d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
ffea8e2027
freebsd-skq/sys/dev/hme/if_hme.c

1739 lines
48 KiB
C
Raw Normal View History

Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
/*-
* Copyright (c) 1999 The NetBSD Foundation, Inc.
Convert to use bus_dmamap_load_mbuf(); decouple software TX descriptors from the hardware descriptors to avoid the overhead of having a DMA map for each of them. Bump the number of hardware descriptors to 128, and use half as many software descriptors for now. Some minor cleanups.
2003-01-06 22:12:57 +00:00
* Copyright (c) 2001-2003 Thomas Moestl <tmm@FreeBSD.org>.
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
* All rights reserved.
*
* This code is derived from software contributed to The NetBSD Foundation
* by Paul Kranenburg.
*
* 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 the NetBSD
* Foundation, Inc. and its contributors.
* 4. Neither the name of The NetBSD Foundation nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. 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 THE FOUNDATION OR CONTRIBUTORS
* 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.
*
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
* from: NetBSD: hme.c,v 1.45 2005/02/18 00:22:11 heas Exp
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
*/
Use __FBSDID(). Also some minor style cleanups.
2003-08-24 17:55:58 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
/*
* HME Ethernet module driver.
*
* The HME is e.g. part of the PCIO PCI multi function device.
* It supports TX gathering and TX and RX checksum offloading.
* RX buffers must be aligned at a programmable offset modulo 16. We choose 2
* for this offset: mbuf clusters are usually on about 2^11 boundaries, 2 bytes
* are skipped to make sure the header after the ethernet header is aligned on a
* natural boundary, so this ensures minimal wastage in the most common case.
*
* Also, apparently, the buffers must extend to a DMA burst boundary beyond the
* maximum packet size (this is not verified). Buffers starting on odd
* boundaries must be mapped so that the burst can start on a natural boundary.
*
Implement TCP/UDP Transmit/Receive checksum offload. Since HME doesn't compensate the checksum for UDP datagram which can yield to 0x0, UDP transmit checksum offload is disabled by default. The UDP Transmit checksum offload can be reactivated by setting special link option link0 with ifconfig(8). Approved by: jake (mentor) Reviewed by: tmm Tested by: Herve Boulouis <amon@sockar.homeip.net>
2004-08-05 02:52:33 +00:00
* STP2002QFP-UG says that Ethernet hardware supports TCP checksum offloading.
* In reality, we can do the same technique for UDP datagram too. However,
* the hardware doesn't compensate the checksum for UDP datagram which can yield
* to 0x0. As a safe guard, UDP checksum offload is disabled by default. It
* can be reactivated by setting special link option link0 with ifconfig(8).
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
*/
Implement TCP/UDP Transmit/Receive checksum offload. Since HME doesn't compensate the checksum for UDP datagram which can yield to 0x0, UDP transmit checksum offload is disabled by default. The UDP Transmit checksum offload can be reactivated by setting special link option link0 with ifconfig(8). Approved by: jake (mentor) Reviewed by: tmm Tested by: Herve Boulouis <amon@sockar.homeip.net>
2004-08-05 02:52:33 +00:00
#define HME_CSUM_FEATURES (CSUM_TCP)
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
#if 0
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
#define HMEDEBUG
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
#endif
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
#define KTR_HME KTR_CT2 /* XXX */
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/bus.h>
Move the new byte order function prototypes from <sys/param.h> to <sys/endian.h>. This puts us in line with NetBSD and OpenBSD.
2002-04-26 22:48:23 +00:00
#include <sys/endian.h>
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
#include <sys/kernel.h>
Add missing <sys/module.h> includes currently relying on nested include in <sys/kernel.h>
2004-06-03 06:10:02 +00:00
#include <sys/module.h>
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
#include <sys/ktr.h>
#include <sys/mbuf.h>
#include <sys/malloc.h>
#include <sys/socket.h>
#include <sys/sockio.h>
Call bpf_mtap() on output, to catch outgoing packets for e.g. tcpdump .
2002-07-30 21:47:14 +00:00
#include <net/bpf.h>
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
#include <net/ethernet.h>
#include <net/if.h>
#include <net/if_arp.h>
#include <net/if_dl.h>
#include <net/if_media.h>
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
#include <net/if_types.h>
The Sun hme hardware supposedly supports Tx frames up to 65535 octets, and Rx frames up to 8191 octets, so it is perfectly capable of supporting vlan(4)-style VLAN natively. Thus, make it support VLAN `oversize' frames. Reviewed by: tmm
2004-05-06 13:38:19 +00:00
#include <net/if_vlan_var.h>
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
Implement TCP/UDP Transmit/Receive checksum offload. Since HME doesn't compensate the checksum for UDP datagram which can yield to 0x0, UDP transmit checksum offload is disabled by default. The UDP Transmit checksum offload can be reactivated by setting special link option link0 with ifconfig(8). Approved by: jake (mentor) Reviewed by: tmm Tested by: Herve Boulouis <amon@sockar.homeip.net>
2004-08-05 02:52:33 +00:00
#include <netinet/in.h>
#include <netinet/in_systm.h>
#include <netinet/ip.h>
#include <netinet/tcp.h>
#include <netinet/udp.h>
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
#include <dev/mii/mii.h>
#include <dev/mii/miivar.h>
#include <machine/bus.h>
s#<foo/bar.h>#<dev/foo/bar.h>#g
2003-08-23 05:51:03 +00:00
#include <dev/hme/if_hmereg.h>
#include <dev/hme/if_hmevar.h>
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
CTASSERT(powerof2(HME_NRXDESC) && HME_NRXDESC >= 32 && HME_NRXDESC <= 256);
CTASSERT(HME_NTXDESC % 16 == 0 && HME_NTXDESC >= 16 && HME_NTXDESC <= 256);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
static void hme_start(struct ifnet *);
Make hme(4) mpsafe - Let hme_start()/hme_init() acquire lock and then call hme_start_locked()/hme_init_locked() respectivly. - Teardown interrupt handler before hme_detach(). - Remove IFF_NEEDSGIANT flag and mark interrupt handler INTR_MPSAFE. - Set callout handler to CALLOUT_MPSAFE. - Add locks in hme MII interface. Reviewed by: jake Tested by: Julian C. Dunn <jdunn at opentrend dot net> MFC after: 2 weeks
2004-11-22 06:46:30 +00:00
static void hme_start_locked(struct ifnet *);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
static void hme_stop(struct hme_softc *);
static int hme_ioctl(struct ifnet *, u_long, caddr_t);
static void hme_tick(void *);
- Use the hme_tick() callout instead of if_slowtimo() for driving hme_watchdog() in order to avoid races accessing if_timer. - Use bus_get_dma_tag() so hme(4) works on platforms requiring it. - Don't bother to set if_mtu to ETHERMTU, ether_ifattach() does that.
2006-12-06 02:07:20 +00:00
static int hme_watchdog(struct hme_softc *);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
static void hme_init(void *);
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
static void hme_init_locked(struct hme_softc *);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
static int hme_add_rxbuf(struct hme_softc *, unsigned int, int);
static int hme_meminit(struct hme_softc *);
static int hme_mac_bitflip(struct hme_softc *, u_int32_t, u_int32_t,
u_int32_t, u_int32_t);
static void hme_mifinit(struct hme_softc *);
static void hme_setladrf(struct hme_softc *, int);
static int hme_mediachange(struct ifnet *);
- In hme_stop() mask all interrupts. - In hme_eint() print MIF register contents on MIF interrupts. - In hme_mifinit() don't bother to preserve the previous MIF config. This was mainly done in order to preserve the PHY select bit (external or internal PHY) but which only needs to be set as appropriate when reading from or writing to the desired PHY in hme_mii_{read,write}reg(). Similarly don't bother to set the PHY select bit in hme_mii_statchg(). - In hme_mii_{read,write}reg() ignore requests to PHYs other than the external and internal PHY one. - Move enabling/disabling the MII drivers of the external transceiver from hme_init_locked() and based on the sheer presence of an external to hme_mifinit() and based on the currently selected media, defaulting to the internal transceiver when the media hasn't been set, yet. Invoke hme_mifinit() from the newly added hme_mediachange_locked() so the setting of the MII drivers is updated when changing media. These changes keep the MII bus from wedging (which manifests in the HME and the PHYs no longer being able to communicate with each other) when the PHY device drivers isolate the unused PHY in two-PHY configurations as present in f.e. Netra t1 100 while changing media, either from hme_init_locked() (see also below) or via ifconfig(8). They also allow for using both transceivers/PHYs. - In the newly added hme_mediachange_locked() also reset the PHYs in two- PHY configurations before invoking mii_mediachg(). This is required for successfully unisolating the previously unused PHY when switching between PHYs. - Now that changing media should no longer cause problems back out rev. 1.27 and re-enable setting the current media in hme_init_locked() (see the commit message of rev. 1.23 for more info). These changes are roughly a merge of NetBSD gem.c rev. 1.32 - 1.35 (1.30 was already fixed differently in our 1.36; 1.31 and 1.32 were wrong) with some parts reworked and things that don't make sense like setting the MII drivers and restoring the previous MIF and XIF settings in hme_mii_{read, write}reg() omitted. MFC after: 2 weeks
2006-12-04 00:51:08 +00:00
static int hme_mediachange_locked(struct hme_softc *);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
static void hme_mediastatus(struct ifnet *, struct ifmediareq *);
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
static int hme_load_txmbuf(struct hme_softc *, struct mbuf **);
Implement TCP/UDP Transmit/Receive checksum offload. Since HME doesn't compensate the checksum for UDP datagram which can yield to 0x0, UDP transmit checksum offload is disabled by default. The UDP Transmit checksum offload can be reactivated by setting special link option link0 with ifconfig(8). Approved by: jake (mentor) Reviewed by: tmm Tested by: Herve Boulouis <amon@sockar.homeip.net>
2004-08-05 02:52:33 +00:00
static void hme_read(struct hme_softc *, int, int, u_int32_t);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
static void hme_eint(struct hme_softc *, u_int);
static void hme_rint(struct hme_softc *);
static void hme_tint(struct hme_softc *);
Implement TCP/UDP Transmit/Receive checksum offload. Since HME doesn't compensate the checksum for UDP datagram which can yield to 0x0, UDP transmit checksum offload is disabled by default. The UDP Transmit checksum offload can be reactivated by setting special link option link0 with ifconfig(8). Approved by: jake (mentor) Reviewed by: tmm Tested by: Herve Boulouis <amon@sockar.homeip.net>
2004-08-05 02:52:33 +00:00
static void hme_rxcksum(struct mbuf *, u_int32_t);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
static void hme_cdma_callback(void *, bus_dma_segment_t *, int, int);
devclass_t hme_devclass;
static int hme_nerr;
DRIVER_MODULE(miibus, hme, miibus_driver, miibus_devclass, 0, 0);
Fix module dependancy name. PR: 46871 Submitted by: Hartmut Brandt <brandt@fokus.gmd.de>
2003-01-08 18:53:38 +00:00
MODULE_DEPEND(hme, miibus, 1, 1, 1);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
#define HME_SPC_READ_4(spc, sc, offs) \
bus_space_read_4((sc)->sc_ ## spc ## t, (sc)->sc_ ## spc ## h, \
- Use bus_space_subregion() rather than arithmetic on bus_space_handle_t. [1] - Properly use the error variable and return it on failure in the attach- routines. Reviewed by: tmm Inspired by: NetBSD [1]
2004-08-12 20:37:02 +00:00
(offs))
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
#define HME_SPC_WRITE_4(spc, sc, offs, v) \
bus_space_write_4((sc)->sc_ ## spc ## t, (sc)->sc_ ## spc ## h, \
- Use bus_space_subregion() rather than arithmetic on bus_space_handle_t. [1] - Properly use the error variable and return it on failure in the attach- routines. Reviewed by: tmm Inspired by: NetBSD [1]
2004-08-12 20:37:02 +00:00
(offs), (v))
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
#define HME_SPC_BARRIER(spc, sc, offs, l, f) \
bus_space_barrier((sc)->sc_ ## spc ## t, (sc)->sc_ ## spc ## h, \
(offs), (l), (f))
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
#define HME_SEB_READ_4(sc, offs) HME_SPC_READ_4(seb, (sc), (offs))
#define HME_SEB_WRITE_4(sc, offs, v) HME_SPC_WRITE_4(seb, (sc), (offs), (v))
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
#define HME_SEB_BARRIER(sc, offs, l, f) \
HME_SPC_BARRIER(seb, (sc), (offs), (l), (f))
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
#define HME_ERX_READ_4(sc, offs) HME_SPC_READ_4(erx, (sc), (offs))
#define HME_ERX_WRITE_4(sc, offs, v) HME_SPC_WRITE_4(erx, (sc), (offs), (v))
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
#define HME_ERX_BARRIER(sc, offs, l, f) \
HME_SPC_BARRIER(erx, (sc), (offs), (l), (f))
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
#define HME_ETX_READ_4(sc, offs) HME_SPC_READ_4(etx, (sc), (offs))
#define HME_ETX_WRITE_4(sc, offs, v) HME_SPC_WRITE_4(etx, (sc), (offs), (v))
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
#define HME_ETX_BARRIER(sc, offs, l, f) \
HME_SPC_BARRIER(etx, (sc), (offs), (l), (f))
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
#define HME_MAC_READ_4(sc, offs) HME_SPC_READ_4(mac, (sc), (offs))
#define HME_MAC_WRITE_4(sc, offs, v) HME_SPC_WRITE_4(mac, (sc), (offs), (v))
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
#define HME_MAC_BARRIER(sc, offs, l, f) \
HME_SPC_BARRIER(mac, (sc), (offs), (l), (f))
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
#define HME_MIF_READ_4(sc, offs) HME_SPC_READ_4(mif, (sc), (offs))
#define HME_MIF_WRITE_4(sc, offs, v) HME_SPC_WRITE_4(mif, (sc), (offs), (v))
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
#define HME_MIF_BARRIER(sc, offs, l, f) \
HME_SPC_BARRIER(mif, (sc), (offs), (l), (f))
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
#define HME_MAXERR 5
#define HME_WHINE(dev, ...) do { \
if (hme_nerr++ < HME_MAXERR) \
device_printf(dev, __VA_ARGS__); \
if (hme_nerr == HME_MAXERR) { \
Fix typo in printf string. MFC after: 1 week Approved by: cperciva (mentor)
2006-05-27 09:28:59 +00:00
device_printf(dev, "too many errors; not reporting " \
"any more\n"); \
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
} \
} while(0)
The Sun hme hardware supposedly supports Tx frames up to 65535 octets, and Rx frames up to 8191 octets, so it is perfectly capable of supporting vlan(4)-style VLAN natively. Thus, make it support VLAN `oversize' frames. Reviewed by: tmm
2004-05-06 13:38:19 +00:00
/* Support oversized VLAN frames. */
#define HME_MAX_FRAMESIZE (ETHER_MAX_LEN + ETHER_VLAN_ENCAP_LEN)
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
int
hme_config(struct hme_softc *sc)
{
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
struct ifnet *ifp;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
struct mii_softc *child;
bus_size_t size;
int error, rdesc, tdesc, i;
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
ifp = sc->sc_ifp = if_alloc(IFT_ETHER);
if (ifp == NULL)
return (ENOSPC);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
/*
* HME common initialization.
*
* hme_softc fields that must be initialized by the front-end:
*
Spelling and style fixes. Obtained from: NetBSD
2004-05-22 01:56:18 +00:00
* the DMA bus tag:
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
* sc_dmatag
*
* the bus handles, tags and offsets (splitted for SBus compatability):
* sc_seb{t,h,o} (Shared Ethernet Block registers)
* sc_erx{t,h,o} (Receiver Unit registers)
* sc_etx{t,h,o} (Transmitter Unit registers)
* sc_mac{t,h,o} (MAC registers)
Spelling and style fixes. Obtained from: NetBSD
2004-05-22 01:56:18 +00:00
* sc_mif{t,h,o} (Management Interface registers)
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
*
* the maximum bus burst size:
* sc_burst
*
*/
- Remove redundant assertions that the driver lock is not held in attach() and detach() since mtx_lock() will assert that already since the driver lock is not recursive. - Move the call to callout_init_mtx() before hme_stop() so that the callout_stop() in hme_stop() doesn't operate on an uninitialized callout structure during attach. Reported by: yongari (2) MFC after: 3 days
2005-08-24 20:28:56 +00:00
callout_init_mtx(&sc->sc_tick_ch, &sc->sc_lock, 0);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
/* Make sure the chip is stopped. */
Make hme(4) mpsafe - Let hme_start()/hme_init() acquire lock and then call hme_start_locked()/hme_init_locked() respectivly. - Teardown interrupt handler before hme_detach(). - Remove IFF_NEEDSGIANT flag and mark interrupt handler INTR_MPSAFE. - Set callout handler to CALLOUT_MPSAFE. - Add locks in hme MII interface. Reviewed by: jake Tested by: Julian C. Dunn <jdunn at opentrend dot net> MFC after: 2 weeks
2004-11-22 06:46:30 +00:00
HME_LOCK(sc);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
hme_stop(sc);
Make hme(4) mpsafe - Let hme_start()/hme_init() acquire lock and then call hme_start_locked()/hme_init_locked() respectivly. - Teardown interrupt handler before hme_detach(). - Remove IFF_NEEDSGIANT flag and mark interrupt handler INTR_MPSAFE. - Set callout handler to CALLOUT_MPSAFE. - Add locks in hme MII interface. Reviewed by: jake Tested by: Julian C. Dunn <jdunn at opentrend dot net> MFC after: 2 weeks
2004-11-22 06:46:30 +00:00
HME_UNLOCK(sc);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
error = bus_dma_tag_create(bus_get_dma_tag(sc->sc_dev), 1, 0,
BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL, NULL,
BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0,
NULL, NULL, &sc->sc_pdmatag);
if (error)
goto fail_ifnet;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
/*
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
* Create control, RX and TX mbuf DMA tags.
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
* Buffer descriptors must be aligned on a 2048 byte boundary;
* take this into account when calculating the size. Note that
* the maximum number of descriptors (256) occupies 2048 bytes,
* so we allocate that much regardless of HME_N*DESC.
*/
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
size = 4096;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
error = bus_dma_tag_create(sc->sc_pdmatag, 2048, 0,
BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL, NULL, size,
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
1, size, 0, busdma_lock_mutex, &sc->sc_lock, &sc->sc_cdmatag);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
if (error)
goto fail_ptag;
error = bus_dma_tag_create(sc->sc_pdmatag, max(0x10, sc->sc_burst), 0,
BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL, NULL, MCLBYTES,
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
1, MCLBYTES, BUS_DMA_ALLOCNOW, NULL, NULL, &sc->sc_rdmatag);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
if (error)
goto fail_ctag;
error = bus_dma_tag_create(sc->sc_pdmatag, max(0x10, sc->sc_burst), 0,
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL, NULL,
MCLBYTES * HME_NTXSEGS, HME_NTXSEGS, MCLBYTES, BUS_DMA_ALLOCNOW,
Mega busdma API commit. Add two new arguments to bus_dma_tag_create(): lockfunc and lockfuncarg. Lockfunc allows a driver to provide a function for managing its locking semantics while using busdma. At the moment, this is used for the asynchronous busdma_swi and callback mechanism. Two lockfunc implementations are provided: busdma_lock_mutex() performs standard mutex operations on the mutex that is specified from lockfuncarg. dftl_lock() is a panic implementation and is defaulted to when NULL, NULL are passed to bus_dma_tag_create(). The only time that NULL, NULL should ever be used is when the driver ensures that bus_dmamap_load() will not be deferred. Drivers that do not provide their own locking can pass busdma_lock_mutex,&Giant args in order to preserve the former behaviour. sparc64 and powerpc do not provide real busdma_swi functions, so this is largely a noop on those platforms. The busdma_swi on is64 is not properly locked yet, so warnings will be emitted on this platform when busdma callback deferrals happen. If anyone gets panics or warnings from dflt_lock() being called, please let me know right away. Reviewed by: tmm, gibbs
2003-07-01 15:52:06 +00:00
NULL, NULL, &sc->sc_tdmatag);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
if (error)
goto fail_rtag;
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
/* Allocate the control DMA buffer. */
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
error = bus_dmamem_alloc(sc->sc_cdmatag, (void **)&sc->sc_rb.rb_membase,
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO, &sc->sc_cdmamap);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
if (error != 0) {
device_printf(sc->sc_dev, "DMA buffer alloc error %d\n", error);
goto fail_ttag;
}
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
/* Load the control DMA buffer. */
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
sc->sc_rb.rb_dmabase = 0;
if ((error = bus_dmamap_load(sc->sc_cdmatag, sc->sc_cdmamap,
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
sc->sc_rb.rb_membase, size, hme_cdma_callback, sc, 0)) != 0 ||
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
sc->sc_rb.rb_dmabase == 0) {
device_printf(sc->sc_dev, "DMA buffer map load error %d\n",
error);
goto fail_free;
}
CTR2(KTR_HME, "hme_config: dma va %p, pa %#lx", sc->sc_rb.rb_membase,
sc->sc_rb.rb_dmabase);
/*
* Prepare the RX descriptors. rdesc serves as marker for the last
* processed descriptor and may be used later on.
