHER_BPF_MTAP(ifp, m_head);
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
}
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)
CSR_WRITE_4(sc, RL_TIMERCNT, 1);
#endif
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
/* 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
#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);
#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.
*/
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
}
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) {
CSR_WRITE_1(sc, RL_CFG3, CSR_READ_1(sc, RL_CFG3) |
RL_CFG3_JUMBO_EN0);
switch (sc->rl_hwrev->rl_rev) {
case RL_HWREV_8168DP:
break;
case RL_HWREV_8168E:
CSR_WRITE_1(sc, RL_CFG4, CSR_READ_1(sc, RL_CFG4) |
0x01);
break;
default:
CSR_WRITE_1(sc, RL_CFG4, CSR_READ_1(sc, RL_CFG4) |
RL_CFG4_JUMBO_EN1);
}
} else {
CSR_WRITE_1(sc, RL_CFG3, CSR_READ_1(sc, RL_CFG3) &
~RL_CFG3_JUMBO_EN0);
switch (sc->rl_hwrev->rl_rev) {
case RL_HWREV_8168DP:
break;
case RL_HWREV_8168E:
CSR_WRITE_1(sc, RL_CFG4, CSR_READ_1(sc, RL_CFG4) &
~0x01);
break;
default:
CSR_WRITE_1(sc, RL_CFG4, CSR_READ_1(sc, RL_CFG4) &
~RL_CFG4_JUMBO_EN1);
}
}
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
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;
RL_LOCK(sc);
re_init_locked(sc);
RL_UNLOCK(sc);
}
static void
re_init_locked(struct rl_softc *sc)
{
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;
uint32_t reg;
uint16_t cfg;
union {
uint32_t align_dummy;
u_char eaddr[ETHER_ADDR_LEN];
} 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
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);
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);
/* Put controller into known state. */
re_reset(sc);
/*
* For C+ mode, initialize the RX descriptors and mbufs.
*/
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);
}
}
re_tx_list_init(sc);
/*
* Enable C+ RX and TX mode, as well as VLAN stripping and
* RX checksum offload. We must configure the C+ register
* before all others.
*/
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;
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);
if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SC ||
sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE) {
reg = 0x000fff00;
if ((CSR_READ_1(sc, RL_CFG2) & RL_CFG2_PCI66MHZ) != 0)
reg |= 0x000000ff;
if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE)
reg |= 0x00f00000;
CSR_WRITE_4(sc, 0x7c, reg);
/* Disable interrupt mitigation. */
CSR_WRITE_2(sc, 0xe2, 0);
}
/*
* 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;
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy 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 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);
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);
/*
* 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));
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy 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 transmit and receive.
*/
CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
/*
* 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
*/
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);
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
/*
* 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
*/
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
/* Configure interrupt moderation. */
if (sc->rl_type == RL_8169) {
/* Magic from vendor. */
CSR_WRITE_2(sc, RL_INTRMOD, 0x5100);
}
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy 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
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy 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);
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);
#ifdef notdef
/* Enable receiver and transmitter. */
CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB);
#endif
/*
* 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));
}
#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.
*/
if (sc->rl_type == RL_8169) {
if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) {
/*
* For controllers that use new jumbo frame scheme,
* set maximum size of jumbo frame depedning on
* 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. */
CSR_WRITE_2(sc, RL_MAXRXPKTLEN, RE_RX_DESC_BUFLEN);
} else
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
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;
mii_mediachg(mii);
CSR_WRITE_1(sc, RL_CFG1, CSR_READ_1(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
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
sc->rl_flags &= ~RL_FLAG_LINK;
sc->rl_watchdog_timer = 0;
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
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;
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);
RL_LOCK(sc);
error = mii_mediachg(mii);
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);
