freebsd-skq/sys/pci/if_de.c

5052 lines
152 KiB
C
Raw Normal View History

/* $NetBSD: if_de.c,v 1.86 1999/06/01 19:17:59 thorpej Exp $ */
/*-
* Copyright (c) 1994-1997 Matt Thomas (matt@3am-software.com)
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. The name of the author may not be used to endorse or promote products
* derived from this software without specific prior written permission
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``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 AUTHOR 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.
*
* Id: if_de.c,v 1.94 1997/07/03 16:55:07 thomas Exp
*/
/*
* DEC 21040 PCI Ethernet Controller
*
* Written by Matt Thomas
* BPF support code stolen directly from if_ec.c
*
* This driver supports the DEC DE435 or any other PCI
* board which support 21040, 21041, or 21140 (mostly).
*/
2003-06-11 06:34:30 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#define TULIP_HDR_DATA
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
#include "opt_ddb.h"
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/endian.h>
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
#include <sys/ktr.h>
#include <sys/mbuf.h>
#include <sys/socket.h>
#include <sys/sockio.h>
#include <sys/malloc.h>
#include <sys/kernel.h>
2004-05-30 20:00:41 +00:00
#include <sys/module.h>
#include <sys/eventhandler.h>
#include <machine/bus.h>
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
#include <machine/bus_dma.h>
#include <machine/resource.h>
#include <sys/bus.h>
#include <sys/rman.h>
#include <net/if.h>
#include <net/if_arp.h>
#include <net/ethernet.h>
#include <net/if_media.h>
#include <net/if_types.h>
#include <net/if_dl.h>
#include <net/bpf.h>
#ifdef INET
#include <netinet/in.h>
#include <netinet/if_ether.h>
#endif
#include <vm/vm.h>
#include <net/if_var.h>
#include <vm/pmap.h>
#include <dev/pci/pcivar.h>
#include <dev/pci/pcireg.h>
#include <pci/dc21040reg.h>
1999-08-09 14:43:39 +00:00
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
#ifdef DDB
#include <ddb/ddb.h>
#endif
/*
* Intel CPUs should use I/O mapped access.
*/
1999-08-09 14:43:39 +00:00
#if defined(__i386__)
#define TULIP_IOMAPPED
#endif
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
#if 0
/* This enables KTR traces at KTR_DEV. */
#define KTR_TULIP KTR_DEV
#else
#define KTR_TULIP 0
#endif
#if 0
/*
* This turns on all sort of debugging stuff and make the
* driver much larger.
*/
#define TULIP_DEBUG
#endif
1997-08-03 13:00:42 +00:00
#if 0
#define TULIP_PERFSTATS
#endif
#define TULIP_HZ 10
#include <pci/if_devar.h>
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
#define SYNC_NONE 0
#define SYNC_RX 1
#define SYNC_TX 2
/*
* This module supports
* the DEC 21040 PCI Ethernet Controller.
* the DEC 21041 PCI Ethernet Controller.
* the DEC 21140 PCI Fast Ethernet Controller.
*/
2004-07-05 22:27:26 +00:00
static void tulip_addr_filter(tulip_softc_t * const sc);
static int tulip_ifmedia_change(struct ifnet * const ifp);
static void tulip_ifmedia_status(struct ifnet * const ifp,
struct ifmediareq *req);
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
static void tulip_init(void *);
static void tulip_init_locked(tulip_softc_t * const sc);
2004-07-05 22:27:26 +00:00
static void tulip_intr_shared(void *arg);
static void tulip_intr_normal(void *arg);
static void tulip_mii_autonegotiate(tulip_softc_t * const sc,
const unsigned phyaddr);
static int tulip_mii_map_abilities(tulip_softc_t * const sc,
unsigned abilities);
static tulip_media_t
tulip_mii_phy_readspecific(tulip_softc_t * const sc);
static unsigned tulip_mii_readreg(tulip_softc_t * const sc, unsigned devaddr,
unsigned regno);
static void tulip_mii_writereg(tulip_softc_t * const sc, unsigned devaddr,
unsigned regno, unsigned data);
static void tulip_reset(tulip_softc_t * const sc);
static void tulip_rx_intr(tulip_softc_t * const sc);
static int tulip_srom_decode(tulip_softc_t * const sc);
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
static void tulip_start(struct ifnet *ifp);
static void tulip_start_locked(tulip_softc_t * const sc);
2004-07-05 22:27:26 +00:00
static struct mbuf *
tulip_txput(tulip_softc_t * const sc, struct mbuf *m);
static void tulip_txput_setup(tulip_softc_t * const sc);
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
struct mbuf * tulip_dequeue_mbuf(tulip_ringinfo_t *ri, tulip_descinfo_t *di,