*/
for (rdesc = 0; rdesc < HME_NRXDESC; rdesc++) {
Miscellaneous fixes: - always reinitialize the rx descriptors, even if the mbuf is kept. This should fix the hangs on ifconfig that were observed - on an rx overflow, reinitialize the descriptor so that the interface will not hang - correct some bus_dmamap_sync() calls - correct some debug messages - some minor nits
2002-07-14 12:09:48 +00:00
sc->sc_rb.rb_rxdesc[rdesc].hrx_m = NULL;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
error = bus_dmamap_create(sc->sc_rdmatag, 0,
&sc->sc_rb.rb_rxdesc[rdesc].hrx_dmamap);
if (error != 0)
goto fail_rxdesc;
}
error = bus_dmamap_create(sc->sc_rdmatag, 0,
&sc->sc_rb.rb_spare_dmamap);
if (error != 0)
goto fail_rxdesc;
/* Same for the TX descs. */
Convert to use bus_dmamap_load_mbuf(); decouple software TX descriptors from the hardware descriptors to avoid the overhead of having a DMA map for each of them. Bump the number of hardware descriptors to 128, and use half as many software descriptors for now. Some minor cleanups.
2003-01-06 22:12:57 +00:00
for (tdesc = 0; tdesc < HME_NTXQ; tdesc++) {
Miscellaneous fixes: - always reinitialize the rx descriptors, even if the mbuf is kept. This should fix the hangs on ifconfig that were observed - on an rx overflow, reinitialize the descriptor so that the interface will not hang - correct some bus_dmamap_sync() calls - correct some debug messages - some minor nits
2002-07-14 12:09:48 +00:00
sc->sc_rb.rb_txdesc[tdesc].htx_m = NULL;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
error = bus_dmamap_create(sc->sc_tdmatag, 0,
&sc->sc_rb.rb_txdesc[tdesc].htx_dmamap);
if (error != 0)
goto fail_txdesc;
}
Implement TCP/UDP Transmit/Receive checksum offload. Since HME doesn't compensate the checksum for UDP datagram which can yield to 0x0, UDP transmit checksum offload is disabled by default. The UDP Transmit checksum offload can be reactivated by setting special link option link0 with ifconfig(8). Approved by: jake (mentor) Reviewed by: tmm Tested by: Herve Boulouis <amon@sockar.homeip.net>
2004-08-05 02:52:33 +00:00
sc->sc_csum_features = HME_CSUM_FEATURES;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
/* Initialize ifnet structure. */
ifp->if_softc = sc;
Replace the if_name and if_unit members of struct ifnet with new members if_xname, if_dname, and if_dunit. if_xname is the name of the interface and if_dname/unit are the driver name and instance. This change paves the way for interface renaming and enhanced pseudo device creation and configuration symantics. Approved By: re (in principle) Reviewed By: njl, imp Tested On: i386, amd64, sparc64 Obtained From: NetBSD (if_xname)
2003-10-31 18:32:15 +00:00
if_initname(ifp, device_get_name(sc->sc_dev),
device_get_unit(sc->sc_dev));
Make hme(4) mpsafe - Let hme_start()/hme_init() acquire lock and then call hme_start_locked()/hme_init_locked() respectivly. - Teardown interrupt handler before hme_detach(). - Remove IFF_NEEDSGIANT flag and mark interrupt handler INTR_MPSAFE. - Set callout handler to CALLOUT_MPSAFE. - Add locks in hme MII interface. Reviewed by: jake Tested by: Julian C. Dunn <jdunn at opentrend dot net> MFC after: 2 weeks
2004-11-22 06:46:30 +00:00
ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
ifp->if_start = hme_start;
ifp->if_ioctl = hme_ioctl;
ifp->if_init = hme_init;
Second part of ALTQ driver modifications, covering: an(4), ath(4), hme(4), ndis(4), vr(4) and wi(4) Please help testing: http://people.freebsd.org/~mlaier/ALTQ_driver/ Tested by: Vaidas Damosevicius (an, ath, wi) Roman Divacky (vr) Submitted by: yongari (hme)
2004-08-01 23:58:04 +00:00
IFQ_SET_MAXLEN(&ifp->if_snd, HME_NTXQ);
ifp->if_snd.ifq_drv_maxlen = HME_NTXQ;
IFQ_SET_READY(&ifp->if_snd);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
hme_mifinit(sc);
if ((error = mii_phy_probe(sc->sc_dev, &sc->sc_miibus, hme_mediachange,
hme_mediastatus)) != 0) {
device_printf(sc->sc_dev, "phy probe failed: %d\n", error);
goto fail_rxdesc;
}
sc->sc_mii = device_get_softc(sc->sc_miibus);
/*
* Walk along the list of attached MII devices and
- In hme_stop() mask all interrupts. - In hme_eint() print MIF register contents on MIF interrupts. - In hme_mifinit() don't bother to preserve the previous MIF config. This was mainly done in order to preserve the PHY select bit (external or internal PHY) but which only needs to be set as appropriate when reading from or writing to the desired PHY in hme_mii_{read,write}reg(). Similarly don't bother to set the PHY select bit in hme_mii_statchg(). - In hme_mii_{read,write}reg() ignore requests to PHYs other than the external and internal PHY one. - Move enabling/disabling the MII drivers of the external transceiver from hme_init_locked() and based on the sheer presence of an external to hme_mifinit() and based on the currently selected media, defaulting to the internal transceiver when the media hasn't been set, yet. Invoke hme_mifinit() from the newly added hme_mediachange_locked() so the setting of the MII drivers is updated when changing media. These changes keep the MII bus from wedging (which manifests in the HME and the PHYs no longer being able to communicate with each other) when the PHY device drivers isolate the unused PHY in two-PHY configurations as present in f.e. Netra t1 100 while changing media, either from hme_init_locked() (see also below) or via ifconfig(8). They also allow for using both transceivers/PHYs. - In the newly added hme_mediachange_locked() also reset the PHYs in two- PHY configurations before invoking mii_mediachg(). This is required for successfully unisolating the previously unused PHY when switching between PHYs. - Now that changing media should no longer cause problems back out rev. 1.27 and re-enable setting the current media in hme_init_locked() (see the commit message of rev. 1.23 for more info). These changes are roughly a merge of NetBSD gem.c rev. 1.32 - 1.35 (1.30 was already fixed differently in our 1.36; 1.31 and 1.32 were wrong) with some parts reworked and things that don't make sense like setting the MII drivers and restoring the previous MIF and XIF settings in hme_mii_{read, write}reg() omitted. MFC after: 2 weeks
2006-12-04 00:51:08 +00:00
* establish an `MII instance' to `PHY number'
* mapping. We'll use this mapping to enable the MII
* drivers of the external transceiver according to
* the currently selected media.
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
*/
- In hme_stop() mask all interrupts. - In hme_eint() print MIF register contents on MIF interrupts. - In hme_mifinit() don't bother to preserve the previous MIF config. This was mainly done in order to preserve the PHY select bit (external or internal PHY) but which only needs to be set as appropriate when reading from or writing to the desired PHY in hme_mii_{read,write}reg(). Similarly don't bother to set the PHY select bit in hme_mii_statchg(). - In hme_mii_{read,write}reg() ignore requests to PHYs other than the external and internal PHY one. - Move enabling/disabling the MII drivers of the external transceiver from hme_init_locked() and based on the sheer presence of an external to hme_mifinit() and based on the currently selected media, defaulting to the internal transceiver when the media hasn't been set, yet. Invoke hme_mifinit() from the newly added hme_mediachange_locked() so the setting of the MII drivers is updated when changing media. These changes keep the MII bus from wedging (which manifests in the HME and the PHYs no longer being able to communicate with each other) when the PHY device drivers isolate the unused PHY in two-PHY configurations as present in f.e. Netra t1 100 while changing media, either from hme_init_locked() (see also below) or via ifconfig(8). They also allow for using both transceivers/PHYs. - In the newly added hme_mediachange_locked() also reset the PHYs in two- PHY configurations before invoking mii_mediachg(). This is required for successfully unisolating the previously unused PHY when switching between PHYs. - Now that changing media should no longer cause problems back out rev. 1.27 and re-enable setting the current media in hme_init_locked() (see the commit message of rev. 1.23 for more info). These changes are roughly a merge of NetBSD gem.c rev. 1.32 - 1.35 (1.30 was already fixed differently in our 1.36; 1.31 and 1.32 were wrong) with some parts reworked and things that don't make sense like setting the MII drivers and restoring the previous MIF and XIF settings in hme_mii_{read, write}reg() omitted. MFC after: 2 weeks
2006-12-04 00:51:08 +00:00
sc->sc_phys[0] = sc->sc_phys[1] = -1;
LIST_FOREACH(child, &sc->sc_mii->mii_phys, mii_list) {
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
/*
* Note: we support just two PHYs: the built-in
* internal device and an external on the MII
* connector.
*/
- In hme_stop() mask all interrupts. - In hme_eint() print MIF register contents on MIF interrupts. - In hme_mifinit() don't bother to preserve the previous MIF config. This was mainly done in order to preserve the PHY select bit (external or internal PHY) but which only needs to be set as appropriate when reading from or writing to the desired PHY in hme_mii_{read,write}reg(). Similarly don't bother to set the PHY select bit in hme_mii_statchg(). - In hme_mii_{read,write}reg() ignore requests to PHYs other than the external and internal PHY one. - Move enabling/disabling the MII drivers of the external transceiver from hme_init_locked() and based on the sheer presence of an external to hme_mifinit() and based on the currently selected media, defaulting to the internal transceiver when the media hasn't been set, yet. Invoke hme_mifinit() from the newly added hme_mediachange_locked() so the setting of the MII drivers is updated when changing media. These changes keep the MII bus from wedging (which manifests in the HME and the PHYs no longer being able to communicate with each other) when the PHY device drivers isolate the unused PHY in two-PHY configurations as present in f.e. Netra t1 100 while changing media, either from hme_init_locked() (see also below) or via ifconfig(8). They also allow for using both transceivers/PHYs. - In the newly added hme_mediachange_locked() also reset the PHYs in two- PHY configurations before invoking mii_mediachg(). This is required for successfully unisolating the previously unused PHY when switching between PHYs. - Now that changing media should no longer cause problems back out rev. 1.27 and re-enable setting the current media in hme_init_locked() (see the commit message of rev. 1.23 for more info). These changes are roughly a merge of NetBSD gem.c rev. 1.32 - 1.35 (1.30 was already fixed differently in our 1.36; 1.31 and 1.32 were wrong) with some parts reworked and things that don't make sense like setting the MII drivers and restoring the previous MIF and XIF settings in hme_mii_{read, write}reg() omitted. MFC after: 2 weeks
2006-12-04 00:51:08 +00:00
if ((child->mii_phy != HME_PHYAD_EXTERNAL &&
child->mii_phy != HME_PHYAD_INTERNAL) ||
child->mii_inst > 1) {
Spelling and style fixes. Obtained from: NetBSD
2004-05-22 01:56:18 +00:00
device_printf(sc->sc_dev, "cannot accommodate "
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
"MII device %s at phy %d, instance %d\n",
device_get_name(child->mii_dev),
child->mii_phy, child->mii_inst);
continue;
}
sc->sc_phys[child->mii_inst] = child->mii_phy;
}
/* Attach the interface. */
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
ether_ifattach(ifp, sc->sc_enaddr);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
The Sun hme hardware supposedly supports Tx frames up to 65535 octets, and Rx frames up to 8191 octets, so it is perfectly capable of supporting vlan(4)-style VLAN natively. Thus, make it support VLAN `oversize' frames. Reviewed by: tmm
2004-05-06 13:38:19 +00:00
/*
Implement TCP/UDP Transmit/Receive checksum offload. Since HME doesn't compensate the checksum for UDP datagram which can yield to 0x0, UDP transmit checksum offload is disabled by default. The UDP Transmit checksum offload can be reactivated by setting special link option link0 with ifconfig(8). Approved by: jake (mentor) Reviewed by: tmm Tested by: Herve Boulouis <amon@sockar.homeip.net>
2004-08-05 02:52:33 +00:00
* Tell the upper layer(s) we support long frames/checksum offloads.
The Sun hme hardware supposedly supports Tx frames up to 65535 octets, and Rx frames up to 8191 octets, so it is perfectly capable of supporting vlan(4)-style VLAN natively. Thus, make it support VLAN `oversize' frames. Reviewed by: tmm
2004-05-06 13:38:19 +00:00
*/
ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
Implement TCP/UDP Transmit/Receive checksum offload. Since HME doesn't compensate the checksum for UDP datagram which can yield to 0x0, UDP transmit checksum offload is disabled by default. The UDP Transmit checksum offload can be reactivated by setting special link option link0 with ifconfig(8). Approved by: jake (mentor) Reviewed by: tmm Tested by: Herve Boulouis <amon@sockar.homeip.net>
2004-08-05 02:52:33 +00:00
ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_HWCSUM;
ifp->if_hwassist |= sc->sc_csum_features;
ifp->if_capenable |= IFCAP_VLAN_MTU | IFCAP_HWCSUM;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
return (0);
fail_txdesc:
for (i = 0; i < tdesc; i++) {
bus_dmamap_destroy(sc->sc_tdmatag,
sc->sc_rb.rb_txdesc[i].htx_dmamap);
}
bus_dmamap_destroy(sc->sc_rdmatag, sc->sc_rb.rb_spare_dmamap);
fail_rxdesc:
for (i = 0; i < rdesc; i++) {
bus_dmamap_destroy(sc->sc_rdmatag,
sc->sc_rb.rb_rxdesc[i].hrx_dmamap);
}
bus_dmamap_unload(sc->sc_cdmatag, sc->sc_cdmamap);
fail_free:
bus_dmamem_free(sc->sc_cdmatag, sc->sc_rb.rb_membase, sc->sc_cdmamap);
fail_ttag:
bus_dma_tag_destroy(sc->sc_tdmatag);
fail_rtag:
bus_dma_tag_destroy(sc->sc_rdmatag);
fail_ctag:
bus_dma_tag_destroy(sc->sc_cdmatag);
fail_ptag:
bus_dma_tag_destroy(sc->sc_pdmatag);
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
fail_ifnet:
if_free(ifp);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
return (error);
}
Add detach, shutdown, suspend and resume methods. Tested by: jake
2003-01-09 00:45:10 +00:00
void
hme_detach(struct hme_softc *sc)
{
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
struct ifnet *ifp = sc->sc_ifp;
Add detach, shutdown, suspend and resume methods. Tested by: jake
2003-01-09 00:45:10 +00:00
int i;
Make hme(4) mpsafe - Let hme_start()/hme_init() acquire lock and then call hme_start_locked()/hme_init_locked() respectivly. - Teardown interrupt handler before hme_detach(). - Remove IFF_NEEDSGIANT flag and mark interrupt handler INTR_MPSAFE. - Set callout handler to CALLOUT_MPSAFE. - Add locks in hme MII interface. Reviewed by: jake Tested by: Julian C. Dunn <jdunn at opentrend dot net> MFC after: 2 weeks
2004-11-22 06:46:30 +00:00
HME_LOCK(sc);
Add detach, shutdown, suspend and resume methods. Tested by: jake
2003-01-09 00:45:10 +00:00
hme_stop(sc);
Make hme(4) mpsafe - Let hme_start()/hme_init() acquire lock and then call hme_start_locked()/hme_init_locked() respectivly. - Teardown interrupt handler before hme_detach(). - Remove IFF_NEEDSGIANT flag and mark interrupt handler INTR_MPSAFE. - Set callout handler to CALLOUT_MPSAFE. - Add locks in hme MII interface. Reviewed by: jake Tested by: Julian C. Dunn <jdunn at opentrend dot net> MFC after: 2 weeks
2004-11-22 06:46:30 +00:00
HME_UNLOCK(sc);
Add callout_drain()'s to foo_detach() after calling foo_stop() to make sure that if softclock is running on another CPU and is blocked on our driver lock, we will wait until it has acquired the lock, seen that it was cancelled, dropped the lock, and awakened us so that we can safely destroy the mutex. MFC after: 3 days
2005-08-17 17:44:32 +00:00
callout_drain(&sc->sc_tick_ch);
Move hme_stop() after ether_ifdetach() and if_free() to prevent a memory-referenced-after-free panic if the hme interface fails to attach. Patch obtained from: marius Suggested same thing: brooks MFC after: 3 days
2005-09-08 13:50:16 +00:00
ether_ifdetach(ifp);
if_free(ifp);
Add detach, shutdown, suspend and resume methods. Tested by: jake
2003-01-09 00:45:10 +00:00
device_delete_child(sc->sc_dev, sc->sc_miibus);
for (i = 0; i < HME_NTXQ; i++) {
bus_dmamap_destroy(sc->sc_tdmatag,
sc->sc_rb.rb_txdesc[i].htx_dmamap);
}
bus_dmamap_destroy(sc->sc_rdmatag, sc->sc_rb.rb_spare_dmamap);
for (i = 0; i < HME_NRXDESC; i++) {
bus_dmamap_destroy(sc->sc_rdmatag,
sc->sc_rb.rb_rxdesc[i].hrx_dmamap);
}
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
bus_dmamap_sync(sc->sc_cdmatag, sc->sc_cdmamap,
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
Add detach, shutdown, suspend and resume methods. Tested by: jake
2003-01-09 00:45:10 +00:00
bus_dmamap_unload(sc->sc_cdmatag, sc->sc_cdmamap);
bus_dmamem_free(sc->sc_cdmatag, sc->sc_rb.rb_membase, sc->sc_cdmamap);
bus_dma_tag_destroy(sc->sc_tdmatag);
bus_dma_tag_destroy(sc->sc_rdmatag);
bus_dma_tag_destroy(sc->sc_cdmatag);
bus_dma_tag_destroy(sc->sc_pdmatag);
}
void
hme_suspend(struct hme_softc *sc)
{
Make hme(4) mpsafe - Let hme_start()/hme_init() acquire lock and then call hme_start_locked()/hme_init_locked() respectivly. - Teardown interrupt handler before hme_detach(). - Remove IFF_NEEDSGIANT flag and mark interrupt handler INTR_MPSAFE. - Set callout handler to CALLOUT_MPSAFE. - Add locks in hme MII interface. Reviewed by: jake Tested by: Julian C. Dunn <jdunn at opentrend dot net> MFC after: 2 weeks
2004-11-22 06:46:30 +00:00
HME_LOCK(sc);
Add detach, shutdown, suspend and resume methods. Tested by: jake
2003-01-09 00:45:10 +00:00
hme_stop(sc);
Make hme(4) mpsafe - Let hme_start()/hme_init() acquire lock and then call hme_start_locked()/hme_init_locked() respectivly. - Teardown interrupt handler before hme_detach(). - Remove IFF_NEEDSGIANT flag and mark interrupt handler INTR_MPSAFE. - Set callout handler to CALLOUT_MPSAFE. - Add locks in hme MII interface. Reviewed by: jake Tested by: Julian C. Dunn <jdunn at opentrend dot net> MFC after: 2 weeks
2004-11-22 06:46:30 +00:00
HME_UNLOCK(sc);
Add detach, shutdown, suspend and resume methods. Tested by: jake
2003-01-09 00:45:10 +00:00
}
void
hme_resume(struct hme_softc *sc)
{
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
struct ifnet *ifp = sc->sc_ifp;
Add detach, shutdown, suspend and resume methods. Tested by: jake
2003-01-09 00:45:10 +00:00
Make hme(4) mpsafe - Let hme_start()/hme_init() acquire lock and then call hme_start_locked()/hme_init_locked() respectivly. - Teardown interrupt handler before hme_detach(). - Remove IFF_NEEDSGIANT flag and mark interrupt handler INTR_MPSAFE. - Set callout handler to CALLOUT_MPSAFE. - Add locks in hme MII interface. Reviewed by: jake Tested by: Julian C. Dunn <jdunn at opentrend dot net> MFC after: 2 weeks
2004-11-22 06:46:30 +00:00
HME_LOCK(sc);
Add detach, shutdown, suspend and resume methods. Tested by: jake
2003-01-09 00:45:10 +00:00
if ((ifp->if_flags & IFF_UP) != 0)
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
hme_init_locked(sc);
Make hme(4) mpsafe - Let hme_start()/hme_init() acquire lock and then call hme_start_locked()/hme_init_locked() respectivly. - Teardown interrupt handler before hme_detach(). - Remove IFF_NEEDSGIANT flag and mark interrupt handler INTR_MPSAFE. - Set callout handler to CALLOUT_MPSAFE. - Add locks in hme MII interface. Reviewed by: jake Tested by: Julian C. Dunn <jdunn at opentrend dot net> MFC after: 2 weeks
2004-11-22 06:46:30 +00:00
HME_UNLOCK(sc);
Add detach, shutdown, suspend and resume methods. Tested by: jake
2003-01-09 00:45:10 +00:00
}
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
static void
hme_cdma_callback(void *xsc, bus_dma_segment_t *segs, int nsegs, int error)
{
struct hme_softc *sc = (struct hme_softc *)xsc;
if (error != 0)
return;
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
KASSERT(nsegs == 1,
("%s: too many DMA segments (%d)", __func__, nsegs));
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
sc->sc_rb.rb_dmabase = segs[0].ds_addr;
}
static void
hme_tick(void *arg)
{
struct hme_softc *sc = arg;
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
struct ifnet *ifp;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
- Use callout_init_mtx() to close races between hme_stop() and hme_tick(). - Use the driver lock instead of Giant in a bus dma callback. - Clear IFF_DRV_(RUNNING|OACTIVE) in hme_stop() instead of just clearing RUNNING in hme_ioctl() to be more like other ethernet drivers. - Lock the driver lock around mii operations. - Remove spls. - Cleanup locking in hme_ioctl(). MFC after: 1 week
2005-08-10 20:12:25 +00:00
HME_LOCK_ASSERT(sc, MA_OWNED);
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
ifp = sc->sc_ifp;
/*
* Unload collision counters
*/
ifp->if_collisions +=
HME_MAC_READ_4(sc, HME_MACI_NCCNT) +
HME_MAC_READ_4(sc, HME_MACI_FCCNT) +
HME_MAC_READ_4(sc, HME_MACI_EXCNT) +
HME_MAC_READ_4(sc, HME_MACI_LTCNT);
/*
* then clear the hardware counters.