Take the support for the 8139C+/8169/8169S/8110S chips out of the rl(4) driver and put it in a new re(4) driver. The re(4) driver shares the if_rlreg.h file with rl(4) but is a separate module. (Ultimately I may change this. For now, it's convenient.) rl(4) has been modified so that it will never attach to an 8139C+ chip, leaving it to re(4) instead. Only re(4) has the PCI IDs to match the 8169/8169S/8110S gigE chips. if_re.c contains the same basic code that was originally bolted onto if_rl.c, with the following updates: - Added support for jumbo frames. Currently, there seems to be a limit of approximately 6200 bytes for jumbo frames on transmit. (This was determined via experimentation.) The 8169S/8110S chips apparently are limited to 7.5K frames on transmit. This may require some more work, though the framework to handle jumbo frames on RX is in place: the re_rxeof() routine will gather up frames than span multiple 2K clusters into a single mbuf list. - Fixed bug in re_txeof(): if we reap some of the TX buffers, but there are still some pending, re-arm the timer before exiting re_txeof() so that another timeout interrupt will be generated, just in case re_start() doesn't do it for us. - Handle the 'link state changed' interrupt - Fix a detach bug. If re(4) is loaded as a module, and you do tcpdump -i re0, then you do 'kldunload if_re,' the system will panic after a few seconds. This happens because ether_ifdetach() ends up calling the BPF detach code, which notices the interface is in promiscuous mode and tries to switch promisc mode off while detaching the BPF listner. This ultimately results in a call to re_ioctl() (due to SIOCSIFFLAGS), which in turn calls re_init() to handle the IFF_PROMISC flag change. Unfortunately, calling re_init() here turns the chip back on and restarts the 1-second timeout loop that drives re_tick(). By the time the timeout fires, if_re.ko has been unloaded, which results in a call to invalid code and blows up the system. To fix this, I cleared the IFF_UP flag before calling ether_ifdetach(), which stops the ioctl routine from trying to reset the chip. - Modified comments in re_rxeof() relating to the difference in RX descriptor status bit layout between the 8139C+ and the gigE chips. The layout is different because the frame length field was expanded from 12 bits to 13, and they got rid of one of the status bits to make room. - Add diagnostic code (re_diag()) to test for the case where a user has installed a broken 32-bit 8169 PCI NIC in a 64-bit slot. Some NICs have the REQ64# and ACK64# lines connected even though the board is 32-bit only (in this case, they should be pulled high). This fools the chip into doing 64-bit DMA transfers even though there is no 64-bit data path. To detect this, re_diag() puts the chip into digital loopback mode and sets the receiver to promiscuous mode, then initiates a single 64-byte packet transmission. The frame is echoed back to the host, and if the frame contents are intact, we know DMA is working correctly, otherwise we complain loudly on the console and abort the device attach. (At the moment, I don't know of any way to work around the problem other than physically modifying the board, so until/unless I can think of a software workaround, this will have do to.) - Created re(4) man page - Modified rlphy.c to allow re(4) to attach as well as rl(4). Note that this code works for the sample 8169/Marvell 88E1000 NIC that I have, but probably won't work for the 8169S/8110S chips. RealTek has sent me some sample NICs, but they haven't arrived yet. I will probably need to add an rlgphy 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
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);
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);
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
ifmr->ifm_active = mii->mii_media_active;
ifmr->ifm_status = mii->mii_media_status;
}
static int
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;
uint32_t rev;
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
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:
if (ifr->ifr_mtu < ETHERMIN ||
ifr->ifr_mtu > sc->rl_hwrev->rl_max_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;
break;
}
RL_LOCK(sc);
if (ifp->if_mtu != ifr->ifr_mtu) {
ifp->if_mtu = ifr->ifr_mtu;
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;
}
VLAN_CAPABILITIES(ifp);
}
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:
RL_LOCK(sc);
if ((ifp->if_flags & IFF_UP) != 0) {
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
if (((ifp->if_flags ^ sc->rl_if_flags)
& (IFF_PROMISC | IFF_ALLMULTI)) != 0)
re_set_rxmode(sc);
} 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;
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:
RL_LOCK(sc);
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
re_set_rxmode(sc);
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
{
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)
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 */
if ((mask & IFCAP_TXCSUM) != 0 &&
(ifp->if_capabilities & IFCAP_TXCSUM) != 0) {
ifp->if_capenable ^= IFCAP_TXCSUM;
if ((ifp->if_capenable & IFCAP_TXCSUM) != 0) {
rev = sc->rl_hwrev->rl_rev;
if (rev == RL_HWREV_8168C ||
rev == RL_HWREV_8168C_SPIN2)
ifp->if_hwassist |= CSUM_TCP | CSUM_UDP;
else
ifp->if_hwassist |= RE_CSUM_FEATURES;
} else
2006-11-21 05:41:11 +00:00
ifp->if_hwassist &= ~RE_CSUM_FEATURES;
reinit = 1;
}
if ((mask & IFCAP_RXCSUM) != 0 &&
(ifp->if_capabilities & IFCAP_RXCSUM) != 0) {
ifp->if_capenable ^= IFCAP_RXCSUM;
reinit = 1;
}
if ((mask & IFCAP_TSO4) != 0 &&
(ifp->if_capabilities & IFCAP_TSO) != 0) {
ifp->if_capenable ^= IFCAP_TSO4;
if ((IFCAP_TSO4 & ifp->if_capenable) != 0)
ifp->if_hwassist |= CSUM_TSO;
else
ifp->if_hwassist &= ~CSUM_TSO;