int sync);
static void tulip_dma_map_addr(void *, bus_dma_segment_t *, int, int);
static void tulip_dma_map_rxbuf(void *, bus_dma_segment_t *, int,
bus_size_t, int);
static void
tulip_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
{
u_int32_t *paddr;
if (error)
return;
paddr = arg;
*paddr = segs->ds_addr;
}
static void
tulip_dma_map_rxbuf(void *arg, bus_dma_segment_t *segs, int nseg,
bus_size_t mapsize, int error)
{
tulip_desc_t *desc;
if (error)
return;
desc = arg;
KASSERT(nseg == 1, ("too many DMA segments"));
KASSERT(segs[0].ds_len >= TULIP_RX_BUFLEN, ("receive buffer too small"));
desc->d_addr1 = segs[0].ds_addr;
desc->d_length1 = TULIP_RX_BUFLEN;
#ifdef not_needed
/* These should already always be zero. */
desc->d_addr2 = 0;
desc->d_length2 = 0;
#endif
}
struct mbuf *
tulip_dequeue_mbuf(tulip_ringinfo_t *ri, tulip_descinfo_t *di, int sync)
{
struct mbuf *m;
m = di->di_mbuf;
if (m != NULL) {
switch (sync) {
case SYNC_NONE:
break;
case SYNC_RX:
TULIP_RXMAP_POSTSYNC(ri, di);
break;
case SYNC_TX:
TULIP_TXMAP_POSTSYNC(ri, di);
break;
default:
panic("bad sync flag: %d", sync);
}
bus_dmamap_unload(ri->ri_data_tag, *di->di_map);
di->di_mbuf = NULL;
}
return (m);
}
2004-07-05 22:27:26 +00:00
static void
tulip_timeout_callback(void *arg)
{
tulip_softc_t * const sc = arg;
1997-08-03 13:00:42 +00:00
TULIP_PERFSTART(timeout)
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
TULIP_LOCK_ASSERT(sc);
1997-08-03 13:00:42 +00:00
sc->tulip_flags &= ~TULIP_TIMEOUTPENDING;
sc->tulip_probe_timeout -= 1000 / TULIP_HZ;
(sc->tulip_boardsw->bd_media_poll)(sc, TULIP_MEDIAPOLL_TIMER);
1997-08-03 13:00:42 +00:00
TULIP_PERFEND(timeout);
}
static void
tulip_timeout(tulip_softc_t * const sc)
{
TULIP_LOCK_ASSERT(sc);
if (sc->tulip_flags & TULIP_TIMEOUTPENDING)
return;
sc->tulip_flags |= TULIP_TIMEOUTPENDING;
callout_reset(&sc->tulip_callout, (hz + TULIP_HZ / 2) / TULIP_HZ,
tulip_timeout_callback, sc);
}
static int
tulip_txprobe(tulip_softc_t * const sc)
{
struct mbuf *m;
/*
* Before we are sure this is the right media we need
* to send a small packet to make sure there's carrier.
1997-08-03 13:00:42 +00:00
* Strangely, BNC and AUI will "see" receive data if
* either is connected so the transmit is the only way
* to verify the connectivity.
*/
TULIP_LOCK_ASSERT(sc);
MGETHDR(m, M_DONTWAIT, MT_DATA);
if (m == NULL)
return 0;
/*
* Construct a LLC TEST message which will point to ourselves.
*/
bcopy(IF_LLADDR(sc->tulip_ifp), mtod(m, struct ether_header *)->ether_dhost, 6);
bcopy(IF_LLADDR(sc->tulip_ifp), mtod(m, struct ether_header *)->ether_shost, 6);
mtod(m, struct ether_header *)->ether_type = htons(3);
mtod(m, unsigned char *)[14] = 0;
mtod(m, unsigned char *)[15] = 0;
mtod(m, unsigned char *)[16] = 0xE3; /* LLC Class1 TEST (no poll) */
m->m_len = m->m_pkthdr.len = sizeof(struct ether_header) + 3;
/*
* send it!
*/
sc->tulip_cmdmode |= TULIP_CMD_TXRUN;
1997-08-03 13:00:42 +00:00
sc->tulip_intrmask |= TULIP_STS_TXINTR;
sc->tulip_flags |= TULIP_TXPROBE_ACTIVE;
TULIP_CSR_WRITE(sc, csr_command, sc->tulip_cmdmode);
1997-08-03 13:00:42 +00:00
TULIP_CSR_WRITE(sc, csr_intr, sc->tulip_intrmask);
if ((m = tulip_txput(sc, m)) != NULL)
m_freem(m);
sc->tulip_probe.probe_txprobes++;
return 1;
}
static void
tulip_media_set(tulip_softc_t * const sc, tulip_media_t media)
{
const tulip_media_info_t *mi = sc->tulip_mediums[media];
TULIP_LOCK_ASSERT(sc);
if (mi == NULL)
return;
/*
* If we are switching media, make sure we don't think there's
* any stale RX activity
*/
sc->tulip_flags &= ~TULIP_RXACT;
if (mi->mi_type == TULIP_MEDIAINFO_SIA) {
TULIP_CSR_WRITE(sc, csr_sia_connectivity, TULIP_SIACONN_RESET);
TULIP_CSR_WRITE(sc, csr_sia_tx_rx, mi->mi_sia_tx_rx);
if (sc->tulip_features & TULIP_HAVE_SIAGP) {
TULIP_CSR_WRITE(sc, csr_sia_general, mi->mi_sia_gp_control|mi->mi_sia_general);
DELAY(50);
TULIP_CSR_WRITE(sc, csr_sia_general, mi->mi_sia_gp_data|mi->mi_sia_general);
} else {
TULIP_CSR_WRITE(sc, csr_sia_general, mi->mi_sia_general);
}
TULIP_CSR_WRITE(sc, csr_sia_connectivity, mi->mi_sia_connectivity);
} else if (mi->mi_type == TULIP_MEDIAINFO_GPR) {
#define TULIP_GPR_CMDBITS (TULIP_CMD_PORTSELECT|TULIP_CMD_PCSFUNCTION|TULIP_CMD_SCRAMBLER|TULIP_CMD_TXTHRSHLDCTL)
/*
* If the cmdmode bits don't match the currently operating mode,
* set the cmdmode appropriately and reset the chip.