*/
HME_MAC_WRITE_4(sc, HME_MACI_NCCNT, 0);
HME_MAC_WRITE_4(sc, HME_MACI_FCCNT, 0);
HME_MAC_WRITE_4(sc, HME_MACI_EXCNT, 0);
HME_MAC_WRITE_4(sc, HME_MACI_LTCNT, 0);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
mii_tick(sc->sc_mii);
- Use the hme_tick() callout instead of if_slowtimo() for driving hme_watchdog() in order to avoid races accessing if_timer. - Use bus_get_dma_tag() so hme(4) works on platforms requiring it. - Don't bother to set if_mtu to ETHERMTU, ether_ifattach() does that.
2006-12-06 02:07:20 +00:00
if (hme_watchdog(sc) == EJUSTRETURN)
return;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
callout_reset(&sc->sc_tick_ch, hz, hme_tick, sc);
}
static void
hme_stop(struct hme_softc *sc)
{
u_int32_t v;
int n;
callout_stop(&sc->sc_tick_ch);
- Use the hme_tick() callout instead of if_slowtimo() for driving hme_watchdog() in order to avoid races accessing if_timer. - Use bus_get_dma_tag() so hme(4) works on platforms requiring it. - Don't bother to set if_mtu to ETHERMTU, ether_ifattach() does that.
2006-12-06 02:07:20 +00:00
sc->sc_wdog_timer = 0;
- Use callout_init_mtx() to close races between hme_stop() and hme_tick(). - Use the driver lock instead of Giant in a bus dma callback. - Clear IFF_DRV_(RUNNING|OACTIVE) in hme_stop() instead of just clearing RUNNING in hme_ioctl() to be more like other ethernet drivers. - Lock the driver lock around mii operations. - Remove spls. - Cleanup locking in hme_ioctl(). MFC after: 1 week
2005-08-10 20:12:25 +00:00
sc->sc_ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
sc->sc_flags &= ~HME_LINK;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
- In hme_stop() mask all interrupts. - In hme_eint() print MIF register contents on MIF interrupts. - In hme_mifinit() don't bother to preserve the previous MIF config. This was mainly done in order to preserve the PHY select bit (external or internal PHY) but which only needs to be set as appropriate when reading from or writing to the desired PHY in hme_mii_{read,write}reg(). Similarly don't bother to set the PHY select bit in hme_mii_statchg(). - In hme_mii_{read,write}reg() ignore requests to PHYs other than the external and internal PHY one. - Move enabling/disabling the MII drivers of the external transceiver from hme_init_locked() and based on the sheer presence of an external to hme_mifinit() and based on the currently selected media, defaulting to the internal transceiver when the media hasn't been set, yet. Invoke hme_mifinit() from the newly added hme_mediachange_locked() so the setting of the MII drivers is updated when changing media. These changes keep the MII bus from wedging (which manifests in the HME and the PHYs no longer being able to communicate with each other) when the PHY device drivers isolate the unused PHY in two-PHY configurations as present in f.e. Netra t1 100 while changing media, either from hme_init_locked() (see also below) or via ifconfig(8). They also allow for using both transceivers/PHYs. - In the newly added hme_mediachange_locked() also reset the PHYs in two- PHY configurations before invoking mii_mediachg(). This is required for successfully unisolating the previously unused PHY when switching between PHYs. - Now that changing media should no longer cause problems back out rev. 1.27 and re-enable setting the current media in hme_init_locked() (see the commit message of rev. 1.23 for more info). These changes are roughly a merge of NetBSD gem.c rev. 1.32 - 1.35 (1.30 was already fixed differently in our 1.36; 1.31 and 1.32 were wrong) with some parts reworked and things that don't make sense like setting the MII drivers and restoring the previous MIF and XIF settings in hme_mii_{read, write}reg() omitted. MFC after: 2 weeks
2006-12-04 00:51:08 +00:00
/* Mask all interrupts */
HME_SEB_WRITE_4(sc, HME_SEBI_IMASK, 0xffffffff);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
/* Reset transmitter and receiver */
HME_SEB_WRITE_4(sc, HME_SEBI_RESET, HME_SEB_RESET_ETX |
HME_SEB_RESET_ERX);
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
HME_SEB_BARRIER(sc, HME_SEBI_RESET, 4,
BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
for (n = 0; n < 20; n++) {
v = HME_SEB_READ_4(sc, HME_SEBI_RESET);
if ((v & (HME_SEB_RESET_ETX | HME_SEB_RESET_ERX)) == 0)
return;
DELAY(20);
}
device_printf(sc->sc_dev, "hme_stop: reset failed\n");
}
Miscellaneous fixes: - always reinitialize the rx descriptors, even if the mbuf is kept. This should fix the hangs on ifconfig that were observed - on an rx overflow, reinitialize the descriptor so that the interface will not hang - correct some bus_dmamap_sync() calls - correct some debug messages - some minor nits
2002-07-14 12:09:48 +00:00
/*
* Discard the contents of an mbuf in the RX ring, freeing the buffer in the
* ring for subsequent use.
*/
bus_dmamap_sync() overhaul: - Remove NetBSD-style or-ed together BUS_DMASYNC operations, in some cases relaxing the (intended) syncing operation a bit. - Add syncs before reading the descriptor rings. - Try to combine syncs where possible to avoid overhead. - Sync all maps before unloading them.
2003-01-21 17:22:52 +00:00
static __inline void
hme_discard_rxbuf(struct hme_softc *sc, int ix)
Miscellaneous fixes: - always reinitialize the rx descriptors, even if the mbuf is kept. This should fix the hangs on ifconfig that were observed - on an rx overflow, reinitialize the descriptor so that the interface will not hang - correct some bus_dmamap_sync() calls - correct some debug messages - some minor nits
2002-07-14 12:09:48 +00:00
{
/*
* Dropped a packet, reinitialize the descriptor and turn the
* ownership back to the hardware.
*/
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
HME_XD_SETFLAGS(sc->sc_flags & HME_PCI, sc->sc_rb.rb_rxd,
ix, HME_XD_OWN | HME_XD_ENCODE_RSIZE(HME_DESC_RXLEN(sc,
&sc->sc_rb.rb_rxdesc[ix])));
Miscellaneous fixes: - always reinitialize the rx descriptors, even if the mbuf is kept. This should fix the hangs on ifconfig that were observed - on an rx overflow, reinitialize the descriptor so that the interface will not hang - correct some bus_dmamap_sync() calls - correct some debug messages - some minor nits
2002-07-14 12:09:48 +00:00
}
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
static int
hme_add_rxbuf(struct hme_softc *sc, unsigned int ri, int keepold)
{
struct hme_rxdesc *rd;
struct mbuf *m;
Use bus_dmamap_load_mbuf_sg() for loading rx buffers.
2005-01-15 22:05:59 +00:00
bus_dma_segment_t segs[1];
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
bus_dmamap_t map;
Convert to use bus_dmamap_load_mbuf(); decouple software TX descriptors from the hardware descriptors to avoid the overhead of having a DMA map for each of them. Bump the number of hardware descriptors to 128, and use half as many software descriptors for now. Some minor cleanups.
2003-01-06 22:12:57 +00:00
uintptr_t b;
Use bus_dmamap_load_mbuf_sg() for loading rx buffers.
2005-01-15 22:05:59 +00:00
int a, unmap, nsegs;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
rd = &sc->sc_rb.rb_rxdesc[ri];
unmap = rd->hrx_m != NULL;
Miscellaneous fixes: - always reinitialize the rx descriptors, even if the mbuf is kept. This should fix the hangs on ifconfig that were observed - on an rx overflow, reinitialize the descriptor so that the interface will not hang - correct some bus_dmamap_sync() calls - correct some debug messages - some minor nits
2002-07-14 12:09:48 +00:00
if (unmap && keepold) {
/*
* Reinitialize the descriptor flags, as they may have been
* altered by the hardware.
*/
bus_dmamap_sync() overhaul: - Remove NetBSD-style or-ed together BUS_DMASYNC operations, in some cases relaxing the (intended) syncing operation a bit. - Add syncs before reading the descriptor rings. - Try to combine syncs where possible to avoid overhead. - Sync all maps before unloading them.
2003-01-21 17:22:52 +00:00
hme_discard_rxbuf(sc, ri);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
return (0);
Miscellaneous fixes: - always reinitialize the rx descriptors, even if the mbuf is kept. This should fix the hangs on ifconfig that were observed - on an rx overflow, reinitialize the descriptor so that the interface will not hang - correct some bus_dmamap_sync() calls - correct some debug messages - some minor nits
2002-07-14 12:09:48 +00:00
}
Back out M_* changes, per decision of the TRB. Approved by: trb
2003-02-19 05:47:46 +00:00
if ((m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR)) == NULL)
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
return (ENOBUFS);
Convert to use bus_dmamap_load_mbuf(); decouple software TX descriptors from the hardware descriptors to avoid the overhead of having a DMA map for each of them. Bump the number of hardware descriptors to 128, and use half as many software descriptors for now. Some minor cleanups.
2003-01-06 22:12:57 +00:00
m->m_len = m->m_pkthdr.len = m->m_ext.ext_size;
b = mtod(m, uintptr_t);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
/*
* Required alignment boundary. At least 16 is needed, but since
* the mapping must be done in a way that a burst can start on a
* natural boundary we might need to extend this.
*/
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
a = imax(HME_MINRXALIGN, sc->sc_burst);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
/*
Convert to use bus_dmamap_load_mbuf(); decouple software TX descriptors from the hardware descriptors to avoid the overhead of having a DMA map for each of them. Bump the number of hardware descriptors to 128, and use half as many software descriptors for now. Some minor cleanups.
2003-01-06 22:12:57 +00:00
* Make sure the buffer suitably aligned. The 2 byte offset is removed
* when the mbuf is handed up. XXX: this ensures at least 16 byte
* alignment of the header adjacent to the ethernet header, which
* should be sufficient in all cases. Nevertheless, this second-guesses
* ALIGN().
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
*/
Convert to use bus_dmamap_load_mbuf(); decouple software TX descriptors from the hardware descriptors to avoid the overhead of having a DMA map for each of them. Bump the number of hardware descriptors to 128, and use half as many software descriptors for now. Some minor cleanups.
2003-01-06 22:12:57 +00:00
m_adj(m, roundup2(b, a) - b);
Use bus_dmamap_load_mbuf_sg() for loading rx buffers.
2005-01-15 22:05:59 +00:00
if (bus_dmamap_load_mbuf_sg(sc->sc_rdmatag, sc->sc_rb.rb_spare_dmamap,
m, segs, &nsegs, 0) != 0) {
Convert to use bus_dmamap_load_mbuf(); decouple software TX descriptors from the hardware descriptors to avoid the overhead of having a DMA map for each of them. Bump the number of hardware descriptors to 128, and use half as many software descriptors for now. Some minor cleanups.
2003-01-06 22:12:57 +00:00
m_freem(m);
return (ENOBUFS);
}
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
/* If nsegs is wrong then the stack is corrupt. */
KASSERT(nsegs == 1,
("%s: too many DMA segments (%d)", __func__, nsegs));
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
if (unmap) {
bus_dmamap_sync(sc->sc_rdmatag, rd->hrx_dmamap,
BUS_DMASYNC_POSTREAD);
bus_dmamap_unload(sc->sc_rdmatag, rd->hrx_dmamap);
}
map = rd->hrx_dmamap;
rd->hrx_dmamap = sc->sc_rb.rb_spare_dmamap;
sc->sc_rb.rb_spare_dmamap = map;
bus_dmamap_sync(sc->sc_rdmatag, rd->hrx_dmamap, BUS_DMASYNC_PREREAD);
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
HME_XD_SETADDR(sc->sc_flags & HME_PCI, sc->sc_rb.rb_rxd, ri,
segs[0].ds_addr);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
rd->hrx_m = m;
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
HME_XD_SETFLAGS(sc->sc_flags & HME_PCI, sc->sc_rb.rb_rxd, ri,
HME_XD_OWN | HME_XD_ENCODE_RSIZE(HME_DESC_RXLEN(sc, rd)));
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
return (0);
}
static int
hme_meminit(struct hme_softc *sc)
{
struct hme_ring *hr = &sc->sc_rb;
struct hme_txdesc *td;
bus_addr_t dma;
caddr_t p;
unsigned int i;
int error;
p = hr->rb_membase;
dma = hr->rb_dmabase;
/*
* Allocate transmit descriptors
*/
hr->rb_txd = p;
hr->rb_txddma = dma;
p += HME_NTXDESC * HME_XD_SIZE;
dma += HME_NTXDESC * HME_XD_SIZE;
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
/*
* We have reserved descriptor space until the next 2048 byte
* boundary.
*/
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
dma = (bus_addr_t)roundup((u_long)dma, 2048);
p = (caddr_t)roundup((u_long)p, 2048);
/*
* Allocate receive descriptors
*/
hr->rb_rxd = p;
hr->rb_rxddma = dma;
p += HME_NRXDESC * HME_XD_SIZE;
dma += HME_NRXDESC * HME_XD_SIZE;
/* Again move forward to the next 2048 byte boundary.*/
dma = (bus_addr_t)roundup((u_long)dma, 2048);
p = (caddr_t)roundup((u_long)p, 2048);
/*
* Initialize transmit buffer descriptors
*/
for (i = 0; i < HME_NTXDESC; i++) {
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
HME_XD_SETADDR(sc->sc_flags & HME_PCI, hr->rb_txd, i, 0);
HME_XD_SETFLAGS(sc->sc_flags & HME_PCI, hr->rb_txd, i, 0);
Convert to use bus_dmamap_load_mbuf(); decouple software TX descriptors from the hardware descriptors to avoid the overhead of having a DMA map for each of them. Bump the number of hardware descriptors to 128, and use half as many software descriptors for now. Some minor cleanups.
2003-01-06 22:12:57 +00:00
}
STAILQ_INIT(&sc->sc_rb.rb_txfreeq);
STAILQ_INIT(&sc->sc_rb.rb_txbusyq);
for (i = 0; i < HME_NTXQ; i++) {
td = &sc->sc_rb.rb_txdesc[i];
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
if (td->htx_m != NULL) {
bus_dmamap_sync() overhaul: - Remove NetBSD-style or-ed together BUS_DMASYNC operations, in some cases relaxing the (intended) syncing operation a bit. - Add syncs before reading the descriptor rings. - Try to combine syncs where possible to avoid overhead. - Sync all maps before unloading them.
2003-01-21 17:22:52 +00:00
bus_dmamap_sync(sc->sc_tdmatag, td->htx_dmamap,
BUS_DMASYNC_POSTWRITE);
Convert to use bus_dmamap_load_mbuf(); decouple software TX descriptors from the hardware descriptors to avoid the overhead of having a DMA map for each of them. Bump the number of hardware descriptors to 128, and use half as many software descriptors for now. Some minor cleanups.
2003-01-06 22:12:57 +00:00
bus_dmamap_unload(sc->sc_tdmatag, td->htx_dmamap);
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
m_freem(td->htx_m);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
td->htx_m = NULL;
}
Convert to use bus_dmamap_load_mbuf(); decouple software TX descriptors from the hardware descriptors to avoid the overhead of having a DMA map for each of them. Bump the number of hardware descriptors to 128, and use half as many software descriptors for now. Some minor cleanups.
2003-01-06 22:12:57 +00:00
STAILQ_INSERT_TAIL(&sc->sc_rb.rb_txfreeq, td, htx_q);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
}
/*
* Initialize receive buffer descriptors
*/
for (i = 0; i < HME_NRXDESC; i++) {
error = hme_add_rxbuf(sc, i, 1);
if (error != 0)
return (error);
}
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
bus_dmamap_sync(sc->sc_cdmatag, sc->sc_cdmamap,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
Miscellaneous fixes: - always reinitialize the rx descriptors, even if the mbuf is kept. This should fix the hangs on ifconfig that were observed - on an rx overflow, reinitialize the descriptor so that the interface will not hang - correct some bus_dmamap_sync() calls - correct some debug messages - some minor nits
2002-07-14 12:09:48 +00:00
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
hr->rb_tdhead = hr->rb_tdtail = 0;
hr->rb_td_nbusy = 0;
hr->rb_rdtail = 0;
Miscellaneous fixes: - always reinitialize the rx descriptors, even if the mbuf is kept. This should fix the hangs on ifconfig that were observed - on an rx overflow, reinitialize the descriptor so that the interface will not hang - correct some bus_dmamap_sync() calls - correct some debug messages - some minor nits
2002-07-14 12:09:48 +00:00
CTR2(KTR_HME, "hme_meminit: tx ring va %p, pa %#lx", hr->rb_txd,
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
hr->rb_txddma);
Miscellaneous fixes: - always reinitialize the rx descriptors, even if the mbuf is kept. This should fix the hangs on ifconfig that were observed - on an rx overflow, reinitialize the descriptor so that the interface will not hang - correct some bus_dmamap_sync() calls - correct some debug messages - some minor nits
2002-07-14 12:09:48 +00:00
CTR2(KTR_HME, "hme_meminit: rx ring va %p, pa %#lx", hr->rb_rxd,
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
hr->rb_rxddma);
CTR2(KTR_HME, "rx entry 1: flags %x, address %x",
*(u_int32_t *)hr->rb_rxd, *(u_int32_t *)(hr->rb_rxd + 4));
CTR2(KTR_HME, "tx entry 1: flags %x, address %x",
*(u_int32_t *)hr->rb_txd, *(u_int32_t *)(hr->rb_txd + 4));
return (0);
}
static int
hme_mac_bitflip(struct hme_softc *sc, u_int32_t reg, u_int32_t val,
u_int32_t clr, u_int32_t set)
{
int i = 0;
val &= ~clr;
val |= set;
HME_MAC_WRITE_4(sc, reg, val);
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
HME_MAC_BARRIER(sc, reg, 4,
BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
if (clr == 0 && set == 0)
return (1); /* just write, no bits to wait for */
do {
DELAY(100);
i++;
val = HME_MAC_READ_4(sc, reg);
if (i > 40) {
/* After 3.5ms, we should have been done. */
device_printf(sc->sc_dev, "timeout while writing to "
"MAC configuration register\n");
return (0);
}
} while ((val & clr) != 0 && (val & set) != set);
return (1);
}
/*
* Initialization of interface; set up initialization block
* and transmit/receive descriptor rings.
*/
static void
hme_init(void *xsc)
Make hme(4) mpsafe - Let hme_start()/hme_init() acquire lock and then call hme_start_locked()/hme_init_locked() respectivly. - Teardown interrupt handler before hme_detach(). - Remove IFF_NEEDSGIANT flag and mark interrupt handler INTR_MPSAFE. - Set callout handler to CALLOUT_MPSAFE. - Add locks in hme MII interface. Reviewed by: jake Tested by: Julian C. Dunn <jdunn at opentrend dot net> MFC after: 2 weeks
2004-11-22 06:46:30 +00:00
{
struct hme_softc *sc = (struct hme_softc *)xsc;
HME_LOCK(sc);
hme_init_locked(sc);
HME_UNLOCK(sc);
}
static void
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
hme_init_locked(struct hme_softc *sc)
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
{
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
struct ifnet *ifp = sc->sc_ifp;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
u_int8_t *ea;
Implement TCP/UDP Transmit/Receive checksum offload. Since HME doesn't compensate the checksum for UDP datagram which can yield to 0x0, UDP transmit checksum offload is disabled by default. The UDP Transmit checksum offload can be reactivated by setting special link option link0 with ifconfig(8). Approved by: jake (mentor) Reviewed by: tmm Tested by: Herve Boulouis <amon@sockar.homeip.net>
2004-08-05 02:52:33 +00:00
u_int32_t n, v;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
Make hme(4) mpsafe - Let hme_start()/hme_init() acquire lock and then call hme_start_locked()/hme_init_locked() respectivly. - Teardown interrupt handler before hme_detach(). - Remove IFF_NEEDSGIANT flag and mark interrupt handler INTR_MPSAFE. - Set callout handler to CALLOUT_MPSAFE. - Add locks in hme MII interface. Reviewed by: jake Tested by: Julian C. Dunn <jdunn at opentrend dot net> MFC after: 2 weeks
2004-11-22 06:46:30 +00:00
HME_LOCK_ASSERT(sc, MA_OWNED);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
/*
* Initialization sequence. The numbered steps below correspond
* to the sequence outlined in section 6.3.5.1 in the Ethernet
* Channel Engine manual (part of the PCIO manual).
* See also the STP2002-STQ document from Sun Microsystems.