if (ifp->if_mtu > RL_TSO_MTU &&
(ifp->if_capenable & IFCAP_TSO4) != 0) {
ifp->if_capenable &= ~IFCAP_TSO4;
ifp->if_hwassist &= ~CSUM_TSO;
}
}
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;
}
if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 &&
(mask & (IFCAP_HWCSUM | IFCAP_TSO4 |
IFCAP_VLAN_HWTSO)) != 0)
reinit = 1;
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;
}
if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING) {
ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
re_init(sc);
}
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;
}
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
re_watchdog(struct rl_softc *sc)
{
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
RL_LOCK_ASSERT(sc);
if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0)
return;
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);
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))
re_start_locked(ifp);
return;
}
if_printf(ifp, "watchdog timeout\n");
ifp->if_oerrors++;
re_rxeof(sc, NULL);
ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
re_init_locked(sc);
if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
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
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
{
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
RL_LOCK_ASSERT(sc);
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
sc->rl_watchdog_timer = 0;
callout_stop(&sc->rl_stat_callout);
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
if ((sc->rl_flags & RL_FLAG_CMDSTOP) != 0)
CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_STOPREQ | RL_CMD_TX_ENB |
RL_CMD_RX_ENB);
else
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);
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) {
bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag,
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
}
}
}
/*
* Device suspend routine. Stop the interface and save some PCI
* settings in case the BIOS doesn't restore them properly on
* resume.
*/
static int
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);
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);
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;
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
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);
RL_LOCK(sc);
ifp = sc->rl_ifp;
/* 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
/*
* Clear WOL matching such that normal Rx filtering
* wouldn't interfere with WOL patterns.
*/
re_clrwol(sc);
/* 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;
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.
*/
static int
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);
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);
/*
* Mark interface as down since otherwise we will panic if
* interrupt comes in later on, which can happen in some
* cases.
*/
sc->rl_ifp->if_flags &= ~IFF_UP;
re_setwol(sc);
RL_UNLOCK(sc);
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
re_setwol(struct rl_softc *sc)
{
struct ifnet *ifp;
int pmc;
uint16_t pmstat;
uint8_t v;
RL_LOCK_ASSERT(sc);
if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
return;
ifp = sc->rl_ifp;
/* 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);
}
if ((ifp->if_capenable & IFCAP_WOL) != 0 &&
(sc->rl_flags & RL_FLAG_WOLRXENB) != 0)
CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RX_ENB);
/* Enable config register write. */
CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
/* Enable PME. */
v = CSR_READ_1(sc, RL_CFG1);
v &= ~RL_CFG1_PME;
if ((ifp->if_capenable & IFCAP_WOL) != 0)
v |= RL_CFG1_PME;
CSR_WRITE_1(sc, RL_CFG1, v);
v = CSR_READ_1(sc, RL_CFG3);
v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
v |= RL_CFG3_WOL_MAGIC;
CSR_WRITE_1(sc, RL_CFG3, v);
/* Config register write done. */
CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
v = CSR_READ_1(sc, RL_CFG5);
v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
v &= ~RL_CFG5_WOL_LANWAKE;
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;
CSR_WRITE_1(sc, RL_CFG5, v);
if ((ifp->if_capenable & IFCAP_WOL) != 0 &&
(sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0)
CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) & ~0x80);
/*
* 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
re_clrwol(struct rl_softc *sc)
{
int pmc;
uint8_t v;
RL_LOCK_ASSERT(sc);
if (pci_find_cap(sc->rl_dev, PCIY_PMG, &pmc) != 0)
return;
/* Enable config register write. */
CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE);
v = CSR_READ_1(sc, RL_CFG3);
v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC);
CSR_WRITE_1(sc, RL_CFG3, v);
/* Config register write done. */
CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF);
v = CSR_READ_1(sc, RL_CFG5);
v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST);
v &= ~RL_CFG5_WOL_LANWAKE;
CSR_WRITE_1(sc, RL_CFG5, v);
}
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;
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;
}
}
}
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);
if ((sc->rl_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
RL_UNLOCK(sc);
goto done;
}
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,
"DUMP statistics request timedout\n");
return (ETIMEDOUT);
}
done:
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));
}