*/
if (((mi->mi_cmdmode ^ TULIP_CSR_READ(sc, csr_command)) & TULIP_GPR_CMDBITS) != 0) {
sc->tulip_cmdmode &= ~TULIP_GPR_CMDBITS;
sc->tulip_cmdmode |= mi->mi_cmdmode;
tulip_reset(sc);
}
TULIP_CSR_WRITE(sc, csr_gp, TULIP_GP_PINSET|sc->tulip_gpinit);
DELAY(10);
TULIP_CSR_WRITE(sc, csr_gp, (u_int8_t) mi->mi_gpdata);
} else if (mi->mi_type == TULIP_MEDIAINFO_SYM) {
/*
* If the cmdmode bits don't match the currently operating mode,
* set the cmdmode appropriately and reset the chip.
*/
if (((mi->mi_cmdmode ^ TULIP_CSR_READ(sc, csr_command)) & TULIP_GPR_CMDBITS) != 0) {
sc->tulip_cmdmode &= ~TULIP_GPR_CMDBITS;
sc->tulip_cmdmode |= mi->mi_cmdmode;
tulip_reset(sc);
}
TULIP_CSR_WRITE(sc, csr_sia_general, mi->mi_gpcontrol);
TULIP_CSR_WRITE(sc, csr_sia_general, mi->mi_gpdata);
} else if (mi->mi_type == TULIP_MEDIAINFO_MII
&& sc->tulip_probe_state != TULIP_PROBE_INACTIVE) {
int idx;
if (sc->tulip_features & TULIP_HAVE_SIAGP) {
const u_int8_t *dp;
dp = &sc->tulip_rombuf[mi->mi_reset_offset];
for (idx = 0; idx < mi->mi_reset_length; idx++, dp += 2) {
DELAY(10);
TULIP_CSR_WRITE(sc, csr_sia_general, (dp[0] + 256 * dp[1]) << 16);
}
sc->tulip_phyaddr = mi->mi_phyaddr;
dp = &sc->tulip_rombuf[mi->mi_gpr_offset];
for (idx = 0; idx < mi->mi_gpr_length; idx++, dp += 2) {
DELAY(10);
TULIP_CSR_WRITE(sc, csr_sia_general, (dp[0] + 256 * dp[1]) << 16);
}
} else {
for (idx = 0; idx < mi->mi_reset_length; idx++) {
DELAY(10);
TULIP_CSR_WRITE(sc, csr_gp, sc->tulip_rombuf[mi->mi_reset_offset + idx]);
}
sc->tulip_phyaddr = mi->mi_phyaddr;
for (idx = 0; idx < mi->mi_gpr_length; idx++) {
DELAY(10);
TULIP_CSR_WRITE(sc, csr_gp, sc->tulip_rombuf[mi->mi_gpr_offset + idx]);
}
}
if (sc->tulip_flags & TULIP_TRYNWAY) {
tulip_mii_autonegotiate(sc, sc->tulip_phyaddr);
} else if ((sc->tulip_flags & TULIP_DIDNWAY) == 0) {
u_int32_t data = tulip_mii_readreg(sc, sc->tulip_phyaddr, PHYREG_CONTROL);
data &= ~(PHYCTL_SELECT_100MB|PHYCTL_FULL_DUPLEX|PHYCTL_AUTONEG_ENABLE);
sc->tulip_flags &= ~TULIP_DIDNWAY;
if (TULIP_IS_MEDIA_FD(media))
data |= PHYCTL_FULL_DUPLEX;
if (TULIP_IS_MEDIA_100MB(media))
data |= PHYCTL_SELECT_100MB;
tulip_mii_writereg(sc, sc->tulip_phyaddr, PHYREG_CONTROL, data);
}
}
}
static void
tulip_linkup(tulip_softc_t * const sc, tulip_media_t media)
{
TULIP_LOCK_ASSERT(sc);
if ((sc->tulip_flags & TULIP_LINKUP) == 0)
sc->tulip_flags |= TULIP_PRINTLINKUP;
sc->tulip_flags |= TULIP_LINKUP;
sc->tulip_ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
#if 0 /* XXX how does with work with ifmedia? */
if ((sc->tulip_flags & TULIP_DIDNWAY) == 0) {
if (sc->tulip_ifp->if_flags & IFF_FULLDUPLEX) {
if (TULIP_CAN_MEDIA_FD(media)
&& sc->tulip_mediums[TULIP_FD_MEDIA_OF(media)] != NULL)
media = TULIP_FD_MEDIA_OF(media);
} else {
if (TULIP_IS_MEDIA_FD(media)
&& sc->tulip_mediums[TULIP_HD_MEDIA_OF(media)] != NULL)
media = TULIP_HD_MEDIA_OF(media);
}
}
#endif
if (sc->tulip_media != media) {
#ifdef TULIP_DEBUG
sc->tulip_dbg.dbg_last_media = sc->tulip_media;
#endif
sc->tulip_media = media;
sc->tulip_flags |= TULIP_PRINTMEDIA;
if (TULIP_IS_MEDIA_FD(sc->tulip_media)) {
sc->tulip_cmdmode |= TULIP_CMD_FULLDUPLEX;
} else if (sc->tulip_chipid != TULIP_21041 || (sc->tulip_flags & TULIP_DIDNWAY) == 0) {
sc->tulip_cmdmode &= ~TULIP_CMD_FULLDUPLEX;
}
}
/*
* We could set probe_timeout to 0 but setting to 3000 puts this
* in one central place and the only matters is tulip_link is
* followed by a tulip_timeout. Therefore setting it should not
* result in aberrant behavour.