*/
/* step 1 & 2. Reset the Ethernet Channel */
hme_stop(sc);
/* Re-initialize the MIF */
hme_mifinit(sc);
#if 0
/* Mask all MIF interrupts, just in case */
HME_MIF_WRITE_4(sc, HME_MIFI_IMASK, 0xffff);
#endif
/* step 3. Setup data structures in host memory */
if (hme_meminit(sc) != 0) {
device_printf(sc->sc_dev, "out of buffers; init aborted.");
return;
}
/* step 4. TX MAC registers & counters */
HME_MAC_WRITE_4(sc, HME_MACI_NCCNT, 0);
HME_MAC_WRITE_4(sc, HME_MACI_FCCNT, 0);
HME_MAC_WRITE_4(sc, HME_MACI_EXCNT, 0);
HME_MAC_WRITE_4(sc, HME_MACI_LTCNT, 0);
The Sun hme hardware supposedly supports Tx frames up to 65535 octets, and Rx frames up to 8191 octets, so it is perfectly capable of supporting vlan(4)-style VLAN natively. Thus, make it support VLAN `oversize' frames. Reviewed by: tmm
2004-05-06 13:38:19 +00:00
HME_MAC_WRITE_4(sc, HME_MACI_TXSIZE, HME_MAX_FRAMESIZE);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
/* Load station MAC address */
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
ea = IF_LLADDR(ifp);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
HME_MAC_WRITE_4(sc, HME_MACI_MACADDR0, (ea[0] << 8) | ea[1]);
HME_MAC_WRITE_4(sc, HME_MACI_MACADDR1, (ea[2] << 8) | ea[3]);
HME_MAC_WRITE_4(sc, HME_MACI_MACADDR2, (ea[4] << 8) | ea[5]);
/*
* Init seed for backoff
* (source suggested by manual: low 10 bits of MAC address)
*/
v = ((ea[4] << 8) | ea[5]) & 0x3fff;
HME_MAC_WRITE_4(sc, HME_MACI_RANDSEED, v);
/* Note: Accepting power-on default for other MAC registers here.. */
/* step 5. RX MAC registers & counters */
hme_setladrf(sc, 0);
/* step 6 & 7. Program Descriptor Ring Base Addresses */
HME_ETX_WRITE_4(sc, HME_ETXI_RING, sc->sc_rb.rb_txddma);
/* Transmit Descriptor ring size: in increments of 16 */
HME_ETX_WRITE_4(sc, HME_ETXI_RSIZE, HME_NTXDESC / 16 - 1);
HME_ERX_WRITE_4(sc, HME_ERXI_RING, sc->sc_rb.rb_rxddma);
The Sun hme hardware supposedly supports Tx frames up to 65535 octets, and Rx frames up to 8191 octets, so it is perfectly capable of supporting vlan(4)-style VLAN natively. Thus, make it support VLAN `oversize' frames. Reviewed by: tmm
2004-05-06 13:38:19 +00:00
HME_MAC_WRITE_4(sc, HME_MACI_RXSIZE, HME_MAX_FRAMESIZE);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
/* step 8. Global Configuration & Interrupt Mask */
HME_SEB_WRITE_4(sc, HME_SEBI_IMASK,
~(/*HME_SEB_STAT_GOTFRAME | HME_SEB_STAT_SENTFRAME |*/
HME_SEB_STAT_HOSTTOTX |
HME_SEB_STAT_RXTOHOST |
HME_SEB_STAT_TXALL |
HME_SEB_STAT_TXPERR |
HME_SEB_STAT_RCNTEXP |
HME_SEB_STAT_ALL_ERRORS ));
switch (sc->sc_burst) {
default:
v = 0;
break;
case 16:
v = HME_SEB_CFG_BURST16;
break;
case 32:
v = HME_SEB_CFG_BURST32;
break;
case 64:
v = HME_SEB_CFG_BURST64;
break;
}
Implement TCP/UDP Transmit/Receive checksum offload. Since HME doesn't compensate the checksum for UDP datagram which can yield to 0x0, UDP transmit checksum offload is disabled by default. The UDP Transmit checksum offload can be reactivated by setting special link option link0 with ifconfig(8). Approved by: jake (mentor) Reviewed by: tmm Tested by: Herve Boulouis <amon@sockar.homeip.net>
2004-08-05 02:52:33 +00:00
/*
* Blindly setting 64bit transfers may hang PCI cards(Cheerio?).
* Allowing 64bit transfers breaks TX checksum offload as well.
* Don't know this comes from hardware bug or driver's DMAing
* scheme.
*
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
* if (sc->sc_flags & HME_PCI == 0)
* v |= HME_SEB_CFG_64BIT;
Implement TCP/UDP Transmit/Receive checksum offload. Since HME doesn't compensate the checksum for UDP datagram which can yield to 0x0, UDP transmit checksum offload is disabled by default. The UDP Transmit checksum offload can be reactivated by setting special link option link0 with ifconfig(8). Approved by: jake (mentor) Reviewed by: tmm Tested by: Herve Boulouis <amon@sockar.homeip.net>
2004-08-05 02:52:33 +00:00
*/
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
HME_SEB_WRITE_4(sc, HME_SEBI_CFG, v);
/* step 9. ETX Configuration: use mostly default values */
/* Enable DMA */
v = HME_ETX_READ_4(sc, HME_ETXI_CFG);
v |= HME_ETX_CFG_DMAENABLE;
HME_ETX_WRITE_4(sc, HME_ETXI_CFG, v);
/* step 10. ERX Configuration */
v = HME_ERX_READ_4(sc, HME_ERXI_CFG);
/* Encode Receive Descriptor ring size: four possible values */
v &= ~HME_ERX_CFG_RINGSIZEMSK;
switch (HME_NRXDESC) {
case 32:
v |= HME_ERX_CFG_RINGSIZE32;
break;
case 64:
v |= HME_ERX_CFG_RINGSIZE64;
break;
case 128:
v |= HME_ERX_CFG_RINGSIZE128;
break;
case 256:
v |= HME_ERX_CFG_RINGSIZE256;
break;
default:
printf("hme: invalid Receive Descriptor ring size\n");
break;
}
Convert to use bus_dmamap_load_mbuf(); decouple software TX descriptors from the hardware descriptors to avoid the overhead of having a DMA map for each of them. Bump the number of hardware descriptors to 128, and use half as many software descriptors for now. Some minor cleanups.
2003-01-06 22:12:57 +00:00
/* Enable DMA, fix RX first byte offset. */
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
v &= ~HME_ERX_CFG_FBO_MASK;
Convert to use bus_dmamap_load_mbuf(); decouple software TX descriptors from the hardware descriptors to avoid the overhead of having a DMA map for each of them. Bump the number of hardware descriptors to 128, and use half as many software descriptors for now. Some minor cleanups.
2003-01-06 22:12:57 +00:00
v |= HME_ERX_CFG_DMAENABLE | (HME_RXOFFS << HME_ERX_CFG_FBO_SHIFT);
Implement TCP/UDP Transmit/Receive checksum offload. Since HME doesn't compensate the checksum for UDP datagram which can yield to 0x0, UDP transmit checksum offload is disabled by default. The UDP Transmit checksum offload can be reactivated by setting special link option link0 with ifconfig(8). Approved by: jake (mentor) Reviewed by: tmm Tested by: Herve Boulouis <amon@sockar.homeip.net>
2004-08-05 02:52:33 +00:00
/* RX TCP/UDP checksum offset */
n = (ETHER_HDR_LEN + sizeof(struct ip)) / 2;
n = (n << HME_ERX_CFG_CSUMSTART_SHIFT) & HME_ERX_CFG_CSUMSTART_MASK;
v |= n;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
CTR1(KTR_HME, "hme_init: programming ERX_CFG to %x", (u_int)v);
HME_ERX_WRITE_4(sc, HME_ERXI_CFG, v);
/* step 11. XIF Configuration */
v = HME_MAC_READ_4(sc, HME_MACI_XIF);
v |= HME_MAC_XIF_OE;
CTR1(KTR_HME, "hme_init: programming XIF to %x", (u_int)v);
HME_MAC_WRITE_4(sc, HME_MACI_XIF, v);
/* step 12. RX_MAC Configuration Register */
v = HME_MAC_READ_4(sc, HME_MACI_RXCFG);
v |= HME_MAC_RXCFG_ENABLE;
v &= ~(HME_MAC_RXCFG_DCRCS);
CTR1(KTR_HME, "hme_init: programming RX_MAC to %x", (u_int)v);
HME_MAC_WRITE_4(sc, HME_MACI_RXCFG, v);
/* step 13. TX_MAC Configuration Register */
v = HME_MAC_READ_4(sc, HME_MACI_TXCFG);
v |= (HME_MAC_TXCFG_ENABLE | HME_MAC_TXCFG_DGIVEUP);
CTR1(KTR_HME, "hme_init: programming TX_MAC to %x", (u_int)v);
HME_MAC_WRITE_4(sc, HME_MACI_TXCFG, v);
/* step 14. Issue Transmit Pending command */
#ifdef HMEDEBUG
/* Debug: double-check. */
CTR4(KTR_HME, "hme_init: tx ring %#x, rsz %#x, rx ring %#x, "
"rxsize %#x", HME_ETX_READ_4(sc, HME_ETXI_RING),
HME_ETX_READ_4(sc, HME_ETXI_RSIZE),
HME_ERX_READ_4(sc, HME_ERXI_RING),
HME_MAC_READ_4(sc, HME_MACI_RXSIZE));
CTR3(KTR_HME, "hme_init: intr mask %#x, erx cfg %#x, etx cfg %#x",
HME_SEB_READ_4(sc, HME_SEBI_IMASK),
HME_ERX_READ_4(sc, HME_ERXI_CFG),
HME_ETX_READ_4(sc, HME_ETXI_CFG));
CTR2(KTR_HME, "hme_init: mac rxcfg %#x, maci txcfg %#x",
HME_MAC_READ_4(sc, HME_MACI_RXCFG),
HME_MAC_READ_4(sc, HME_MACI_TXCFG));
#endif
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
ifp->if_drv_flags |= IFF_DRV_RUNNING;
ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
In hme_init() call mii_mediachg() to make sure the current media is set. This is part 2/2 of fixing autonegotiation on hme(4) using DP83840A PHYs. It appears to also fix the occasional problems to establish a link on hme(4) using LU6612 PHYs and shouldn't hurt on those using QS6612 PHYs. Obtained from: NetBSD
2004-05-29 18:29:53 +00:00
/* Set the current media. */
- In hme_stop() mask all interrupts. - In hme_eint() print MIF register contents on MIF interrupts. - In hme_mifinit() don't bother to preserve the previous MIF config. This was mainly done in order to preserve the PHY select bit (external or internal PHY) but which only needs to be set as appropriate when reading from or writing to the desired PHY in hme_mii_{read,write}reg(). Similarly don't bother to set the PHY select bit in hme_mii_statchg(). - In hme_mii_{read,write}reg() ignore requests to PHYs other than the external and internal PHY one. - Move enabling/disabling the MII drivers of the external transceiver from hme_init_locked() and based on the sheer presence of an external to hme_mifinit() and based on the currently selected media, defaulting to the internal transceiver when the media hasn't been set, yet. Invoke hme_mifinit() from the newly added hme_mediachange_locked() so the setting of the MII drivers is updated when changing media. These changes keep the MII bus from wedging (which manifests in the HME and the PHYs no longer being able to communicate with each other) when the PHY device drivers isolate the unused PHY in two-PHY configurations as present in f.e. Netra t1 100 while changing media, either from hme_init_locked() (see also below) or via ifconfig(8). They also allow for using both transceivers/PHYs. - In the newly added hme_mediachange_locked() also reset the PHYs in two- PHY configurations before invoking mii_mediachg(). This is required for successfully unisolating the previously unused PHY when switching between PHYs. - Now that changing media should no longer cause problems back out rev. 1.27 and re-enable setting the current media in hme_init_locked() (see the commit message of rev. 1.23 for more info). These changes are roughly a merge of NetBSD gem.c rev. 1.32 - 1.35 (1.30 was already fixed differently in our 1.36; 1.31 and 1.32 were wrong) with some parts reworked and things that don't make sense like setting the MII drivers and restoring the previous MIF and XIF settings in hme_mii_{read, write}reg() omitted. MFC after: 2 weeks
2006-12-04 00:51:08 +00:00
hme_mediachange_locked(sc);
In hme_init() call mii_mediachg() to make sure the current media is set. This is part 2/2 of fixing autonegotiation on hme(4) using DP83840A PHYs. It appears to also fix the occasional problems to establish a link on hme(4) using LU6612 PHYs and shouldn't hurt on those using QS6612 PHYs. Obtained from: NetBSD
2004-05-29 18:29:53 +00:00
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
/* Start the one second timer. */
- Use the hme_tick() callout instead of if_slowtimo() for driving hme_watchdog() in order to avoid races accessing if_timer. - Use bus_get_dma_tag() so hme(4) works on platforms requiring it. - Don't bother to set if_mtu to ETHERMTU, ether_ifattach() does that.
2006-12-06 02:07:20 +00:00
sc->sc_wdog_timer = 0;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
callout_reset(&sc->sc_tick_ch, hz, hme_tick, sc);
Implement TCP/UDP Transmit/Receive checksum offload. Since HME doesn't compensate the checksum for UDP datagram which can yield to 0x0, UDP transmit checksum offload is disabled by default. The UDP Transmit checksum offload can be reactivated by setting special link option link0 with ifconfig(8). Approved by: jake (mentor) Reviewed by: tmm Tested by: Herve Boulouis <amon@sockar.homeip.net>
2004-08-05 02:52:33 +00:00
}
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
/*
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
* Routine to DMA map an mbuf chain, set up the descriptor rings
* accordingly and start the transmission.
* Returns 0 on success, -1 if there were not enough free descriptors
* to map the packet, or an errno otherwise.
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
*
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
* XXX: this relies on the fact that segments returned by
* bus_dmamap_load_mbuf_sg() are readable from the nearest burst
* boundary on (i.e. potentially before ds_addr) to the first
* boundary beyond the end. This is usually a safe assumption to
* make, but is not documented.
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
*/
static int
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
hme_load_txmbuf(struct hme_softc *sc, struct mbuf **m0)
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
{
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
bus_dma_segment_t segs[HME_NTXSEGS];
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
struct hme_txdesc *htx;
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
struct ip *ip;
struct mbuf *m;
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
caddr_t txd;
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
int error, i, nsegs, pci, ri, si;
uint32_t cflags, flags;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
if ((htx = STAILQ_FIRST(&sc->sc_rb.rb_txfreeq)) == NULL)
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
return (ENOBUFS);
cflags = 0;
if (((*m0)->m_pkthdr.csum_flags & sc->sc_csum_features) != 0) {
if (M_WRITABLE(*m0) == 0) {
m = m_dup(*m0, M_DONTWAIT);
m_freem(*m0);
*m0 = m;
if (m == NULL)
return (ENOBUFS);
}
i = sizeof(struct ether_header);
m = m_pullup(*m0, i + sizeof(struct ip));
if (m == NULL) {
*m0 = NULL;
return (ENOBUFS);
}
ip = (struct ip *)(mtod(m, caddr_t) + i);
i += (ip->ip_hl << 2);
cflags = i << HME_XD_TXCKSUM_SSHIFT |
((i + m->m_pkthdr.csum_data) << HME_XD_TXCKSUM_OSHIFT) |
HME_XD_TXCKSUM;
*m0 = m;
}
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
error = bus_dmamap_load_mbuf_sg(sc->sc_tdmatag, htx->htx_dmamap,
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
*m0, segs, &nsegs, 0);
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
if (error == EFBIG) {
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
m = m_collapse(*m0, M_DONTWAIT, HME_NTXSEGS);
Fix invalid reference of mbuf chains. Use proper pointer dereference to inform modified mbuf chains to caller. While I'm here perform checksum offload setup after loading DMA maps as m_defrag(9) can return new mbuf chains. In collaboration with: glebius
2006-08-12 01:19:37 +00:00
if (m == NULL) {
m_freem(*m0);
*m0 = NULL;
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
return (ENOMEM);
}
Fix invalid reference of mbuf chains. Use proper pointer dereference to inform modified mbuf chains to caller. While I'm here perform checksum offload setup after loading DMA maps as m_defrag(9) can return new mbuf chains. In collaboration with: glebius
2006-08-12 01:19:37 +00:00
*m0 = m;
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
error = bus_dmamap_load_mbuf_sg(sc->sc_tdmatag, htx->htx_dmamap,
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
*m0, segs, &nsegs, 0);
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
if (error != 0) {
Fix invalid reference of mbuf chains. Use proper pointer dereference to inform modified mbuf chains to caller. While I'm here perform checksum offload setup after loading DMA maps as m_defrag(9) can return new mbuf chains. In collaboration with: glebius
2006-08-12 01:19:37 +00:00
m_freem(*m0);
*m0 = NULL;
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
return (error);
}
} else if (error != 0)
return (error);
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
/* If nsegs is wrong then the stack is corrupt. */
KASSERT(nsegs <= HME_NTXSEGS,
("%s: too many DMA segments (%d)", __func__, nsegs));
if (nsegs == 0) {
Fix invalid reference of mbuf chains. Use proper pointer dereference to inform modified mbuf chains to caller. While I'm here perform checksum offload setup after loading DMA maps as m_defrag(9) can return new mbuf chains. In collaboration with: glebius
2006-08-12 01:19:37 +00:00
m_freem(*m0);
*m0 = NULL;
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
return (EIO);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
}
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
if (sc->sc_rb.rb_td_nbusy + nsegs >= HME_NTXDESC) {
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
bus_dmamap_unload(sc->sc_tdmatag, htx->htx_dmamap);
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
/* Retry with m_collapse(9)? */
return (ENOBUFS);
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
}
bus_dmamap_sync(sc->sc_tdmatag, htx->htx_dmamap, BUS_DMASYNC_PREWRITE);
si = ri = sc->sc_rb.rb_tdhead;
txd = sc->sc_rb.rb_txd;
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
pci = sc->sc_flags & HME_PCI;
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
CTR2(KTR_HME, "hme_load_mbuf: next desc is %d (%#x)", ri,
HME_XD_GETFLAGS(pci, txd, ri));
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
for (i = 0; i < nsegs; i++) {
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
/* Fill the ring entry. */
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
flags = HME_XD_ENCODE_TSIZE(segs[i].ds_len);
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
if (i == 0)
flags |= HME_XD_SOP | cflags;
else
flags |= HME_XD_OWN | cflags;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
CTR3(KTR_HME, "hme_load_mbuf: activating ri %d, si %d (%#x)",
ri, si, flags);
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
HME_XD_SETADDR(pci, txd, ri, segs[i].ds_addr);
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
HME_XD_SETFLAGS(pci, txd, ri, flags);
sc->sc_rb.rb_td_nbusy++;
htx->htx_lastdesc = ri;
ri = (ri + 1) % HME_NTXDESC;
}
sc->sc_rb.rb_tdhead = ri;
/* set EOP on the last descriptor */
ri = (ri + HME_NTXDESC - 1) % HME_NTXDESC;
flags = HME_XD_GETFLAGS(pci, txd, ri);
flags |= HME_XD_EOP;
CTR3(KTR_HME, "hme_load_mbuf: setting EOP ri %d, si %d (%#x)", ri, si,
flags);
HME_XD_SETFLAGS(pci, txd, ri, flags);
/* Turn the first descriptor ownership to the hme */
flags = HME_XD_GETFLAGS(pci, txd, si);
flags |= HME_XD_OWN;
CTR2(KTR_HME, "hme_load_mbuf: setting OWN for 1st desc ri %d, (%#x)",
ri, flags);
HME_XD_SETFLAGS(pci, txd, si, flags);
STAILQ_REMOVE_HEAD(&sc->sc_rb.rb_txfreeq, htx_q);
STAILQ_INSERT_TAIL(&sc->sc_rb.rb_txbusyq, htx, htx_q);
Fix invalid reference of mbuf chains. Use proper pointer dereference to inform modified mbuf chains to caller. While I'm here perform checksum offload setup after loading DMA maps as m_defrag(9) can return new mbuf chains. In collaboration with: glebius
2006-08-12 01:19:37 +00:00
htx->htx_m = *m0;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
/* start the transmission. */
HME_ETX_WRITE_4(sc, HME_ETXI_PENDING, HME_ETX_TP_DMAWAKEUP);
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
return (0);
}
/*
* Pass a packet to the higher levels.
*/
static void
Implement TCP/UDP Transmit/Receive checksum offload. Since HME doesn't compensate the checksum for UDP datagram which can yield to 0x0, UDP transmit checksum offload is disabled by default. The UDP Transmit checksum offload can be reactivated by setting special link option link0 with ifconfig(8). Approved by: jake (mentor) Reviewed by: tmm Tested by: Herve Boulouis <amon@sockar.homeip.net>
2004-08-05 02:52:33 +00:00
hme_read(struct hme_softc *sc, int ix, int len, u_int32_t flags)
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
{
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
struct ifnet *ifp = sc->sc_ifp;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
struct mbuf *m;
if (len <= sizeof(struct ether_header) ||
The Sun hme hardware supposedly supports Tx frames up to 65535 octets, and Rx frames up to 8191 octets, so it is perfectly capable of supporting vlan(4)-style VLAN natively. Thus, make it support VLAN `oversize' frames. Reviewed by: tmm
2004-05-06 13:38:19 +00:00
len > HME_MAX_FRAMESIZE) {
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
#ifdef HMEDEBUG
HME_WHINE(sc->sc_dev, "invalid packet size %d; dropping\n",
len);
#endif
Miscellaneous fixes: - always reinitialize the rx descriptors, even if the mbuf is kept. This should fix the hangs on ifconfig that were observed - on an rx overflow, reinitialize the descriptor so that the interface will not hang - correct some bus_dmamap_sync() calls - correct some debug messages - some minor nits
2002-07-14 12:09:48 +00:00
ifp->if_ierrors++;
bus_dmamap_sync() overhaul: - Remove NetBSD-style or-ed together BUS_DMASYNC operations, in some cases relaxing the (intended) syncing operation a bit. - Add syncs before reading the descriptor rings. - Try to combine syncs where possible to avoid overhead. - Sync all maps before unloading them.