*/
sc->tulip_probe_timeout = 3000;
sc->tulip_probe_state = TULIP_PROBE_INACTIVE;
sc->tulip_flags &= ~(TULIP_TXPROBE_ACTIVE|TULIP_TRYNWAY);
if (sc->tulip_flags & TULIP_INRESET) {
tulip_media_set(sc, sc->tulip_media);
} else if (sc->tulip_probe_media != sc->tulip_media) {
/*
* No reason to change media if we have the right media.
*/
tulip_reset(sc);
}
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
tulip_init_locked(sc);
}
static void
tulip_media_print(tulip_softc_t * const sc)
{
struct ifnet *ifp = sc->tulip_ifp;
TULIP_LOCK_ASSERT(sc);
if ((sc->tulip_flags & TULIP_LINKUP) == 0)
return;
if (sc->tulip_flags & TULIP_PRINTMEDIA) {
if_printf(ifp, "enabling %s port\n",
tulip_mediums[sc->tulip_media]);
sc->tulip_flags &= ~(TULIP_PRINTMEDIA|TULIP_PRINTLINKUP);
} else if (sc->tulip_flags & TULIP_PRINTLINKUP) {
if_printf(ifp, "link up\n");
sc->tulip_flags &= ~TULIP_PRINTLINKUP;
}
}
#if defined(TULIP_DO_GPR_SENSE)
static tulip_media_t
tulip_21140_gpr_media_sense(tulip_softc_t * const sc)
{
struct ifnet *ifp sc->tulip_ifp;
tulip_media_t maybe_media = TULIP_MEDIA_UNKNOWN;
tulip_media_t last_media = TULIP_MEDIA_UNKNOWN;
tulip_media_t media;
TULIP_LOCK_ASSERT(sc);
/*
* If one of the media blocks contained a default media flag,
* use that.
*/
for (media = TULIP_MEDIA_UNKNOWN; media < TULIP_MEDIA_MAX; media++) {
const tulip_media_info_t *mi;
/*
* Media is not supported (or is full-duplex).
*/
if ((mi = sc->tulip_mediums[media]) == NULL || TULIP_IS_MEDIA_FD(media))
continue;
if (mi->mi_type != TULIP_MEDIAINFO_GPR)
continue;
/*
* Remember the media is this is the "default" media.
*/
if (mi->mi_default && maybe_media == TULIP_MEDIA_UNKNOWN)
maybe_media = media;
/*
* No activity mask? Can't see if it is active if there's no mask.
*/
if (mi->mi_actmask == 0)
continue;
/*
* Does the activity data match?
*/
if ((TULIP_CSR_READ(sc, csr_gp) & mi->mi_actmask) != mi->mi_actdata)
continue;
#if defined(TULIP_DEBUG)
if_printf(ifp, "gpr_media_sense: %s: 0x%02x & 0x%02x == 0x%02x\n",
tulip_mediums[media],
TULIP_CSR_READ(sc, csr_gp) & 0xFF,
mi->mi_actmask, mi->mi_actdata);
#endif
/*
* It does! If this is the first media we detected, then
* remember this media. If isn't the first, then there were
* multiple matches which we equate to no match (since we don't
* which to select (if any).
*/
if (last_media == TULIP_MEDIA_UNKNOWN) {
last_media = media;
} else if (last_media != media) {
last_media = TULIP_MEDIA_UNKNOWN;
}
}
return (last_media != TULIP_MEDIA_UNKNOWN) ? last_media : maybe_media;
}
#endif /* TULIP_DO_GPR_SENSE */
static tulip_link_status_t
tulip_media_link_monitor(tulip_softc_t * const sc)
{
struct ifnet *ifp = sc->tulip_ifp;
const tulip_media_info_t * const mi = sc->tulip_mediums[sc->tulip_media];
tulip_link_status_t linkup = TULIP_LINK_DOWN;
TULIP_LOCK_ASSERT(sc);
if (mi == NULL) {
#if defined(DIAGNOSTIC) || defined(TULIP_DEBUG)
panic("tulip_media_link_monitor: %s: botch at line %d\n",
tulip_mediums[sc->tulip_media],__LINE__);
#else
return TULIP_LINK_UNKNOWN;
#endif
}
/*
* Have we seen some packets? If so, the link must be good.
*/
if ((sc->tulip_flags & (TULIP_RXACT|TULIP_LINKUP)) == (TULIP_RXACT|TULIP_LINKUP)) {
sc->tulip_flags &= ~TULIP_RXACT;
sc->tulip_probe_timeout = 3000;
return TULIP_LINK_UP;
}
sc->tulip_flags &= ~TULIP_RXACT;
if (mi->mi_type == TULIP_MEDIAINFO_MII) {
u_int32_t status;
/*
* Read the PHY status register.
*/
status = tulip_mii_readreg(sc, sc->tulip_phyaddr, PHYREG_STATUS);
if (status & PHYSTS_AUTONEG_DONE) {
/*
* If the PHY has completed autonegotiation, see the if the
* remote systems abilities have changed. If so, upgrade or
* downgrade as appropriate.
*/
u_int32_t abilities = tulip_mii_readreg(sc, sc->tulip_phyaddr, PHYREG_AUTONEG_ABILITIES);
abilities = (abilities << 6) & status;
if (abilities != sc->tulip_abilities) {
#if defined(TULIP_DEBUG)
loudprintf("%s(phy%d): autonegotiation changed: 0x%04x -> 0x%04x\n",
ifp->if_xname, sc->tulip_phyaddr,
sc->tulip_abilities, abilities);
#endif
if (tulip_mii_map_abilities(sc, abilities)) {
tulip_linkup(sc, sc->tulip_probe_media);
return TULIP_LINK_UP;
}
/*
* if we had selected media because of autonegotiation,
* we need to probe for the new media.
*/
sc->tulip_probe_state = TULIP_PROBE_INACTIVE;
if (sc->tulip_flags & TULIP_DIDNWAY)
return TULIP_LINK_DOWN;
}
}
/*
* The link is now up. If was down, say its back up.
*/
if ((status & (PHYSTS_LINK_UP|PHYSTS_REMOTE_FAULT)) == PHYSTS_LINK_UP)
linkup = TULIP_LINK_UP;
} else if (mi->mi_type == TULIP_MEDIAINFO_GPR) {
/*
* No activity sensor? Assume all's well.
*/
if (mi->mi_actmask == 0)
return TULIP_LINK_UNKNOWN;
/*
* Does the activity data match?
*/
if ((TULIP_CSR_READ(sc, csr_gp) & mi->mi_actmask) == mi->mi_actdata)
linkup = TULIP_LINK_UP;
} else if (mi->mi_type == TULIP_MEDIAINFO_SIA) {
/*
* Assume non TP ok for now.
*/
if (!TULIP_IS_MEDIA_TP(sc->tulip_media))
return TULIP_LINK_UNKNOWN;
if ((TULIP_CSR_READ(sc, csr_sia_status) & TULIP_SIASTS_LINKFAIL) == 0)
linkup = TULIP_LINK_UP;
#if defined(TULIP_DEBUG)
if (sc->tulip_probe_timeout <= 0)
if_printf(ifp, "sia status = 0x%08x\n",
TULIP_CSR_READ(sc, csr_sia_status));
#endif
} else if (mi->mi_type == TULIP_MEDIAINFO_SYM) {
return TULIP_LINK_UNKNOWN;
}
/*
* We will wait for 3 seconds until the link goes into suspect mode.
*/
if (sc->tulip_flags & TULIP_LINKUP) {
if (linkup == TULIP_LINK_UP)
sc->tulip_probe_timeout = 3000;
if (sc->tulip_probe_timeout > 0)
return TULIP_LINK_UP;
sc->tulip_flags &= ~TULIP_LINKUP;
if_printf(ifp, "link down: cable problem?\n");
}
#if defined(TULIP_DEBUG)
sc->tulip_dbg.dbg_link_downed++;
#endif
return TULIP_LINK_DOWN;
}
static void
tulip_media_poll(tulip_softc_t * const sc, tulip_mediapoll_event_t event)
{
struct ifnet *ifp = sc->tulip_ifp;
TULIP_LOCK_ASSERT(sc);
#if defined(TULIP_DEBUG)
sc->tulip_dbg.dbg_events[event]++;
#endif
if (sc->tulip_probe_state == TULIP_PROBE_INACTIVE
&& event == TULIP_MEDIAPOLL_TIMER) {
switch (tulip_media_link_monitor(sc)) {
case TULIP_LINK_DOWN: {
/*
* Link Monitor failed. Probe for new media.
*/
event = TULIP_MEDIAPOLL_LINKFAIL;
break;
}
case TULIP_LINK_UP: {
/*
* Check again soon.
*/
tulip_timeout(sc);
return;
}
case TULIP_LINK_UNKNOWN: {
/*
* We can't tell so don't bother.
*/
return;
}
}
}
if (event == TULIP_MEDIAPOLL_LINKFAIL) {
if (sc->tulip_probe_state == TULIP_PROBE_INACTIVE) {
if (TULIP_DO_AUTOSENSE(sc)) {
#if defined(TULIP_DEBUG)
sc->tulip_dbg.dbg_link_failures++;
#endif
sc->tulip_media = TULIP_MEDIA_UNKNOWN;
if (sc->tulip_ifp->if_flags & IFF_UP)
tulip_reset(sc); /* restart probe */
}
return;
}
#if defined(TULIP_DEBUG)
sc->tulip_dbg.dbg_link_pollintrs++;
#endif
}
if (event == TULIP_MEDIAPOLL_START) {
sc->tulip_ifp->if_drv_flags |= IFF_DRV_OACTIVE;
if (sc->tulip_probe_state != TULIP_PROBE_INACTIVE)
return;
sc->tulip_probe_mediamask = 0;
sc->tulip_probe_passes = 0;
#if defined(TULIP_DEBUG)
sc->tulip_dbg.dbg_media_probes++;
#endif
/*
* If the SROM contained an explicit media to use, use it.
*/
sc->tulip_cmdmode &= ~(TULIP_CMD_RXRUN|TULIP_CMD_FULLDUPLEX);
sc->tulip_flags |= TULIP_TRYNWAY|TULIP_PROBE1STPASS;
sc->tulip_flags &= ~(TULIP_DIDNWAY|TULIP_PRINTMEDIA|TULIP_PRINTLINKUP);
/*
* connidx is defaulted to a media_unknown type.