2003-01-21 17:22:52 +00:00
hme_discard_rxbuf(sc, ix);
Miscellaneous fixes: - always reinitialize the rx descriptors, even if the mbuf is kept. This should fix the hangs on ifconfig that were observed - on an rx overflow, reinitialize the descriptor so that the interface will not hang - correct some bus_dmamap_sync() calls - correct some debug messages - some minor nits
2002-07-14 12:09:48 +00:00
return;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
}
m = sc->sc_rb.rb_rxdesc[ix].hrx_m;
Convert to use bus_dmamap_load_mbuf(); decouple software TX descriptors from the hardware descriptors to avoid the overhead of having a DMA map for each of them. Bump the number of hardware descriptors to 128, and use half as many software descriptors for now. Some minor cleanups.
2003-01-06 22:12:57 +00:00
CTR1(KTR_HME, "hme_read: len %d", len);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
if (hme_add_rxbuf(sc, ix, 0) != 0) {
/*
* hme_add_rxbuf will leave the old buffer in the ring until
* it is sure that a new buffer can be mapped. If it can not,
* drop the packet, but leave the interface up.
*/
Miscellaneous fixes: - always reinitialize the rx descriptors, even if the mbuf is kept. This should fix the hangs on ifconfig that were observed - on an rx overflow, reinitialize the descriptor so that the interface will not hang - correct some bus_dmamap_sync() calls - correct some debug messages - some minor nits
2002-07-14 12:09:48 +00:00
ifp->if_iqdrops++;
bus_dmamap_sync() overhaul: - Remove NetBSD-style or-ed together BUS_DMASYNC operations, in some cases relaxing the (intended) syncing operation a bit. - Add syncs before reading the descriptor rings. - Try to combine syncs where possible to avoid overhead. - Sync all maps before unloading them.
2003-01-21 17:22:52 +00:00
hme_discard_rxbuf(sc, ix);
Miscellaneous fixes: - always reinitialize the rx descriptors, even if the mbuf is kept. This should fix the hangs on ifconfig that were observed - on an rx overflow, reinitialize the descriptor so that the interface will not hang - correct some bus_dmamap_sync() calls - correct some debug messages - some minor nits
2002-07-14 12:09:48 +00:00
return;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
}
ifp->if_ipackets++;
m->m_pkthdr.rcvif = ifp;
Convert to use bus_dmamap_load_mbuf(); decouple software TX descriptors from the hardware descriptors to avoid the overhead of having a DMA map for each of them. Bump the number of hardware descriptors to 128, and use half as many software descriptors for now. Some minor cleanups.
2003-01-06 22:12:57 +00:00
m->m_pkthdr.len = m->m_len = len + HME_RXOFFS;
m_adj(m, HME_RXOFFS);
Implement TCP/UDP Transmit/Receive checksum offload. Since HME doesn't compensate the checksum for UDP datagram which can yield to 0x0, UDP transmit checksum offload is disabled by default. The UDP Transmit checksum offload can be reactivated by setting special link option link0 with ifconfig(8). Approved by: jake (mentor) Reviewed by: tmm Tested by: Herve Boulouis <amon@sockar.homeip.net>
2004-08-05 02:52:33 +00:00
/* RX TCP/UDP checksum */
if (ifp->if_capenable & IFCAP_RXCSUM)
hme_rxcksum(m, flags);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
/* Pass the packet up. */
Make hme(4) mpsafe - Let hme_start()/hme_init() acquire lock and then call hme_start_locked()/hme_init_locked() respectivly. - Teardown interrupt handler before hme_detach(). - Remove IFF_NEEDSGIANT flag and mark interrupt handler INTR_MPSAFE. - Set callout handler to CALLOUT_MPSAFE. - Add locks in hme MII interface. Reviewed by: jake Tested by: Julian C. Dunn <jdunn at opentrend dot net> MFC after: 2 weeks
2004-11-22 06:46:30 +00:00
HME_UNLOCK(sc);
network interface driver changes: o don't strip the Ethernet header from inbound packets; pass packets up the stack intact (required significant changes to some drivers) o reference common definitions in net/ethernet.h (e.g. ETHER_ALIGN) o track ether_ifattach/ether_ifdetach API changes o track bpf changes (use BPF_TAP and BPF_MTAP) o track vlan changes (ifnet capabilities, revised processing scheme, etc.) o use if_input to pass packets "up" o call ether_ioctl for default handling of ioctls Reviewed by: many Approved by: re
2002-11-14 23:54:55 +00:00
(*ifp->if_input)(ifp, m);
Make hme(4) mpsafe - Let hme_start()/hme_init() acquire lock and then call hme_start_locked()/hme_init_locked() respectivly. - Teardown interrupt handler before hme_detach(). - Remove IFF_NEEDSGIANT flag and mark interrupt handler INTR_MPSAFE. - Set callout handler to CALLOUT_MPSAFE. - Add locks in hme MII interface. Reviewed by: jake Tested by: Julian C. Dunn <jdunn at opentrend dot net> MFC after: 2 weeks
2004-11-22 06:46:30 +00:00
HME_LOCK(sc);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
}
static void
hme_start(struct ifnet *ifp)
Make hme(4) mpsafe - Let hme_start()/hme_init() acquire lock and then call hme_start_locked()/hme_init_locked() respectivly. - Teardown interrupt handler before hme_detach(). - Remove IFF_NEEDSGIANT flag and mark interrupt handler INTR_MPSAFE. - Set callout handler to CALLOUT_MPSAFE. - Add locks in hme MII interface. Reviewed by: jake Tested by: Julian C. Dunn <jdunn at opentrend dot net> MFC after: 2 weeks
2004-11-22 06:46:30 +00:00
{
struct hme_softc *sc = ifp->if_softc;
HME_LOCK(sc);
hme_start_locked(ifp);
HME_UNLOCK(sc);
}
static void
hme_start_locked(struct ifnet *ifp)
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
{
struct hme_softc *sc = (struct hme_softc *)ifp->if_softc;
struct mbuf *m;
int error, enq = 0;
Propagate rename of IFF_OACTIVE and IFF_RUNNING to IFF_DRV_OACTIVE and IFF_DRV_RUNNING, as well as the move from ifnet.if_flags to ifnet.if_drv_flags. Device drivers are now responsible for synchronizing access to these flags, as they are in if_drv_flags. This helps prevent races between the network stack and device driver in maintaining the interface flags field. Many __FreeBSD__ and __FreeBSD_version checks maintained and continued; some less so. Reviewed by: pjd, bz MFC after: 7 days
2005-08-09 10:20:02 +00:00
if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
IFF_DRV_RUNNING || (sc->sc_flags & HME_LINK) == 0)
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
return;
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
for (; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) &&
sc->sc_rb.rb_td_nbusy < HME_NTXDESC - 1;) {
Second part of ALTQ driver modifications, covering: an(4), ath(4), hme(4), ndis(4), vr(4) and wi(4) Please help testing: http://people.freebsd.org/~mlaier/ALTQ_driver/ Tested by: Vaidas Damosevicius (an, ath, wi) Roman Divacky (vr) Submitted by: yongari (hme)
2004-08-01 23:58:04 +00:00
IFQ_DRV_DEQUEUE(&ifp->if_snd, m);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
if (m == NULL)
break;
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
error = hme_load_txmbuf(sc, &m);
if (error != 0) {
if (m == NULL)
break;
Propagate rename of IFF_OACTIVE and IFF_RUNNING to IFF_DRV_OACTIVE and IFF_DRV_RUNNING, as well as the move from ifnet.if_flags to ifnet.if_drv_flags. Device drivers are now responsible for synchronizing access to these flags, as they are in if_drv_flags. This helps prevent races between the network stack and device driver in maintaining the interface flags field. Many __FreeBSD__ and __FreeBSD_version checks maintained and continued; some less so. Reviewed by: pjd, bz MFC after: 7 days
2005-08-09 10:20:02 +00:00
ifp->if_drv_flags |= IFF_DRV_OACTIVE;
Second part of ALTQ driver modifications, covering: an(4), ath(4), hme(4), ndis(4), vr(4) and wi(4) Please help testing: http://people.freebsd.org/~mlaier/ALTQ_driver/ Tested by: Vaidas Damosevicius (an, ath, wi) Roman Divacky (vr) Submitted by: yongari (hme)
2004-08-01 23:58:04 +00:00
IFQ_DRV_PREPEND(&ifp->if_snd, m);
Fix some bugs in the handling of the case of not enough tx descriptors being free to send a packet.
2002-03-09 21:50:25 +00:00
break;
Call bpf_mtap() on output, to catch outgoing packets for e.g. tcpdump .
2002-07-30 21:47:14 +00:00
}
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
enq++;
BPF_MTAP(ifp, m);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
}
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
if (enq > 0) {
bus_dmamap_sync() overhaul: - Remove NetBSD-style or-ed together BUS_DMASYNC operations, in some cases relaxing the (intended) syncing operation a bit. - Add syncs before reading the descriptor rings. - Try to combine syncs where possible to avoid overhead. - Sync all maps before unloading them.
2003-01-21 17:22:52 +00:00
bus_dmamap_sync(sc->sc_cdmatag, sc->sc_cdmamap,
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
- Use the hme_tick() callout instead of if_slowtimo() for driving hme_watchdog() in order to avoid races accessing if_timer. - Use bus_get_dma_tag() so hme(4) works on platforms requiring it. - Don't bother to set if_mtu to ETHERMTU, ether_ifattach() does that.
2006-12-06 02:07:20 +00:00
sc->sc_wdog_timer = 5;
bus_dmamap_sync() overhaul: - Remove NetBSD-style or-ed together BUS_DMASYNC operations, in some cases relaxing the (intended) syncing operation a bit. - Add syncs before reading the descriptor rings. - Try to combine syncs where possible to avoid overhead. - Sync all maps before unloading them.
2003-01-21 17:22:52 +00:00
}
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
}
/*
* Transmit interrupt.
*/
static void
hme_tint(struct hme_softc *sc)
{
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
caddr_t txd;
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
struct ifnet *ifp = sc->sc_ifp;
Convert to use bus_dmamap_load_mbuf(); decouple software TX descriptors from the hardware descriptors to avoid the overhead of having a DMA map for each of them. Bump the number of hardware descriptors to 128, and use half as many software descriptors for now. Some minor cleanups.
2003-01-06 22:12:57 +00:00
struct hme_txdesc *htx;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
unsigned int ri, txflags;
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
txd = sc->sc_rb.rb_txd;
Convert to use bus_dmamap_load_mbuf(); decouple software TX descriptors from the hardware descriptors to avoid the overhead of having a DMA map for each of them. Bump the number of hardware descriptors to 128, and use half as many software descriptors for now. Some minor cleanups.
2003-01-06 22:12:57 +00:00
htx = STAILQ_FIRST(&sc->sc_rb.rb_txbusyq);
bus_dmamap_sync() overhaul: - Remove NetBSD-style or-ed together BUS_DMASYNC operations, in some cases relaxing the (intended) syncing operation a bit. - Add syncs before reading the descriptor rings. - Try to combine syncs where possible to avoid overhead. - Sync all maps before unloading them.
2003-01-21 17:22:52 +00:00
bus_dmamap_sync(sc->sc_cdmatag, sc->sc_cdmamap, BUS_DMASYNC_POSTREAD);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
/* Fetch current position in the transmit ring */
for (ri = sc->sc_rb.rb_tdtail;; ri = (ri + 1) % HME_NTXDESC) {
if (sc->sc_rb.rb_td_nbusy <= 0) {
CTR0(KTR_HME, "hme_tint: not busy!");
break;
}
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
txflags = HME_XD_GETFLAGS(sc->sc_flags & HME_PCI, txd, ri);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
CTR2(KTR_HME, "hme_tint: index %d, flags %#x", ri, txflags);
if ((txflags & HME_XD_OWN) != 0)
break;
Convert to use bus_dmamap_load_mbuf(); decouple software TX descriptors from the hardware descriptors to avoid the overhead of having a DMA map for each of them. Bump the number of hardware descriptors to 128, and use half as many software descriptors for now. Some minor cleanups.
2003-01-06 22:12:57 +00:00
CTR0(KTR_HME, "hme_tint: not owned");
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
--sc->sc_rb.rb_td_nbusy;
Propagate rename of IFF_OACTIVE and IFF_RUNNING to IFF_DRV_OACTIVE and IFF_DRV_RUNNING, as well as the move from ifnet.if_flags to ifnet.if_drv_flags. Device drivers are now responsible for synchronizing access to these flags, as they are in if_drv_flags. This helps prevent races between the network stack and device driver in maintaining the interface flags field. Many __FreeBSD__ and __FreeBSD_version checks maintained and continued; some less so. Reviewed by: pjd, bz MFC after: 7 days
2005-08-09 10:20:02 +00:00
ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
/* Complete packet transmitted? */
if ((txflags & HME_XD_EOP) == 0)
continue;
Convert to use bus_dmamap_load_mbuf(); decouple software TX descriptors from the hardware descriptors to avoid the overhead of having a DMA map for each of them. Bump the number of hardware descriptors to 128, and use half as many software descriptors for now. Some minor cleanups.
2003-01-06 22:12:57 +00:00
KASSERT(htx->htx_lastdesc == ri,
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
("%s: ring indices skewed: %d != %d!",
__func__, htx->htx_lastdesc, ri));
Convert to use bus_dmamap_load_mbuf(); decouple software TX descriptors from the hardware descriptors to avoid the overhead of having a DMA map for each of them. Bump the number of hardware descriptors to 128, and use half as many software descriptors for now. Some minor cleanups.
2003-01-06 22:12:57 +00:00
bus_dmamap_sync(sc->sc_tdmatag, htx->htx_dmamap,
BUS_DMASYNC_POSTWRITE);
bus_dmamap_unload(sc->sc_tdmatag, htx->htx_dmamap);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
ifp->if_opackets++;
Convert to use bus_dmamap_load_mbuf(); decouple software TX descriptors from the hardware descriptors to avoid the overhead of having a DMA map for each of them. Bump the number of hardware descriptors to 128, and use half as many software descriptors for now. Some minor cleanups.
2003-01-06 22:12:57 +00:00
m_freem(htx->htx_m);
htx->htx_m = NULL;
STAILQ_REMOVE_HEAD(&sc->sc_rb.rb_txbusyq, htx_q);
STAILQ_INSERT_TAIL(&sc->sc_rb.rb_txfreeq, htx, htx_q);
htx = STAILQ_FIRST(&sc->sc_rb.rb_txbusyq);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
}
- Use the hme_tick() callout instead of if_slowtimo() for driving hme_watchdog() in order to avoid races accessing if_timer. - Use bus_get_dma_tag() so hme(4) works on platforms requiring it. - Don't bother to set if_mtu to ETHERMTU, ether_ifattach() does that.
2006-12-06 02:07:20 +00:00
sc->sc_wdog_timer = sc->sc_rb.rb_td_nbusy > 0 ? 5 : 0;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
/* Update ring */
sc->sc_rb.rb_tdtail = ri;
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
hme_start_locked(ifp);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
}
Implement TCP/UDP Transmit/Receive checksum offload. Since HME doesn't compensate the checksum for UDP datagram which can yield to 0x0, UDP transmit checksum offload is disabled by default. The UDP Transmit checksum offload can be reactivated by setting special link option link0 with ifconfig(8). Approved by: jake (mentor) Reviewed by: tmm Tested by: Herve Boulouis <amon@sockar.homeip.net>
2004-08-05 02:52:33 +00:00
/*
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
* RX TCP/UDP checksum
Implement TCP/UDP Transmit/Receive checksum offload. Since HME doesn't compensate the checksum for UDP datagram which can yield to 0x0, UDP transmit checksum offload is disabled by default. The UDP Transmit checksum offload can be reactivated by setting special link option link0 with ifconfig(8). Approved by: jake (mentor) Reviewed by: tmm Tested by: Herve Boulouis <amon@sockar.homeip.net>
2004-08-05 02:52:33 +00:00
*/
static void
hme_rxcksum(struct mbuf *m, u_int32_t flags)
{
struct ether_header *eh;
struct ip *ip;
struct udphdr *uh;
int32_t hlen, len, pktlen;
u_int16_t cksum, *opts;
u_int32_t temp32;
pktlen = m->m_pkthdr.len;
if (pktlen < sizeof(struct ether_header) + sizeof(struct ip))
return;
eh = mtod(m, struct ether_header *);
if (eh->ether_type != htons(ETHERTYPE_IP))
return;
ip = (struct ip *)(eh + 1);
if (ip->ip_v != IPVERSION)
return;
hlen = ip->ip_hl << 2;
pktlen -= sizeof(struct ether_header);
if (hlen < sizeof(struct ip))
return;
if (ntohs(ip->ip_len) < hlen)
return;
if (ntohs(ip->ip_len) != pktlen)
return;
if (ip->ip_off & htons(IP_MF | IP_OFFMASK))
return; /* can't handle fragmented packet */
switch (ip->ip_p) {
case IPPROTO_TCP:
if (pktlen < (hlen + sizeof(struct tcphdr)))
return;
break;
case IPPROTO_UDP:
if (pktlen < (hlen + sizeof(struct udphdr)))
return;
uh = (struct udphdr *)((caddr_t)ip + hlen);
if (uh->uh_sum == 0)
return; /* no checksum */
break;
default:
return;
}
Backout rev. 1.46. It caused Rx checksum offload breakage on little endian systems. Reported by: joerg Tested by: joerg
2006-03-21 12:21:51 +00:00
cksum = ~(flags & HME_XD_RXCKSUM);
Implement TCP/UDP Transmit/Receive checksum offload. Since HME doesn't compensate the checksum for UDP datagram which can yield to 0x0, UDP transmit checksum offload is disabled by default. The UDP Transmit checksum offload can be reactivated by setting special link option link0 with ifconfig(8). Approved by: jake (mentor) Reviewed by: tmm Tested by: Herve Boulouis <amon@sockar.homeip.net>
2004-08-05 02:52:33 +00:00
/* checksum fixup for IP options */
len = hlen - sizeof(struct ip);
if (len > 0) {
opts = (u_int16_t *)(ip + 1);
for (; len > 0; len -= sizeof(u_int16_t), opts++) {
temp32 = cksum - *opts;
temp32 = (temp32 >> 16) + (temp32 & 65535);
cksum = temp32 & 65535;
}
}
m->m_pkthdr.csum_flags |= CSUM_DATA_VALID;
m->m_pkthdr.csum_data = cksum;
}
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
/*
* Receive interrupt.
*/
static void
hme_rint(struct hme_softc *sc)
{
caddr_t xdr = sc->sc_rb.rb_rxd;
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
struct ifnet *ifp = sc->sc_ifp;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
unsigned int ri, len;
bus_dmamap_sync() overhaul: - Remove NetBSD-style or-ed together BUS_DMASYNC operations, in some cases relaxing the (intended) syncing operation a bit. - Add syncs before reading the descriptor rings. - Try to combine syncs where possible to avoid overhead. - Sync all maps before unloading them.
2003-01-21 17:22:52 +00:00
int progress = 0;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
u_int32_t flags;
/*
* Process all buffers with valid data.
*/
bus_dmamap_sync() overhaul: - Remove NetBSD-style or-ed together BUS_DMASYNC operations, in some cases relaxing the (intended) syncing operation a bit. - Add syncs before reading the descriptor rings. - Try to combine syncs where possible to avoid overhead. - Sync all maps before unloading them.
2003-01-21 17:22:52 +00:00
bus_dmamap_sync(sc->sc_cdmatag, sc->sc_cdmamap, BUS_DMASYNC_POSTREAD);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
for (ri = sc->sc_rb.rb_rdtail;; ri = (ri + 1) % HME_NRXDESC) {
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
flags = HME_XD_GETFLAGS(sc->sc_flags & HME_PCI, xdr, ri);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
CTR2(KTR_HME, "hme_rint: index %d, flags %#x", ri, flags);
if ((flags & HME_XD_OWN) != 0)
break;
bus_dmamap_sync() overhaul: - Remove NetBSD-style or-ed together BUS_DMASYNC operations, in some cases relaxing the (intended) syncing operation a bit. - Add syncs before reading the descriptor rings. - Try to combine syncs where possible to avoid overhead. - Sync all maps before unloading them.
2003-01-21 17:22:52 +00:00
progress++;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
if ((flags & HME_XD_OFL) != 0) {
device_printf(sc->sc_dev, "buffer overflow, ri=%d; "
"flags=0x%x\n", ri, flags);
Miscellaneous fixes: - always reinitialize the rx descriptors, even if the mbuf is kept. This should fix the hangs on ifconfig that were observed - on an rx overflow, reinitialize the descriptor so that the interface will not hang - correct some bus_dmamap_sync() calls - correct some debug messages - some minor nits
2002-07-14 12:09:48 +00:00
ifp->if_ierrors++;
bus_dmamap_sync() overhaul: - Remove NetBSD-style or-ed together BUS_DMASYNC operations, in some cases relaxing the (intended) syncing operation a bit. - Add syncs before reading the descriptor rings. - Try to combine syncs where possible to avoid overhead. - Sync all maps before unloading them.
2003-01-21 17:22:52 +00:00
hme_discard_rxbuf(sc, ri);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
} else {
len = HME_XD_DECODE_RSIZE(flags);
Implement TCP/UDP Transmit/Receive checksum offload. Since HME doesn't compensate the checksum for UDP datagram which can yield to 0x0, UDP transmit checksum offload is disabled by default. The UDP Transmit checksum offload can be reactivated by setting special link option link0 with ifconfig(8). Approved by: jake (mentor) Reviewed by: tmm Tested by: Herve Boulouis <amon@sockar.homeip.net>
2004-08-05 02:52:33 +00:00
hme_read(sc, ri, len, flags);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
}
}
bus_dmamap_sync() overhaul: - Remove NetBSD-style or-ed together BUS_DMASYNC operations, in some cases relaxing the (intended) syncing operation a bit. - Add syncs before reading the descriptor rings. - Try to combine syncs where possible to avoid overhead. - Sync all maps before unloading them.