*/
sc->tulip_probe_media = tulip_srom_conninfo[sc->tulip_connidx].sc_media;
if (sc->tulip_probe_media != TULIP_MEDIA_UNKNOWN) {
tulip_linkup(sc, sc->tulip_probe_media);
tulip_timeout(sc);
return;
}
if (sc->tulip_features & TULIP_HAVE_GPR) {
sc->tulip_probe_state = TULIP_PROBE_GPRTEST;
sc->tulip_probe_timeout = 2000;
} else {
sc->tulip_probe_media = TULIP_MEDIA_MAX;
sc->tulip_probe_timeout = 0;
sc->tulip_probe_state = TULIP_PROBE_MEDIATEST;
}
}
/*
* Ignore txprobe failures or spurious callbacks.
*/
if (event == TULIP_MEDIAPOLL_TXPROBE_FAILED
&& sc->tulip_probe_state != TULIP_PROBE_MEDIATEST) {
sc->tulip_flags &= ~TULIP_TXPROBE_ACTIVE;
return;
}
/*
* If we really transmitted a packet, then that's the media we'll use.
*/
if (event == TULIP_MEDIAPOLL_TXPROBE_OK || event == TULIP_MEDIAPOLL_LINKPASS) {
if (event == TULIP_MEDIAPOLL_LINKPASS) {
/* XXX Check media status just to be sure */
sc->tulip_probe_media = TULIP_MEDIA_10BASET;
#if defined(TULIP_DEBUG)
} else {
sc->tulip_dbg.dbg_txprobes_ok[sc->tulip_probe_media]++;
#endif
}
tulip_linkup(sc, sc->tulip_probe_media);
tulip_timeout(sc);
return;
}
if (sc->tulip_probe_state == TULIP_PROBE_GPRTEST) {
#if defined(TULIP_DO_GPR_SENSE)
/*
* Check for media via the general purpose register.
*
* Try to sense the media via the GPR. If the same value
* occurs 3 times in a row then just use that.
*/
if (sc->tulip_probe_timeout > 0) {
tulip_media_t new_probe_media = tulip_21140_gpr_media_sense(sc);
#if defined(TULIP_DEBUG)
if_printf(ifp, "media_poll: gpr sensing = %s\n",
tulip_mediums[new_probe_media]);
#endif
if (new_probe_media != TULIP_MEDIA_UNKNOWN) {
if (new_probe_media == sc->tulip_probe_media) {
if (--sc->tulip_probe_count == 0)
tulip_linkup(sc, sc->tulip_probe_media);
} else {
sc->tulip_probe_count = 10;
}
}
sc->tulip_probe_media = new_probe_media;
tulip_timeout(sc);
return;
}
#endif /* TULIP_DO_GPR_SENSE */
/*
* Brute force. We cycle through each of the media types
* and try to transmit a packet.
*/
sc->tulip_probe_state = TULIP_PROBE_MEDIATEST;
sc->tulip_probe_media = TULIP_MEDIA_MAX;
sc->tulip_probe_timeout = 0;
tulip_timeout(sc);
return;
}
if (sc->tulip_probe_state != TULIP_PROBE_MEDIATEST
&& (sc->tulip_features & TULIP_HAVE_MII)) {
tulip_media_t old_media = sc->tulip_probe_media;
tulip_mii_autonegotiate(sc, sc->tulip_phyaddr);
switch (sc->tulip_probe_state) {
case TULIP_PROBE_FAILED:
case TULIP_PROBE_MEDIATEST: {
/*
* Try the next media.
*/
sc->tulip_probe_mediamask |= sc->tulip_mediums[sc->tulip_probe_media]->mi_mediamask;
sc->tulip_probe_timeout = 0;
#ifdef notyet
if (sc->tulip_probe_state == TULIP_PROBE_FAILED)
break;
if (sc->tulip_probe_media != tulip_mii_phy_readspecific(sc))
break;
sc->tulip_probe_timeout = TULIP_IS_MEDIA_TP(sc->tulip_probe_media) ? 2500 : 300;
#endif
break;
}
case TULIP_PROBE_PHYAUTONEG: {
return;
}
case TULIP_PROBE_INACTIVE: {
/*
* Only probe if we autonegotiated a media that hasn't failed.
*/
sc->tulip_probe_timeout = 0;
if (sc->tulip_probe_mediamask & TULIP_BIT(sc->tulip_probe_media)) {
sc->tulip_probe_media = old_media;
break;
}
tulip_linkup(sc, sc->tulip_probe_media);
tulip_timeout(sc);
return;
}
default: {
#if defined(DIAGNOSTIC) || defined(TULIP_DEBUG)
panic("tulip_media_poll: botch at line %d\n", __LINE__);
#endif
break;
}
}
}
if (event == TULIP_MEDIAPOLL_TXPROBE_FAILED) {
#if defined(TULIP_DEBUG)
sc->tulip_dbg.dbg_txprobes_failed[sc->tulip_probe_media]++;
#endif
sc->tulip_flags &= ~TULIP_TXPROBE_ACTIVE;
return;
}
/*
* switch to another media if we tried this one enough.
*/
if (/* event == TULIP_MEDIAPOLL_TXPROBE_FAILED || */ sc->tulip_probe_timeout <= 0) {
#if defined(TULIP_DEBUG)
if (sc->tulip_probe_media == TULIP_MEDIA_UNKNOWN) {
if_printf(ifp, "poll media unknown!\n");
sc->tulip_probe_media = TULIP_MEDIA_MAX;
}
#endif
/*
* Find the next media type to check for. Full Duplex
* types are not allowed.
*/
do {
sc->tulip_probe_media -= 1;
if (sc->tulip_probe_media == TULIP_MEDIA_UNKNOWN) {
if (++sc->tulip_probe_passes == 3) {
if_printf(ifp, "autosense failed: cable problem?\n");
if ((sc->tulip_ifp->if_flags & IFF_UP) == 0) {
sc->tulip_ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
sc->tulip_probe_state = TULIP_PROBE_INACTIVE;
return;
}
}
sc->tulip_flags ^= TULIP_TRYNWAY; /* XXX */
sc->tulip_probe_mediamask = 0;
sc->tulip_probe_media = TULIP_MEDIA_MAX - 1;
}
} while (sc->tulip_mediums[sc->tulip_probe_media] == NULL
|| (sc->tulip_probe_mediamask & TULIP_BIT(sc->tulip_probe_media))
|| TULIP_IS_MEDIA_FD(sc->tulip_probe_media));
#if defined(TULIP_DEBUG)
if_printf(ifp, "%s: probing %s\n",
event == TULIP_MEDIAPOLL_TXPROBE_FAILED ? "txprobe failed" : "timeout",
tulip_mediums[sc->tulip_probe_media]);
#endif
sc->tulip_probe_timeout = TULIP_IS_MEDIA_TP(sc->tulip_probe_media) ? 2500 : 1000;
sc->tulip_probe_state = TULIP_PROBE_MEDIATEST;
sc->tulip_probe.probe_txprobes = 0;
tulip_reset(sc);
tulip_media_set(sc, sc->tulip_probe_media);
sc->tulip_flags &= ~TULIP_TXPROBE_ACTIVE;
}
tulip_timeout(sc);
/*
* If this is hanging off a phy, we know are doing NWAY and we have
* forced the phy to a specific speed. Wait for link up before
* before sending a packet.
*/
switch (sc->tulip_mediums[sc->tulip_probe_media]->mi_type) {
case TULIP_MEDIAINFO_MII: {
if (sc->tulip_probe_media != tulip_mii_phy_readspecific(sc))
return;
break;
}
case TULIP_MEDIAINFO_SIA: {
if (TULIP_IS_MEDIA_TP(sc->tulip_probe_media)) {
if (TULIP_CSR_READ(sc, csr_sia_status) & TULIP_SIASTS_LINKFAIL)
return;
tulip_linkup(sc, sc->tulip_probe_media);
#ifdef notyet
if (sc->tulip_features & TULIP_HAVE_MII)
tulip_timeout(sc);
#endif
return;
}
break;
}
case TULIP_MEDIAINFO_RESET:
case TULIP_MEDIAINFO_SYM:
case TULIP_MEDIAINFO_NONE:
case TULIP_MEDIAINFO_GPR: {
break;
}
}
/*
* Try to send a packet.
*/
tulip_txprobe(sc);
}
static void
tulip_media_select(tulip_softc_t * const sc)
{
TULIP_LOCK_ASSERT(sc);
if (sc->tulip_features & TULIP_HAVE_GPR) {
TULIP_CSR_WRITE(sc, csr_gp, TULIP_GP_PINSET|sc->tulip_gpinit);
DELAY(10);
TULIP_CSR_WRITE(sc, csr_gp, sc->tulip_gpdata);
}
/*
* If this board has no media, just return
*/
if (sc->tulip_features & TULIP_HAVE_NOMEDIA)
return;
if (sc->tulip_media == TULIP_MEDIA_UNKNOWN) {
TULIP_CSR_WRITE(sc, csr_intr, sc->tulip_intrmask);
(*sc->tulip_boardsw->bd_media_poll)(sc, TULIP_MEDIAPOLL_START);
} else {
tulip_media_set(sc, sc->tulip_media);
}
}
static void
tulip_21040_mediainfo_init(tulip_softc_t * const sc, tulip_media_t media)
{
TULIP_LOCK_ASSERT(sc);
sc->tulip_cmdmode |= TULIP_CMD_CAPTREFFCT|TULIP_CMD_THRSHLD160
|TULIP_CMD_BACKOFFCTR;
sc->tulip_ifp->if_baudrate = 10000000;
if (media == TULIP_MEDIA_10BASET || media == TULIP_MEDIA_UNKNOWN) {
TULIP_MEDIAINFO_SIA_INIT(sc, &sc->tulip_mediainfo[0], 21040, 10BASET);
TULIP_MEDIAINFO_SIA_INIT(sc, &sc->tulip_mediainfo[1], 21040, 10BASET_FD);
sc->tulip_intrmask |= TULIP_STS_LINKPASS|TULIP_STS_LINKFAIL;
}
if (media == TULIP_MEDIA_AUIBNC || media == TULIP_MEDIA_UNKNOWN) {
TULIP_MEDIAINFO_SIA_INIT(sc, &sc->tulip_mediainfo[2], 21040, AUIBNC);
}
if (media == TULIP_MEDIA_UNKNOWN) {
TULIP_MEDIAINFO_SIA_INIT(sc, &sc->tulip_mediainfo[3], 21040, EXTSIA);
}
}
static void
tulip_21040_media_probe(tulip_softc_t * const sc)
{
TULIP_LOCK_ASSERT(sc);
tulip_21040_mediainfo_init(sc, TULIP_MEDIA_UNKNOWN);
return;
}
static void
tulip_21040_10baset_only_media_probe(tulip_softc_t * const sc)