2003-01-21 17:22:52 +00:00
if (progress) {
bus_dmamap_sync(sc->sc_cdmatag, sc->sc_cdmamap,
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
bus_dmamap_sync() overhaul: - Remove NetBSD-style or-ed together BUS_DMASYNC operations, in some cases relaxing the (intended) syncing operation a bit. - Add syncs before reading the descriptor rings. - Try to combine syncs where possible to avoid overhead. - Sync all maps before unloading them.
2003-01-21 17:22:52 +00:00
}
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
sc->sc_rb.rb_rdtail = ri;
}
static void
hme_eint(struct hme_softc *sc, u_int status)
{
if ((status & HME_SEB_STAT_MIFIRQ) != 0) {
- In hme_stop() mask all interrupts. - In hme_eint() print MIF register contents on MIF interrupts. - In hme_mifinit() don't bother to preserve the previous MIF config. This was mainly done in order to preserve the PHY select bit (external or internal PHY) but which only needs to be set as appropriate when reading from or writing to the desired PHY in hme_mii_{read,write}reg(). Similarly don't bother to set the PHY select bit in hme_mii_statchg(). - In hme_mii_{read,write}reg() ignore requests to PHYs other than the external and internal PHY one. - Move enabling/disabling the MII drivers of the external transceiver from hme_init_locked() and based on the sheer presence of an external to hme_mifinit() and based on the currently selected media, defaulting to the internal transceiver when the media hasn't been set, yet. Invoke hme_mifinit() from the newly added hme_mediachange_locked() so the setting of the MII drivers is updated when changing media. These changes keep the MII bus from wedging (which manifests in the HME and the PHYs no longer being able to communicate with each other) when the PHY device drivers isolate the unused PHY in two-PHY configurations as present in f.e. Netra t1 100 while changing media, either from hme_init_locked() (see also below) or via ifconfig(8). They also allow for using both transceivers/PHYs. - In the newly added hme_mediachange_locked() also reset the PHYs in two- PHY configurations before invoking mii_mediachg(). This is required for successfully unisolating the previously unused PHY when switching between PHYs. - Now that changing media should no longer cause problems back out rev. 1.27 and re-enable setting the current media in hme_init_locked() (see the commit message of rev. 1.23 for more info). These changes are roughly a merge of NetBSD gem.c rev. 1.32 - 1.35 (1.30 was already fixed differently in our 1.36; 1.31 and 1.32 were wrong) with some parts reworked and things that don't make sense like setting the MII drivers and restoring the previous MIF and XIF settings in hme_mii_{read, write}reg() omitted. MFC after: 2 weeks
2006-12-04 00:51:08 +00:00
device_printf(sc->sc_dev, "XXXlink status changed: "
"cfg=%#x, stat=%#x, sm=%#x\n",
HME_MIF_READ_4(sc, HME_MIFI_CFG),
HME_MIF_READ_4(sc, HME_MIFI_STAT),
HME_MIF_READ_4(sc, HME_MIFI_SM));
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
return;
}
- Convert hme(4) to use TX side bus_dmamap_load_mbuf_sg(9). - Move hardware counter reading/zeroing to hme_tick(). This saves 8 register access per interrupt. [1] - Use imax macro for getting max. argument between two integers. - Invoke bus_dmamap_sync(9) first before freeing mbuf. - Check driver queue first to reduce locking operation in hme_start_locked() and interrupt handler. - Simplyfy watchdog timer setup in interrupt handler. - Don't log normal errors such as RX overrun. If we have DMA stuck condition, reinitialize the driver and log it. Reviewed by: marius Obtained from: OpenBSD [1]
2005-10-25 03:56:21 +00:00
/* check for fatal errors that needs reset to unfreeze DMA engine */
if ((status & HME_SEB_STAT_FATAL_ERRORS) != 0) {
HME_WHINE(sc->sc_dev, "error signaled, status=%#x\n", status);
hme_init_locked(sc);
}
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
}
void
hme_intr(void *v)
{
struct hme_softc *sc = (struct hme_softc *)v;
u_int32_t status;
Make hme(4) mpsafe - Let hme_start()/hme_init() acquire lock and then call hme_start_locked()/hme_init_locked() respectivly. - Teardown interrupt handler before hme_detach(). - Remove IFF_NEEDSGIANT flag and mark interrupt handler INTR_MPSAFE. - Set callout handler to CALLOUT_MPSAFE. - Add locks in hme MII interface. Reviewed by: jake Tested by: Julian C. Dunn <jdunn at opentrend dot net> MFC after: 2 weeks
2004-11-22 06:46:30 +00:00
HME_LOCK(sc);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
status = HME_SEB_READ_4(sc, HME_SEBI_STAT);
CTR1(KTR_HME, "hme_intr: status %#x", (u_int)status);
if ((status & HME_SEB_STAT_ALL_ERRORS) != 0)
hme_eint(sc, status);
if ((status & HME_SEB_STAT_RXTOHOST) != 0)
hme_rint(sc);
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
if ((status & (HME_SEB_STAT_TXALL | HME_SEB_STAT_HOSTTOTX)) != 0)
hme_tint(sc);
Make hme(4) mpsafe - Let hme_start()/hme_init() acquire lock and then call hme_start_locked()/hme_init_locked() respectivly. - Teardown interrupt handler before hme_detach(). - Remove IFF_NEEDSGIANT flag and mark interrupt handler INTR_MPSAFE. - Set callout handler to CALLOUT_MPSAFE. - Add locks in hme MII interface. Reviewed by: jake Tested by: Julian C. Dunn <jdunn at opentrend dot net> MFC after: 2 weeks
2004-11-22 06:46:30 +00:00
HME_UNLOCK(sc);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
}
- Use the hme_tick() callout instead of if_slowtimo() for driving hme_watchdog() in order to avoid races accessing if_timer. - Use bus_get_dma_tag() so hme(4) works on platforms requiring it. - Don't bother to set if_mtu to ETHERMTU, ether_ifattach() does that.
2006-12-06 02:07:20 +00:00
static int
hme_watchdog(struct hme_softc *sc)
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
{
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
struct ifnet *ifp = sc->sc_ifp;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
- Use the hme_tick() callout instead of if_slowtimo() for driving hme_watchdog() in order to avoid races accessing if_timer. - Use bus_get_dma_tag() so hme(4) works on platforms requiring it. - Don't bother to set if_mtu to ETHERMTU, ether_ifattach() does that.
2006-12-06 02:07:20 +00:00
HME_LOCK_ASSERT(sc, MA_OWNED);
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
Make hme(4) mpsafe - Let hme_start()/hme_init() acquire lock and then call hme_start_locked()/hme_init_locked() respectivly. - Teardown interrupt handler before hme_detach(). - Remove IFF_NEEDSGIANT flag and mark interrupt handler INTR_MPSAFE. - Set callout handler to CALLOUT_MPSAFE. - Add locks in hme MII interface. Reviewed by: jake Tested by: Julian C. Dunn <jdunn at opentrend dot net> MFC after: 2 weeks
2004-11-22 06:46:30 +00:00
#ifdef HMEDEBUG
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
CTR1(KTR_HME, "hme_watchdog: status %x",
(u_int)HME_SEB_READ_4(sc, HME_SEBI_STAT));
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
#endif
- Use the hme_tick() callout instead of if_slowtimo() for driving hme_watchdog() in order to avoid races accessing if_timer. - Use bus_get_dma_tag() so hme(4) works on platforms requiring it. - Don't bother to set if_mtu to ETHERMTU, ether_ifattach() does that.
2006-12-06 02:07:20 +00:00
if (sc->sc_wdog_timer == 0 || --sc->sc_wdog_timer != 0)
return (0);
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
if ((sc->sc_flags & HME_LINK) != 0)
device_printf(sc->sc_dev, "device timeout\n");
else if (bootverbose)
device_printf(sc->sc_dev, "device timeout (no link)\n");
++ifp->if_oerrors;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
- Use callout_init_mtx() to close races between hme_stop() and hme_tick(). - Use the driver lock instead of Giant in a bus dma callback. - Clear IFF_DRV_(RUNNING|OACTIVE) in hme_stop() instead of just clearing RUNNING in hme_ioctl() to be more like other ethernet drivers. - Lock the driver lock around mii operations. - Remove spls. - Cleanup locking in hme_ioctl(). MFC after: 1 week
2005-08-10 20:12:25 +00:00
hme_init_locked(sc);
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
hme_start_locked(ifp);
- Use the hme_tick() callout instead of if_slowtimo() for driving hme_watchdog() in order to avoid races accessing if_timer. - Use bus_get_dma_tag() so hme(4) works on platforms requiring it. - Don't bother to set if_mtu to ETHERMTU, ether_ifattach() does that.
2006-12-06 02:07:20 +00:00
return (EJUSTRETURN);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
}
/*
* Initialize the MII Management Interface
*/
static void
hme_mifinit(struct hme_softc *sc)
{
u_int32_t v;
- In hme_stop() mask all interrupts. - In hme_eint() print MIF register contents on MIF interrupts. - In hme_mifinit() don't bother to preserve the previous MIF config. This was mainly done in order to preserve the PHY select bit (external or internal PHY) but which only needs to be set as appropriate when reading from or writing to the desired PHY in hme_mii_{read,write}reg(). Similarly don't bother to set the PHY select bit in hme_mii_statchg(). - In hme_mii_{read,write}reg() ignore requests to PHYs other than the external and internal PHY one. - Move enabling/disabling the MII drivers of the external transceiver from hme_init_locked() and based on the sheer presence of an external to hme_mifinit() and based on the currently selected media, defaulting to the internal transceiver when the media hasn't been set, yet. Invoke hme_mifinit() from the newly added hme_mediachange_locked() so the setting of the MII drivers is updated when changing media. These changes keep the MII bus from wedging (which manifests in the HME and the PHYs no longer being able to communicate with each other) when the PHY device drivers isolate the unused PHY in two-PHY configurations as present in f.e. Netra t1 100 while changing media, either from hme_init_locked() (see also below) or via ifconfig(8). They also allow for using both transceivers/PHYs. - In the newly added hme_mediachange_locked() also reset the PHYs in two- PHY configurations before invoking mii_mediachg(). This is required for successfully unisolating the previously unused PHY when switching between PHYs. - Now that changing media should no longer cause problems back out rev. 1.27 and re-enable setting the current media in hme_init_locked() (see the commit message of rev. 1.23 for more info). These changes are roughly a merge of NetBSD gem.c rev. 1.32 - 1.35 (1.30 was already fixed differently in our 1.36; 1.31 and 1.32 were wrong) with some parts reworked and things that don't make sense like setting the MII drivers and restoring the previous MIF and XIF settings in hme_mii_{read, write}reg() omitted. MFC after: 2 weeks
2006-12-04 00:51:08 +00:00
/*
* Configure the MIF in frame mode, polling disabled, internal PHY
* selected.
*/
HME_MIF_WRITE_4(sc, HME_MIFI_CFG, 0);
/*
* If the currently selected media uses the external transceiver,
* enable its MII drivers (which basically isolates the internal
* one and vice versa). In case the current media hasn't been set,
* yet, we default to the internal transceiver.
*/
v = HME_MAC_READ_4(sc, HME_MACI_XIF);
if (sc->sc_mii != NULL && sc->sc_mii->mii_media.ifm_cur != NULL &&
sc->sc_phys[IFM_INST(sc->sc_mii->mii_media.ifm_cur->ifm_media)] ==
HME_PHYAD_EXTERNAL)
v |= HME_MAC_XIF_MIIENABLE;
else
v &= ~HME_MAC_XIF_MIIENABLE;
HME_MAC_WRITE_4(sc, HME_MACI_XIF, v);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
}
/*
* MII interface
*/
int
hme_mii_readreg(device_t dev, int phy, int reg)
{
- In hme_stop() mask all interrupts. - In hme_eint() print MIF register contents on MIF interrupts. - In hme_mifinit() don't bother to preserve the previous MIF config. This was mainly done in order to preserve the PHY select bit (external or internal PHY) but which only needs to be set as appropriate when reading from or writing to the desired PHY in hme_mii_{read,write}reg(). Similarly don't bother to set the PHY select bit in hme_mii_statchg(). - In hme_mii_{read,write}reg() ignore requests to PHYs other than the external and internal PHY one. - Move enabling/disabling the MII drivers of the external transceiver from hme_init_locked() and based on the sheer presence of an external to hme_mifinit() and based on the currently selected media, defaulting to the internal transceiver when the media hasn't been set, yet. Invoke hme_mifinit() from the newly added hme_mediachange_locked() so the setting of the MII drivers is updated when changing media. These changes keep the MII bus from wedging (which manifests in the HME and the PHYs no longer being able to communicate with each other) when the PHY device drivers isolate the unused PHY in two-PHY configurations as present in f.e. Netra t1 100 while changing media, either from hme_init_locked() (see also below) or via ifconfig(8). They also allow for using both transceivers/PHYs. - In the newly added hme_mediachange_locked() also reset the PHYs in two- PHY configurations before invoking mii_mediachg(). This is required for successfully unisolating the previously unused PHY when switching between PHYs. - Now that changing media should no longer cause problems back out rev. 1.27 and re-enable setting the current media in hme_init_locked() (see the commit message of rev. 1.23 for more info). These changes are roughly a merge of NetBSD gem.c rev. 1.32 - 1.35 (1.30 was already fixed differently in our 1.36; 1.31 and 1.32 were wrong) with some parts reworked and things that don't make sense like setting the MII drivers and restoring the previous MIF and XIF settings in hme_mii_{read, write}reg() omitted. MFC after: 2 weeks
2006-12-04 00:51:08 +00:00
struct hme_softc *sc;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
int n;
u_int32_t v;
- In hme_stop() mask all interrupts. - In hme_eint() print MIF register contents on MIF interrupts. - In hme_mifinit() don't bother to preserve the previous MIF config. This was mainly done in order to preserve the PHY select bit (external or internal PHY) but which only needs to be set as appropriate when reading from or writing to the desired PHY in hme_mii_{read,write}reg(). Similarly don't bother to set the PHY select bit in hme_mii_statchg(). - In hme_mii_{read,write}reg() ignore requests to PHYs other than the external and internal PHY one. - Move enabling/disabling the MII drivers of the external transceiver from hme_init_locked() and based on the sheer presence of an external to hme_mifinit() and based on the currently selected media, defaulting to the internal transceiver when the media hasn't been set, yet. Invoke hme_mifinit() from the newly added hme_mediachange_locked() so the setting of the MII drivers is updated when changing media. These changes keep the MII bus from wedging (which manifests in the HME and the PHYs no longer being able to communicate with each other) when the PHY device drivers isolate the unused PHY in two-PHY configurations as present in f.e. Netra t1 100 while changing media, either from hme_init_locked() (see also below) or via ifconfig(8). They also allow for using both transceivers/PHYs. - In the newly added hme_mediachange_locked() also reset the PHYs in two- PHY configurations before invoking mii_mediachg(). This is required for successfully unisolating the previously unused PHY when switching between PHYs. - Now that changing media should no longer cause problems back out rev. 1.27 and re-enable setting the current media in hme_init_locked() (see the commit message of rev. 1.23 for more info). These changes are roughly a merge of NetBSD gem.c rev. 1.32 - 1.35 (1.30 was already fixed differently in our 1.36; 1.31 and 1.32 were wrong) with some parts reworked and things that don't make sense like setting the MII drivers and restoring the previous MIF and XIF settings in hme_mii_{read, write}reg() omitted. MFC after: 2 weeks
2006-12-04 00:51:08 +00:00
/* We can at most have two PHYs. */
if (phy != HME_PHYAD_EXTERNAL && phy != HME_PHYAD_INTERNAL)
return (0);
sc = device_get_softc(dev);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
/* Select the desired PHY in the MIF configuration register */
v = HME_MIF_READ_4(sc, HME_MIFI_CFG);
if (phy == HME_PHYAD_EXTERNAL)
v |= HME_MIF_CFG_PHY;
- In hme_stop() mask all interrupts. - In hme_eint() print MIF register contents on MIF interrupts. - In hme_mifinit() don't bother to preserve the previous MIF config. This was mainly done in order to preserve the PHY select bit (external or internal PHY) but which only needs to be set as appropriate when reading from or writing to the desired PHY in hme_mii_{read,write}reg(). Similarly don't bother to set the PHY select bit in hme_mii_statchg(). - In hme_mii_{read,write}reg() ignore requests to PHYs other than the external and internal PHY one. - Move enabling/disabling the MII drivers of the external transceiver from hme_init_locked() and based on the sheer presence of an external to hme_mifinit() and based on the currently selected media, defaulting to the internal transceiver when the media hasn't been set, yet. Invoke hme_mifinit() from the newly added hme_mediachange_locked() so the setting of the MII drivers is updated when changing media. These changes keep the MII bus from wedging (which manifests in the HME and the PHYs no longer being able to communicate with each other) when the PHY device drivers isolate the unused PHY in two-PHY configurations as present in f.e. Netra t1 100 while changing media, either from hme_init_locked() (see also below) or via ifconfig(8). They also allow for using both transceivers/PHYs. - In the newly added hme_mediachange_locked() also reset the PHYs in two- PHY configurations before invoking mii_mediachg(). This is required for successfully unisolating the previously unused PHY when switching between PHYs. - Now that changing media should no longer cause problems back out rev. 1.27 and re-enable setting the current media in hme_init_locked() (see the commit message of rev. 1.23 for more info). These changes are roughly a merge of NetBSD gem.c rev. 1.32 - 1.35 (1.30 was already fixed differently in our 1.36; 1.31 and 1.32 were wrong) with some parts reworked and things that don't make sense like setting the MII drivers and restoring the previous MIF and XIF settings in hme_mii_{read, write}reg() omitted. MFC after: 2 weeks
2006-12-04 00:51:08 +00:00
else
v &= ~HME_MIF_CFG_PHY;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
HME_MIF_WRITE_4(sc, HME_MIFI_CFG, v);
/* Construct the frame command */
v = (MII_COMMAND_START << HME_MIF_FO_ST_SHIFT) |
HME_MIF_FO_TAMSB |
(MII_COMMAND_READ << HME_MIF_FO_OPC_SHIFT) |
(phy << HME_MIF_FO_PHYAD_SHIFT) |
(reg << HME_MIF_FO_REGAD_SHIFT);
HME_MIF_WRITE_4(sc, HME_MIFI_FO, v);
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
HME_MIF_BARRIER(sc, HME_MIFI_FO, 4,
BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
for (n = 0; n < 100; n++) {
DELAY(1);
v = HME_MIF_READ_4(sc, HME_MIFI_FO);
- In hme_stop() mask all interrupts. - In hme_eint() print MIF register contents on MIF interrupts. - In hme_mifinit() don't bother to preserve the previous MIF config. This was mainly done in order to preserve the PHY select bit (external or internal PHY) but which only needs to be set as appropriate when reading from or writing to the desired PHY in hme_mii_{read,write}reg(). Similarly don't bother to set the PHY select bit in hme_mii_statchg(). - In hme_mii_{read,write}reg() ignore requests to PHYs other than the external and internal PHY one. - Move enabling/disabling the MII drivers of the external transceiver from hme_init_locked() and based on the sheer presence of an external to hme_mifinit() and based on the currently selected media, defaulting to the internal transceiver when the media hasn't been set, yet. Invoke hme_mifinit() from the newly added hme_mediachange_locked() so the setting of the MII drivers is updated when changing media. These changes keep the MII bus from wedging (which manifests in the HME and the PHYs no longer being able to communicate with each other) when the PHY device drivers isolate the unused PHY in two-PHY configurations as present in f.e. Netra t1 100 while changing media, either from hme_init_locked() (see also below) or via ifconfig(8). They also allow for using both transceivers/PHYs. - In the newly added hme_mediachange_locked() also reset the PHYs in two- PHY configurations before invoking mii_mediachg(). This is required for successfully unisolating the previously unused PHY when switching between PHYs. - Now that changing media should no longer cause problems back out rev. 1.27 and re-enable setting the current media in hme_init_locked() (see the commit message of rev. 1.23 for more info). These changes are roughly a merge of NetBSD gem.c rev. 1.32 - 1.35 (1.30 was already fixed differently in our 1.36; 1.31 and 1.32 were wrong) with some parts reworked and things that don't make sense like setting the MII drivers and restoring the previous MIF and XIF settings in hme_mii_{read, write}reg() omitted. MFC after: 2 weeks
2006-12-04 00:51:08 +00:00
if (v & HME_MIF_FO_TALSB)
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
return (v & HME_MIF_FO_DATA);
}
device_printf(sc->sc_dev, "mii_read timeout\n");
return (0);
}
int
hme_mii_writereg(device_t dev, int phy, int reg, int val)
{
- In hme_stop() mask all interrupts. - In hme_eint() print MIF register contents on MIF interrupts. - In hme_mifinit() don't bother to preserve the previous MIF config. This was mainly done in order to preserve the PHY select bit (external or internal PHY) but which only needs to be set as appropriate when reading from or writing to the desired PHY in hme_mii_{read,write}reg(). Similarly don't bother to set the PHY select bit in hme_mii_statchg(). - In hme_mii_{read,write}reg() ignore requests to PHYs other than the external and internal PHY one. - Move enabling/disabling the MII drivers of the external transceiver from hme_init_locked() and based on the sheer presence of an external to hme_mifinit() and based on the currently selected media, defaulting to the internal transceiver when the media hasn't been set, yet. Invoke hme_mifinit() from the newly added hme_mediachange_locked() so the setting of the MII drivers is updated when changing media. These changes keep the MII bus from wedging (which manifests in the HME and the PHYs no longer being able to communicate with each other) when the PHY device drivers isolate the unused PHY in two-PHY configurations as present in f.e. Netra t1 100 while changing media, either from hme_init_locked() (see also below) or via ifconfig(8). They also allow for using both transceivers/PHYs. - In the newly added hme_mediachange_locked() also reset the PHYs in two- PHY configurations before invoking mii_mediachg(). This is required for successfully unisolating the previously unused PHY when switching between PHYs. - Now that changing media should no longer cause problems back out rev. 1.27 and re-enable setting the current media in hme_init_locked() (see the commit message of rev. 1.23 for more info). These changes are roughly a merge of NetBSD gem.c rev. 1.32 - 1.35 (1.30 was already fixed differently in our 1.36; 1.31 and 1.32 were wrong) with some parts reworked and things that don't make sense like setting the MII drivers and restoring the previous MIF and XIF settings in hme_mii_{read, write}reg() omitted. MFC after: 2 weeks
2006-12-04 00:51:08 +00:00
struct hme_softc *sc;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
int n;
u_int32_t v;
- In hme_stop() mask all interrupts. - In hme_eint() print MIF register contents on MIF interrupts. - In hme_mifinit() don't bother to preserve the previous MIF config. This was mainly done in order to preserve the PHY select bit (external or internal PHY) but which only needs to be set as appropriate when reading from or writing to the desired PHY in hme_mii_{read,write}reg(). Similarly don't bother to set the PHY select bit in hme_mii_statchg(). - In hme_mii_{read,write}reg() ignore requests to PHYs other than the external and internal PHY one. - Move enabling/disabling the MII drivers of the external transceiver from hme_init_locked() and based on the sheer presence of an external to hme_mifinit() and based on the currently selected media, defaulting to the internal transceiver when the media hasn't been set, yet. Invoke hme_mifinit() from the newly added hme_mediachange_locked() so the setting of the MII drivers is updated when changing media. These changes keep the MII bus from wedging (which manifests in the HME and the PHYs no longer being able to communicate with each other) when the PHY device drivers isolate the unused PHY in two-PHY configurations as present in f.e. Netra t1 100 while changing media, either from hme_init_locked() (see also below) or via ifconfig(8). They also allow for using both transceivers/PHYs. - In the newly added hme_mediachange_locked() also reset the PHYs in two- PHY configurations before invoking mii_mediachg(). This is required for successfully unisolating the previously unused PHY when switching between PHYs. - Now that changing media should no longer cause problems back out rev. 1.27 and re-enable setting the current media in hme_init_locked() (see the commit message of rev. 1.23 for more info). These changes are roughly a merge of NetBSD gem.c rev. 1.32 - 1.35 (1.30 was already fixed differently in our 1.36; 1.31 and 1.32 were wrong) with some parts reworked and things that don't make sense like setting the MII drivers and restoring the previous MIF and XIF settings in hme_mii_{read, write}reg() omitted. MFC after: 2 weeks
2006-12-04 00:51:08 +00:00
/* We can at most have two PHYs. */
if (phy != HME_PHYAD_EXTERNAL && phy != HME_PHYAD_INTERNAL)
return (0);
sc = device_get_softc(dev);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
/* Select the desired PHY in the MIF configuration register */
v = HME_MIF_READ_4(sc, HME_MIFI_CFG);
if (phy == HME_PHYAD_EXTERNAL)
v |= HME_MIF_CFG_PHY;
- In hme_stop() mask all interrupts. - In hme_eint() print MIF register contents on MIF interrupts. - In hme_mifinit() don't bother to preserve the previous MIF config. This was mainly done in order to preserve the PHY select bit (external or internal PHY) but which only needs to be set as appropriate when reading from or writing to the desired PHY in hme_mii_{read,write}reg(). Similarly don't bother to set the PHY select bit in hme_mii_statchg(). - In hme_mii_{read,write}reg() ignore requests to PHYs other than the external and internal PHY one. - Move enabling/disabling the MII drivers of the external transceiver from hme_init_locked() and based on the sheer presence of an external to hme_mifinit() and based on the currently selected media, defaulting to the internal transceiver when the media hasn't been set, yet. Invoke hme_mifinit() from the newly added hme_mediachange_locked() so the setting of the MII drivers is updated when changing media. These changes keep the MII bus from wedging (which manifests in the HME and the PHYs no longer being able to communicate with each other) when the PHY device drivers isolate the unused PHY in two-PHY configurations as present in f.e. Netra t1 100 while changing media, either from hme_init_locked() (see also below) or via ifconfig(8). They also allow for using both transceivers/PHYs. - In the newly added hme_mediachange_locked() also reset the PHYs in two- PHY configurations before invoking mii_mediachg(). This is required for successfully unisolating the previously unused PHY when switching between PHYs. - Now that changing media should no longer cause problems back out rev. 1.27 and re-enable setting the current media in hme_init_locked() (see the commit message of rev. 1.23 for more info). These changes are roughly a merge of NetBSD gem.c rev. 1.32 - 1.35 (1.30 was already fixed differently in our 1.36; 1.31 and 1.32 were wrong) with some parts reworked and things that don't make sense like setting the MII drivers and restoring the previous MIF and XIF settings in hme_mii_{read, write}reg() omitted. MFC after: 2 weeks
2006-12-04 00:51:08 +00:00
else
v &= ~HME_MIF_CFG_PHY;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
HME_MIF_WRITE_4(sc, HME_MIFI_CFG, v);
/* Construct the frame command */
v = (MII_COMMAND_START << HME_MIF_FO_ST_SHIFT) |
HME_MIF_FO_TAMSB |
(MII_COMMAND_WRITE << HME_MIF_FO_OPC_SHIFT) |
(phy << HME_MIF_FO_PHYAD_SHIFT) |
(reg << HME_MIF_FO_REGAD_SHIFT) |
(val & HME_MIF_FO_DATA);
HME_MIF_WRITE_4(sc, HME_MIFI_FO, v);
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
HME_MIF_BARRIER(sc, HME_MIFI_FO, 4,
BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
for (n = 0; n < 100; n++) {
DELAY(1);
v = HME_MIF_READ_4(sc, HME_MIFI_FO);
- Use callout_init_mtx() to close races between hme_stop() and hme_tick(). - Use the driver lock instead of Giant in a bus dma callback. - Clear IFF_DRV_(RUNNING|OACTIVE) in hme_stop() instead of just clearing RUNNING in hme_ioctl() to be more like other ethernet drivers. - Lock the driver lock around mii operations. - Remove spls. - Cleanup locking in hme_ioctl(). MFC after: 1 week
2005-08-10 20:12:25 +00:00
if (v & HME_MIF_FO_TALSB)
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
return (1);
}
device_printf(sc->sc_dev, "mii_write timeout\n");
return (0);
}
void
hme_mii_statchg(device_t dev)
{
- In hme_stop() mask all interrupts. - In hme_eint() print MIF register contents on MIF interrupts. - In hme_mifinit() don't bother to preserve the previous MIF config. This was mainly done in order to preserve the PHY select bit (external or internal PHY) but which only needs to be set as appropriate when reading from or writing to the desired PHY in hme_mii_{read,write}reg(). Similarly don't bother to set the PHY select bit in hme_mii_statchg(). - In hme_mii_{read,write}reg() ignore requests to PHYs other than the external and internal PHY one. - Move enabling/disabling the MII drivers of the external transceiver from hme_init_locked() and based on the sheer presence of an external to hme_mifinit() and based on the currently selected media, defaulting to the internal transceiver when the media hasn't been set, yet. Invoke hme_mifinit() from the newly added hme_mediachange_locked() so the setting of the MII drivers is updated when changing media. These changes keep the MII bus from wedging (which manifests in the HME and the PHYs no longer being able to communicate with each other) when the PHY device drivers isolate the unused PHY in two-PHY configurations as present in f.e. Netra t1 100 while changing media, either from hme_init_locked() (see also below) or via ifconfig(8). They also allow for using both transceivers/PHYs. - In the newly added hme_mediachange_locked() also reset the PHYs in two- PHY configurations before invoking mii_mediachg(). This is required for successfully unisolating the previously unused PHY when switching between PHYs. - Now that changing media should no longer cause problems back out rev. 1.27 and re-enable setting the current media in hme_init_locked() (see the commit message of rev. 1.23 for more info). These changes are roughly a merge of NetBSD gem.c rev. 1.32 - 1.35 (1.30 was already fixed differently in our 1.36; 1.31 and 1.32 were wrong) with some parts reworked and things that don't make sense like setting the MII drivers and restoring the previous MIF and XIF settings in hme_mii_{read, write}reg() omitted. MFC after: 2 weeks
2006-12-04 00:51:08 +00:00
struct hme_softc *sc;
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
uint32_t rxcfg, txcfg;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
- In hme_stop() mask all interrupts. - In hme_eint() print MIF register contents on MIF interrupts. - In hme_mifinit() don't bother to preserve the previous MIF config. This was mainly done in order to preserve the PHY select bit (external or internal PHY) but which only needs to be set as appropriate when reading from or writing to the desired PHY in hme_mii_{read,write}reg(). Similarly don't bother to set the PHY select bit in hme_mii_statchg(). - In hme_mii_{read,write}reg() ignore requests to PHYs other than the external and internal PHY one. - Move enabling/disabling the MII drivers of the external transceiver from hme_init_locked() and based on the sheer presence of an external to hme_mifinit() and based on the currently selected media, defaulting to the internal transceiver when the media hasn't been set, yet. Invoke hme_mifinit() from the newly added hme_mediachange_locked() so the setting of the MII drivers is updated when changing media. These changes keep the MII bus from wedging (which manifests in the HME and the PHYs no longer being able to communicate with each other) when the PHY device drivers isolate the unused PHY in two-PHY configurations as present in f.e. Netra t1 100 while changing media, either from hme_init_locked() (see also below) or via ifconfig(8). They also allow for using both transceivers/PHYs. - In the newly added hme_mediachange_locked() also reset the PHYs in two- PHY configurations before invoking mii_mediachg(). This is required for successfully unisolating the previously unused PHY when switching between PHYs. - Now that changing media should no longer cause problems back out rev. 1.27 and re-enable setting the current media in hme_init_locked() (see the commit message of rev. 1.23 for more info). These changes are roughly a merge of NetBSD gem.c rev. 1.32 - 1.35 (1.30 was already fixed differently in our 1.36; 1.31 and 1.32 were wrong) with some parts reworked and things that don't make sense like setting the MII drivers and restoring the previous MIF and XIF settings in hme_mii_{read, write}reg() omitted. MFC after: 2 weeks
2006-12-04 00:51:08 +00:00
sc = device_get_softc(dev);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
#ifdef HMEDEBUG
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
if ((sc->sc_ifp->if_flags & IFF_DEBUG) != 0)
- In hme_stop() mask all interrupts. - In hme_eint() print MIF register contents on MIF interrupts. - In hme_mifinit() don't bother to preserve the previous MIF config. This was mainly done in order to preserve the PHY select bit (external or internal PHY) but which only needs to be set as appropriate when reading from or writing to the desired PHY in hme_mii_{read,write}reg(). Similarly don't bother to set the PHY select bit in hme_mii_statchg(). - In hme_mii_{read,write}reg() ignore requests to PHYs other than the external and internal PHY one. - Move enabling/disabling the MII drivers of the external transceiver from hme_init_locked() and based on the sheer presence of an external to hme_mifinit() and based on the currently selected media, defaulting to the internal transceiver when the media hasn't been set, yet. Invoke hme_mifinit() from the newly added hme_mediachange_locked() so the setting of the MII drivers is updated when changing media. These changes keep the MII bus from wedging (which manifests in the HME and the PHYs no longer being able to communicate with each other) when the PHY device drivers isolate the unused PHY in two-PHY configurations as present in f.e. Netra t1 100 while changing media, either from hme_init_locked() (see also below) or via ifconfig(8). They also allow for using both transceivers/PHYs. - In the newly added hme_mediachange_locked() also reset the PHYs in two- PHY configurations before invoking mii_mediachg(). This is required for successfully unisolating the previously unused PHY when switching between PHYs. - Now that changing media should no longer cause problems back out rev. 1.27 and re-enable setting the current media in hme_init_locked() (see the commit message of rev. 1.23 for more info). These changes are roughly a merge of NetBSD gem.c rev. 1.32 - 1.35 (1.30 was already fixed differently in our 1.36; 1.31 and 1.32 were wrong) with some parts reworked and things that don't make sense like setting the MII drivers and restoring the previous MIF and XIF settings in hme_mii_{read, write}reg() omitted. MFC after: 2 weeks
2006-12-04 00:51:08 +00:00
device_printf(sc->sc_dev, "hme_mii_statchg: status change\n");
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
#endif
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
if ((sc->sc_mii->mii_media_status & IFM_ACTIVE) != 0 &&
IFM_SUBTYPE(sc->sc_mii->mii_media_active) != IFM_NONE)
sc->sc_flags |= HME_LINK;
else
sc->sc_flags &= ~HME_LINK;
txcfg = HME_MAC_READ_4(sc, HME_MACI_TXCFG);
if (!hme_mac_bitflip(sc, HME_MACI_TXCFG, txcfg,
HME_MAC_TXCFG_ENABLE, 0))
device_printf(sc->sc_dev, "cannot disable TX MAC\n");
rxcfg = HME_MAC_READ_4(sc, HME_MACI_RXCFG);
if (!hme_mac_bitflip(sc, HME_MACI_RXCFG, rxcfg,
HME_MAC_RXCFG_ENABLE, 0))
device_printf(sc->sc_dev, "cannot disable RX MAC\n");
/* Set the MAC Full Duplex bit appropriately. */
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) & IFM_FDX) != 0)
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
txcfg |= HME_MAC_TXCFG_FULLDPLX;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
else
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
txcfg &= ~HME_MAC_TXCFG_FULLDPLX;
HME_MAC_WRITE_4(sc, HME_MACI_TXCFG, txcfg);
if ((sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) != 0 &&
(sc->sc_flags & HME_LINK) != 0) {
if (!hme_mac_bitflip(sc, HME_MACI_TXCFG, txcfg, 0,
HME_MAC_TXCFG_ENABLE))
device_printf(sc->sc_dev, "cannot enable TX MAC\n");
if (!hme_mac_bitflip(sc, HME_MACI_RXCFG, rxcfg, 0,
HME_MAC_RXCFG_ENABLE))
device_printf(sc->sc_dev, "cannot enable RX MAC\n");
}
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
}
static int
hme_mediachange(struct ifnet *ifp)
{
struct hme_softc *sc = ifp->if_softc;
- Use callout_init_mtx() to close races between hme_stop() and hme_tick(). - Use the driver lock instead of Giant in a bus dma callback. - Clear IFF_DRV_(RUNNING|OACTIVE) in hme_stop() instead of just clearing RUNNING in hme_ioctl() to be more like other ethernet drivers. - Lock the driver lock around mii operations. - Remove spls. - Cleanup locking in hme_ioctl(). MFC after: 1 week
2005-08-10 20:12:25 +00:00
int error;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
- Use callout_init_mtx() to close races between hme_stop() and hme_tick(). - Use the driver lock instead of Giant in a bus dma callback. - Clear IFF_DRV_(RUNNING|OACTIVE) in hme_stop() instead of just clearing RUNNING in hme_ioctl() to be more like other ethernet drivers. - Lock the driver lock around mii operations. - Remove spls. - Cleanup locking in hme_ioctl(). MFC after: 1 week
2005-08-10 20:12:25 +00:00
HME_LOCK(sc);
- In hme_stop() mask all interrupts. - In hme_eint() print MIF register contents on MIF interrupts. - In hme_mifinit() don't bother to preserve the previous MIF config. This was mainly done in order to preserve the PHY select bit (external or internal PHY) but which only needs to be set as appropriate when reading from or writing to the desired PHY in hme_mii_{read,write}reg(). Similarly don't bother to set the PHY select bit in hme_mii_statchg(). - In hme_mii_{read,write}reg() ignore requests to PHYs other than the external and internal PHY one. - Move enabling/disabling the MII drivers of the external transceiver from hme_init_locked() and based on the sheer presence of an external to hme_mifinit() and based on the currently selected media, defaulting to the internal transceiver when the media hasn't been set, yet. Invoke hme_mifinit() from the newly added hme_mediachange_locked() so the setting of the MII drivers is updated when changing media. These changes keep the MII bus from wedging (which manifests in the HME and the PHYs no longer being able to communicate with each other) when the PHY device drivers isolate the unused PHY in two-PHY configurations as present in f.e. Netra t1 100 while changing media, either from hme_init_locked() (see also below) or via ifconfig(8). They also allow for using both transceivers/PHYs. - In the newly added hme_mediachange_locked() also reset the PHYs in two- PHY configurations before invoking mii_mediachg(). This is required for successfully unisolating the previously unused PHY when switching between PHYs. - Now that changing media should no longer cause problems back out rev. 1.27 and re-enable setting the current media in hme_init_locked() (see the commit message of rev. 1.23 for more info). These changes are roughly a merge of NetBSD gem.c rev. 1.32 - 1.35 (1.30 was already fixed differently in our 1.36; 1.31 and 1.32 were wrong) with some parts reworked and things that don't make sense like setting the MII drivers and restoring the previous MIF and XIF settings in hme_mii_{read, write}reg() omitted. MFC after: 2 weeks
2006-12-04 00:51:08 +00:00
error = hme_mediachange_locked(sc);
- Use callout_init_mtx() to close races between hme_stop() and hme_tick(). - Use the driver lock instead of Giant in a bus dma callback. - Clear IFF_DRV_(RUNNING|OACTIVE) in hme_stop() instead of just clearing RUNNING in hme_ioctl() to be more like other ethernet drivers. - Lock the driver lock around mii operations. - Remove spls. - Cleanup locking in hme_ioctl(). MFC after: 1 week
2005-08-10 20:12:25 +00:00
HME_UNLOCK(sc);
return (error);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
}
- In hme_stop() mask all interrupts. - In hme_eint() print MIF register contents on MIF interrupts. - In hme_mifinit() don't bother to preserve the previous MIF config. This was mainly done in order to preserve the PHY select bit (external or internal PHY) but which only needs to be set as appropriate when reading from or writing to the desired PHY in hme_mii_{read,write}reg(). Similarly don't bother to set the PHY select bit in hme_mii_statchg(). - In hme_mii_{read,write}reg() ignore requests to PHYs other than the external and internal PHY one. - Move enabling/disabling the MII drivers of the external transceiver from hme_init_locked() and based on the sheer presence of an external to hme_mifinit() and based on the currently selected media, defaulting to the internal transceiver when the media hasn't been set, yet. Invoke hme_mifinit() from the newly added hme_mediachange_locked() so the setting of the MII drivers is updated when changing media. These changes keep the MII bus from wedging (which manifests in the HME and the PHYs no longer being able to communicate with each other) when the PHY device drivers isolate the unused PHY in two-PHY configurations as present in f.e. Netra t1 100 while changing media, either from hme_init_locked() (see also below) or via ifconfig(8). They also allow for using both transceivers/PHYs. - In the newly added hme_mediachange_locked() also reset the PHYs in two- PHY configurations before invoking mii_mediachg(). This is required for successfully unisolating the previously unused PHY when switching between PHYs. - Now that changing media should no longer cause problems back out rev. 1.27 and re-enable setting the current media in hme_init_locked() (see the commit message of rev. 1.23 for more info). These changes are roughly a merge of NetBSD gem.c rev. 1.32 - 1.35 (1.30 was already fixed differently in our 1.36; 1.31 and 1.32 were wrong) with some parts reworked and things that don't make sense like setting the MII drivers and restoring the previous MIF and XIF settings in hme_mii_{read, write}reg() omitted. MFC after: 2 weeks
2006-12-04 00:51:08 +00:00
static int
hme_mediachange_locked(struct hme_softc *sc)
{
struct mii_softc *child;
HME_LOCK_ASSERT(sc, MA_OWNED);
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
- In hme_stop() mask all interrupts. - In hme_eint() print MIF register contents on MIF interrupts. - In hme_mifinit() don't bother to preserve the previous MIF config. This was mainly done in order to preserve the PHY select bit (external or internal PHY) but which only needs to be set as appropriate when reading from or writing to the desired PHY in hme_mii_{read,write}reg(). Similarly don't bother to set the PHY select bit in hme_mii_statchg(). - In hme_mii_{read,write}reg() ignore requests to PHYs other than the external and internal PHY one. - Move enabling/disabling the MII drivers of the external transceiver from hme_init_locked() and based on the sheer presence of an external to hme_mifinit() and based on the currently selected media, defaulting to the internal transceiver when the media hasn't been set, yet. Invoke hme_mifinit() from the newly added hme_mediachange_locked() so the setting of the MII drivers is updated when changing media. These changes keep the MII bus from wedging (which manifests in the HME and the PHYs no longer being able to communicate with each other) when the PHY device drivers isolate the unused PHY in two-PHY configurations as present in f.e. Netra t1 100 while changing media, either from hme_init_locked() (see also below) or via ifconfig(8). They also allow for using both transceivers/PHYs. - In the newly added hme_mediachange_locked() also reset the PHYs in two- PHY configurations before invoking mii_mediachg(). This is required for successfully unisolating the previously unused PHY when switching between PHYs. - Now that changing media should no longer cause problems back out rev. 1.27 and re-enable setting the current media in hme_init_locked() (see the commit message of rev. 1.23 for more info). These changes are roughly a merge of NetBSD gem.c rev. 1.32 - 1.35 (1.30 was already fixed differently in our 1.36; 1.31 and 1.32 were wrong) with some parts reworked and things that don't make sense like setting the MII drivers and restoring the previous MIF and XIF settings in hme_mii_{read, write}reg() omitted. MFC after: 2 weeks
2006-12-04 00:51:08 +00:00
#ifdef HMEDEBUG
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
if ((sc->sc_ifp->if_flags & IFF_DEBUG) != 0)
- In hme_stop() mask all interrupts. - In hme_eint() print MIF register contents on MIF interrupts. - In hme_mifinit() don't bother to preserve the previous MIF config. This was mainly done in order to preserve the PHY select bit (external or internal PHY) but which only needs to be set as appropriate when reading from or writing to the desired PHY in hme_mii_{read,write}reg(). Similarly don't bother to set the PHY select bit in hme_mii_statchg(). - In hme_mii_{read,write}reg() ignore requests to PHYs other than the external and internal PHY one. - Move enabling/disabling the MII drivers of the external transceiver from hme_init_locked() and based on the sheer presence of an external to hme_mifinit() and based on the currently selected media, defaulting to the internal transceiver when the media hasn't been set, yet. Invoke hme_mifinit() from the newly added hme_mediachange_locked() so the setting of the MII drivers is updated when changing media. These changes keep the MII bus from wedging (which manifests in the HME and the PHYs no longer being able to communicate with each other) when the PHY device drivers isolate the unused PHY in two-PHY configurations as present in f.e. Netra t1 100 while changing media, either from hme_init_locked() (see also below) or via ifconfig(8). They also allow for using both transceivers/PHYs. - In the newly added hme_mediachange_locked() also reset the PHYs in two- PHY configurations before invoking mii_mediachg(). This is required for successfully unisolating the previously unused PHY when switching between PHYs. - Now that changing media should no longer cause problems back out rev. 1.27 and re-enable setting the current media in hme_init_locked() (see the commit message of rev. 1.23 for more info). These changes are roughly a merge of NetBSD gem.c rev. 1.32 - 1.35 (1.30 was already fixed differently in our 1.36; 1.31 and 1.32 were wrong) with some parts reworked and things that don't make sense like setting the MII drivers and restoring the previous MIF and XIF settings in hme_mii_{read, write}reg() omitted. MFC after: 2 weeks
2006-12-04 00:51:08 +00:00
device_printf(sc->sc_dev, "hme_mediachange_locked");
#endif
hme_mifinit(sc);
/*
* If both PHYs are present reset them. This is required for
* unisolating the previously isolated PHY when switching PHYs.
* As the above hme_mifinit() call will set the MII drivers in
* the XIF configuration register accoring to the currently
* selected media, there should be no window during which the
* data paths of both transceivers are open at the same time,
* even if the PHY device drivers use MIIF_NOISOLATE.
*/
if (sc->sc_phys[0] != -1 && sc->sc_phys[1] != -1)
LIST_FOREACH(child, &sc->sc_mii->mii_phys, mii_list)
mii_phy_reset(child);
return (mii_mediachg(sc->sc_mii));
}
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
static void
hme_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr)
{
struct hme_softc *sc = ifp->if_softc;
Make hme(4) mpsafe - Let hme_start()/hme_init() acquire lock and then call hme_start_locked()/hme_init_locked() respectivly. - Teardown interrupt handler before hme_detach(). - Remove IFF_NEEDSGIANT flag and mark interrupt handler INTR_MPSAFE. - Set callout handler to CALLOUT_MPSAFE. - Add locks in hme MII interface. Reviewed by: jake Tested by: Julian C. Dunn <jdunn at opentrend dot net> MFC after: 2 weeks
2004-11-22 06:46:30 +00:00
HME_LOCK(sc);
if ((ifp->if_flags & IFF_UP) == 0) {
HME_UNLOCK(sc);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
return;
Make hme(4) mpsafe - Let hme_start()/hme_init() acquire lock and then call hme_start_locked()/hme_init_locked() respectivly. - Teardown interrupt handler before hme_detach(). - Remove IFF_NEEDSGIANT flag and mark interrupt handler INTR_MPSAFE. - Set callout handler to CALLOUT_MPSAFE. - Add locks in hme MII interface. Reviewed by: jake Tested by: Julian C. Dunn <jdunn at opentrend dot net> MFC after: 2 weeks
2004-11-22 06:46:30 +00:00
}
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
mii_pollstat(sc->sc_mii);
ifmr->ifm_active = sc->sc_mii->mii_media_active;
ifmr->ifm_status = sc->sc_mii->mii_media_status;
Make hme(4) mpsafe - Let hme_start()/hme_init() acquire lock and then call hme_start_locked()/hme_init_locked() respectivly. - Teardown interrupt handler before hme_detach(). - Remove IFF_NEEDSGIANT flag and mark interrupt handler INTR_MPSAFE. - Set callout handler to CALLOUT_MPSAFE. - Add locks in hme MII interface. Reviewed by: jake Tested by: Julian C. Dunn <jdunn at opentrend dot net> MFC after: 2 weeks
2004-11-22 06:46:30 +00:00
HME_UNLOCK(sc);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
}
/*
* Process an ioctl request.
*/
static int
hme_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
{
struct hme_softc *sc = ifp->if_softc;
struct ifreq *ifr = (struct ifreq *)data;
- Use callout_init_mtx() to close races between hme_stop() and hme_tick(). - Use the driver lock instead of Giant in a bus dma callback. - Clear IFF_DRV_(RUNNING|OACTIVE) in hme_stop() instead of just clearing RUNNING in hme_ioctl() to be more like other ethernet drivers. - Lock the driver lock around mii operations. - Remove spls. - Cleanup locking in hme_ioctl(). MFC after: 1 week
2005-08-10 20:12:25 +00:00
int error = 0;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
switch (cmd) {
case SIOCSIFFLAGS:
- Use callout_init_mtx() to close races between hme_stop() and hme_tick(). - Use the driver lock instead of Giant in a bus dma callback. - Clear IFF_DRV_(RUNNING|OACTIVE) in hme_stop() instead of just clearing RUNNING in hme_ioctl() to be more like other ethernet drivers. - Lock the driver lock around mii operations. - Remove spls. - Cleanup locking in hme_ioctl(). MFC after: 1 week
2005-08-10 20:12:25 +00:00
HME_LOCK(sc);
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
if ((ifp->if_flags & IFF_UP) != 0) {
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0 &&
((ifp->if_flags ^ sc->sc_ifflags) &
(IFF_ALLMULTI | IFF_PROMISC)) != 0)
hme_setladrf(sc, 1);
else
hme_init_locked(sc);
} else if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
hme_stop(sc);
Implement TCP/UDP Transmit/Receive checksum offload. Since HME doesn't compensate the checksum for UDP datagram which can yield to 0x0, UDP transmit checksum offload is disabled by default. The UDP Transmit checksum offload can be reactivated by setting special link option link0 with ifconfig(8). Approved by: jake (mentor) Reviewed by: tmm Tested by: Herve Boulouis <amon@sockar.homeip.net>
2004-08-05 02:52:33 +00:00
if ((ifp->if_flags & IFF_LINK0) != 0)
sc->sc_csum_features |= CSUM_UDP;
else
sc->sc_csum_features &= ~CSUM_UDP;
if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
ifp->if_hwassist = sc->sc_csum_features;
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
sc->sc_ifflags = ifp->if_flags;
- Use callout_init_mtx() to close races between hme_stop() and hme_tick(). - Use the driver lock instead of Giant in a bus dma callback. - Clear IFF_DRV_(RUNNING|OACTIVE) in hme_stop() instead of just clearing RUNNING in hme_ioctl() to be more like other ethernet drivers. - Lock the driver lock around mii operations. - Remove spls. - Cleanup locking in hme_ioctl(). MFC after: 1 week
2005-08-10 20:12:25 +00:00
HME_UNLOCK(sc);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
break;
case SIOCADDMULTI:
case SIOCDELMULTI:
- Use callout_init_mtx() to close races between hme_stop() and hme_tick(). - Use the driver lock instead of Giant in a bus dma callback. - Clear IFF_DRV_(RUNNING|OACTIVE) in hme_stop() instead of just clearing RUNNING in hme_ioctl() to be more like other ethernet drivers. - Lock the driver lock around mii operations. - Remove spls. - Cleanup locking in hme_ioctl(). MFC after: 1 week
2005-08-10 20:12:25 +00:00
HME_LOCK(sc);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
hme_setladrf(sc, 1);
- Use callout_init_mtx() to close races between hme_stop() and hme_tick(). - Use the driver lock instead of Giant in a bus dma callback. - Clear IFF_DRV_(RUNNING|OACTIVE) in hme_stop() instead of just clearing RUNNING in hme_ioctl() to be more like other ethernet drivers. - Lock the driver lock around mii operations. - Remove spls. - Cleanup locking in hme_ioctl(). MFC after: 1 week
2005-08-10 20:12:25 +00:00
HME_UNLOCK(sc);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
error = 0;
break;
case SIOCGIFMEDIA:
case SIOCSIFMEDIA:
error = ifmedia_ioctl(ifp, ifr, &sc->sc_mii->mii_media, cmd);
break;
Implement TCP/UDP Transmit/Receive checksum offload. Since HME doesn't compensate the checksum for UDP datagram which can yield to 0x0, UDP transmit checksum offload is disabled by default. The UDP Transmit checksum offload can be reactivated by setting special link option link0 with ifconfig(8). Approved by: jake (mentor) Reviewed by: tmm Tested by: Herve Boulouis <amon@sockar.homeip.net>
2004-08-05 02:52:33 +00:00
case SIOCSIFCAP:
- Use callout_init_mtx() to close races between hme_stop() and hme_tick(). - Use the driver lock instead of Giant in a bus dma callback. - Clear IFF_DRV_(RUNNING|OACTIVE) in hme_stop() instead of just clearing RUNNING in hme_ioctl() to be more like other ethernet drivers. - Lock the driver lock around mii operations. - Remove spls. - Cleanup locking in hme_ioctl(). MFC after: 1 week
2005-08-10 20:12:25 +00:00
HME_LOCK(sc);
Implement TCP/UDP Transmit/Receive checksum offload. Since HME doesn't compensate the checksum for UDP datagram which can yield to 0x0, UDP transmit checksum offload is disabled by default. The UDP Transmit checksum offload can be reactivated by setting special link option link0 with ifconfig(8). Approved by: jake (mentor) Reviewed by: tmm Tested by: Herve Boulouis <amon@sockar.homeip.net>
2004-08-05 02:52:33 +00:00
ifp->if_capenable = ifr->ifr_reqcap;
if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
ifp->if_hwassist = sc->sc_csum_features;
else
ifp->if_hwassist = 0;
- Use callout_init_mtx() to close races between hme_stop() and hme_tick(). - Use the driver lock instead of Giant in a bus dma callback. - Clear IFF_DRV_(RUNNING|OACTIVE) in hme_stop() instead of just clearing RUNNING in hme_ioctl() to be more like other ethernet drivers. - Lock the driver lock around mii operations. - Remove spls. - Cleanup locking in hme_ioctl(). MFC after: 1 week
2005-08-10 20:12:25 +00:00
HME_UNLOCK(sc);
Implement TCP/UDP Transmit/Receive checksum offload. Since HME doesn't compensate the checksum for UDP datagram which can yield to 0x0, UDP transmit checksum offload is disabled by default. The UDP Transmit checksum offload can be reactivated by setting special link option link0 with ifconfig(8). Approved by: jake (mentor) Reviewed by: tmm Tested by: Herve Boulouis <amon@sockar.homeip.net>
2004-08-05 02:52:33 +00:00
break;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
default:
network interface driver changes: o don't strip the Ethernet header from inbound packets; pass packets up the stack intact (required significant changes to some drivers) o reference common definitions in net/ethernet.h (e.g. ETHER_ALIGN) o track ether_ifattach/ether_ifdetach API changes o track bpf changes (use BPF_TAP and BPF_MTAP) o track vlan changes (ifnet capabilities, revised processing scheme, etc.) o use if_input to pass packets "up" o call ether_ioctl for default handling of ioctls Reviewed by: many Approved by: re
2002-11-14 23:54:55 +00:00
error = ether_ioctl(ifp, cmd, data);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
break;
}
return (error);
}
/*
* Set up the logical address filter.
*/
static void
hme_setladrf(struct hme_softc *sc, int reenable)
{
Stop embedding struct ifnet at the top of driver softcs. Instead the struct ifnet or the layer 2 common structure it was embedded in have been replaced with a struct ifnet pointer to be filled by a call to the new function, if_alloc(). The layer 2 common structure is also allocated via if_alloc() based on the interface type. It is hung off the new struct ifnet member, if_l2com. This change removes the size of these structures from the kernel ABI and will allow us to better manage them as interfaces come and go. Other changes of note: - Struct arpcom is no longer referenced in normal interface code. Instead the Ethernet address is accessed via the IFP2ENADDR() macro. To enforce this ac_enaddr has been renamed to _ac_enaddr. - The second argument to ether_ifattach is now always the mac address from driver private storage rather than sometimes being ac_enaddr. Reviewed by: sobomax, sam
2005-06-10 16:49:24 +00:00
struct ifnet *ifp = sc->sc_ifp;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
struct ifmultiaddr *inm;
u_int32_t crc;
u_int32_t hash[4];
u_int32_t macc;
Make hme(4) mpsafe - Let hme_start()/hme_init() acquire lock and then call hme_start_locked()/hme_init_locked() respectivly. - Teardown interrupt handler before hme_detach(). - Remove IFF_NEEDSGIANT flag and mark interrupt handler INTR_MPSAFE. - Set callout handler to CALLOUT_MPSAFE. - Add locks in hme MII interface. Reviewed by: jake Tested by: Julian C. Dunn <jdunn at opentrend dot net> MFC after: 2 weeks
2004-11-22 06:46:30 +00:00
HME_LOCK_ASSERT(sc, MA_OWNED);
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
/* Clear the hash table. */
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
hash[3] = hash[2] = hash[1] = hash[0] = 0;
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
/* Get the current RX configuration. */
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
macc = HME_MAC_READ_4(sc, HME_MACI_RXCFG);
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
/*
* Turn off promiscuous mode, promiscuous group mode (all multicast),
* and hash filter. Depending on the case, the right bit will be
* enabled.
*/
macc &= ~(HME_MAC_RXCFG_PGRP | HME_MAC_RXCFG_PMISC);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
/*
* Disable the receiver while changing it's state as the documentation
* mandates.
* We then must wait until the bit clears in the register. This should
* take at most 3.5ms.
*/
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
if (!hme_mac_bitflip(sc, HME_MACI_RXCFG, macc,
HME_MAC_RXCFG_ENABLE, 0))
device_printf(sc->sc_dev, "cannot disable RX MAC\n");
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
/* Disable the hash filter before writing to the filter registers. */
if (!hme_mac_bitflip(sc, HME_MACI_RXCFG, macc,
HME_MAC_RXCFG_HENABLE, 0))
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
device_printf(sc->sc_dev, "cannot disable hash filter\n");
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
/* Make the RX MAC really SIMPLEX. */
According to STP2002QFP User's Guide, it seems that driver should program RXMAC to discard frames with SA field matching the stations's MAC address. Experimentation shows that HME receives its own frames when it operates at 10Mbps half-duplex. With this change HME runs at 10Mbps half-duplx should work with IPv6. (No more "DAD detected duplicate IPv6 address".) Reported by: jacques brierre <jbrierre AT bellsouth DOT net> Reviewed by: marius
2005-05-23 05:45:36 +00:00
macc |= HME_MAC_RXCFG_ME;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
if (reenable)
macc |= HME_MAC_RXCFG_ENABLE;
else
macc &= ~HME_MAC_RXCFG_ENABLE;
if ((ifp->if_flags & IFF_PROMISC) != 0) {
macc |= HME_MAC_RXCFG_PMISC;
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
goto chipit;
}
if ((ifp->if_flags & IFF_ALLMULTI) != 0) {
macc |= HME_MAC_RXCFG_PGRP;
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
goto chipit;
}
macc |= HME_MAC_RXCFG_HENABLE;
/*
* Set up multicast address filter by passing all multicast addresses
* through a crc generator, and then using the high order 6 bits as an
* index into the 64 bit logical address filter. The high order bit
* selects the word, while the rest of the bits select the bit within
* the word.
*/
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
IF_ADDR_LOCK(ifp);
TAILQ_FOREACH(inm, &ifp->if_multiaddrs, ifma_link) {
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
if (inm->ifma_addr->sa_family != AF_LINK)
continue;
- Add a LLADDR() forgotten in the conversion to ether_crc32_le(). - Remove a variable no longer used after the conversion. - While here, save on another one no longer really necessary after the conversion.
2004-06-10 00:06:04 +00:00
crc = ether_crc32_le(LLADDR((struct sockaddr_dl *)
inm->ifma_addr), ETHER_ADDR_LEN);
Replace handrolled CRC calculation with ether_crc32_[lb]e().
2004-06-09 14:34:04 +00:00
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
/* Just want the 6 most significant bits. */
crc >>= 26;
/* Set the corresponding bit in the filter. */
hash[crc >> 4] |= 1 << (crc & 0xf);
}
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
IF_ADDR_UNLOCK(ifp);
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
chipit:
/* Now load the hash table into the chip */
HME_MAC_WRITE_4(sc, HME_MACI_HASHTAB0, hash[0]);
HME_MAC_WRITE_4(sc, HME_MACI_HASHTAB1, hash[1]);
HME_MAC_WRITE_4(sc, HME_MACI_HASHTAB2, hash[2]);
HME_MAC_WRITE_4(sc, HME_MACI_HASHTAB3, hash[3]);
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
if (!hme_mac_bitflip(sc, HME_MACI_RXCFG, macc, 0,
According to STP2002QFP User's Guide, it seems that driver should program RXMAC to discard frames with SA field matching the stations's MAC address. Experimentation shows that HME receives its own frames when it operates at 10Mbps half-duplex. With this change HME runs at 10Mbps half-duplx should work with IPv6. (No more "DAD detected duplicate IPv6 address".) Reported by: jacques brierre <jbrierre AT bellsouth DOT net> Reviewed by: marius
2005-05-23 05:45:36 +00:00
macc & (HME_MAC_RXCFG_ENABLE | HME_MAC_RXCFG_HENABLE |
o Disable HMEDEBUG by default. o Add CTASSERTs ensuring that HME_NRXDESC and HME_NTXDESC are set to legal values. o Use appropriate maxsize, nsegments and maxsegsize parameters when creating DMA tags and correct some comments related to them. o The FreeBSD bus_dmamap_sync(9) supports ored together flags for quite some time now so collapse calls accordingly. o Add missing BUS_DMASYNC_PREREAD when syncing the control DMA maps in hme_rint() and hme_start_locked(). o Keep state of the link state and use it to enable or disable the MAC in hme_mii_statchg() accordingly as well as to return early from hme_start_locked() in case the link is down. o Introduce a sc_flags and use it to replace individual members like sc_pci. o Add bus_barrier(9) calls to hme_mac_bitflip(), hme_mii_readreg(), hme_mii_writereg() and hme_stop() to ensure the respective bit has been written before we starting polling on it and for the right bits to change. o Rather just returning in case hme_mac_bitflip() fails and leaving us in an undefined state report the problem and move on; chances are the requested configuration will become active shortly after. o Don't call hme_start_locked() in hme_init_locked() unconditionally but only after calls to hme_init_locked() when it's appropriate, i.e. in hme_watchdog(). o Add a KASSERT which asserts nsegs is valid also to hme_load_txmbuf(). o In hme_load_txmbuf(): - use a maximum of the newly introduced HME_NTXSEGS segments instead of the incorrect HME_NTXQ, which reflects the maximum TX queue length, for loading the mbufs and put the DMA segments back onto the stack instead of the softc as 16 should be ok there. - use the common errno(2) return values instead of homegrown ones, - given that hme_load_txmbuf() is allowed to fail resulting in a packet drop for quite some time now implement the functionality of hme_txcksum() by means of m_pullup(9), which de-obfuscates the code and allows to always retrieve the correct length of the IP header, [1] - also add a KASSERT which asserts nsegs is valid, - take advantage of m_collapse(9) instead of m_defrag(9) for performance reasons. o Don't bother to check whether the interface is running or whether its queue is empty before calling hme_start_locked() in hme_tint(), the former will check these anyway. o In hme_intr() call hme_rint() before hme_tint() as gem_tint() may take quite a while to return when it calls hme_start_locked(). o Get rid of sc_debug and just check if_flags for IFF_DEBUG directly. o Add a shadow sc_ifflags so we don't reset the chip when unnecessary. o Handle IFF_ALLMULTI correctly. [2] o Use PCIR_BAR instead of a homegrown macro. o Replace sc_enaddr[6] with sc_enaddr[ETHER_ADDR_LEN]. o Use the maximum of 256 TX descriptors for better performance as using all of them has no additional static cost rather than using just half of them. Reported by: rwatson [2] Suggested by: yongari [1] Reviewed by: yongari MFC after: 1 month
2008-04-24 23:12:03 +00:00
HME_MAC_RXCFG_ME)))
device_printf(sc->sc_dev, "cannot configure RX MAC\n");
Add a driver for the Sun HME PCI/SBus ethernet adaptor, which is onboard in most machines of the Sun Ultra series. This is a port of the NetBSD driver which I enhanced to make use of the gather functionality and the configurable RX buffer offset to avoid copying all received/sent packet (instead, packets will be directly DMAd from mbuf chains and into mbuf clusters now).
2002-02-27 17:35:48 +00:00
}
Reference in New Issue Copy Permalink