{
TULIP_LOCK_ASSERT(sc);
tulip_21040_mediainfo_init(sc, TULIP_MEDIA_10BASET);
tulip_media_set(sc, TULIP_MEDIA_10BASET);
sc->tulip_media = TULIP_MEDIA_10BASET;
}
static void
tulip_21040_10baset_only_media_select(tulip_softc_t * const sc)
{
TULIP_LOCK_ASSERT(sc);
sc->tulip_flags |= TULIP_LINKUP;
if (sc->tulip_media == TULIP_MEDIA_10BASET_FD) {
sc->tulip_cmdmode |= TULIP_CMD_FULLDUPLEX;
sc->tulip_flags &= ~TULIP_SQETEST;
} else {
sc->tulip_cmdmode &= ~TULIP_CMD_FULLDUPLEX;
sc->tulip_flags |= TULIP_SQETEST;
}
tulip_media_set(sc, sc->tulip_media);
}
static void
tulip_21040_auibnc_only_media_probe(tulip_softc_t * const sc)
{
TULIP_LOCK_ASSERT(sc);
tulip_21040_mediainfo_init(sc, TULIP_MEDIA_AUIBNC);
sc->tulip_flags |= TULIP_SQETEST|TULIP_LINKUP;
tulip_media_set(sc, TULIP_MEDIA_AUIBNC);
sc->tulip_media = TULIP_MEDIA_AUIBNC;
}
static void
tulip_21040_auibnc_only_media_select(tulip_softc_t * const sc)
{
TULIP_LOCK_ASSERT(sc);
tulip_media_set(sc, TULIP_MEDIA_AUIBNC);
sc->tulip_cmdmode &= ~TULIP_CMD_FULLDUPLEX;
}
static const tulip_boardsw_t tulip_21040_boardsw = {
TULIP_21040_GENERIC,
tulip_21040_media_probe,
tulip_media_select,
tulip_media_poll,
};
static const tulip_boardsw_t tulip_21040_10baset_only_boardsw = {
TULIP_21040_GENERIC,
tulip_21040_10baset_only_media_probe,
tulip_21040_10baset_only_media_select,
NULL,
};
static const tulip_boardsw_t tulip_21040_auibnc_only_boardsw = {
TULIP_21040_GENERIC,
tulip_21040_auibnc_only_media_probe,
tulip_21040_auibnc_only_media_select,
NULL,
};
static void
tulip_21041_mediainfo_init(tulip_softc_t * const sc)
{
tulip_media_info_t * const mi = sc->tulip_mediainfo;
TULIP_LOCK_ASSERT(sc);
#ifdef notyet
if (sc->tulip_revinfo >= 0x20) {
TULIP_MEDIAINFO_SIA_INIT(sc, &mi[0], 21041P2, 10BASET);
TULIP_MEDIAINFO_SIA_INIT(sc, &mi[1], 21041P2, 10BASET_FD);
TULIP_MEDIAINFO_SIA_INIT(sc, &mi[0], 21041P2, AUI);
TULIP_MEDIAINFO_SIA_INIT(sc, &mi[1], 21041P2, BNC);
return;
}
#endif
TULIP_MEDIAINFO_SIA_INIT(sc, &mi[0], 21041, 10BASET);
TULIP_MEDIAINFO_SIA_INIT(sc, &mi[1], 21041, 10BASET_FD);
TULIP_MEDIAINFO_SIA_INIT(sc, &mi[2], 21041, AUI);
TULIP_MEDIAINFO_SIA_INIT(sc, &mi[3], 21041, BNC);
}
static void
tulip_21041_media_probe(tulip_softc_t * const sc)
{
TULIP_LOCK_ASSERT(sc);
sc->tulip_ifp->if_baudrate = 10000000;
sc->tulip_cmdmode |= TULIP_CMD_CAPTREFFCT|TULIP_CMD_ENHCAPTEFFCT
|TULIP_CMD_THRSHLD160|TULIP_CMD_BACKOFFCTR;
sc->tulip_intrmask |= TULIP_STS_LINKPASS|TULIP_STS_LINKFAIL;
tulip_21041_mediainfo_init(sc);
}
static void
tulip_21041_media_poll(tulip_softc_t * const sc,
const tulip_mediapoll_event_t event)
{
u_int32_t sia_status;
TULIP_LOCK_ASSERT(sc);
#if defined(TULIP_DEBUG)
sc->tulip_dbg.dbg_events[event]++;
#endif
if (event == TULIP_MEDIAPOLL_LINKFAIL) {
if (sc->tulip_probe_state != TULIP_PROBE_INACTIVE
|| !TULIP_DO_AUTOSENSE(sc))
return;
sc->tulip_media = TULIP_MEDIA_UNKNOWN;
tulip_reset(sc); /* start probe */
return;
}
/*
* If we've been been asked to start a poll or link change interrupt
* restart the probe (and reset the tulip to a known state).
*/
if (event == TULIP_MEDIAPOLL_START) {
sc->tulip_ifp->if_drv_flags |= IFF_DRV_OACTIVE;
sc->tulip_cmdmode &= ~(TULIP_CMD_FULLDUPLEX|TULIP_CMD_RXRUN);
#ifdef notyet
if (sc->tulip_revinfo >= 0x20) {
sc->tulip_cmdmode |= TULIP_CMD_FULLDUPLEX;
sc->tulip_flags |= TULIP_DIDNWAY;
}
#endif
TULIP_CSR_WRITE(sc, csr_command, sc->tulip_cmdmode);
sc->tulip_probe_state = TULIP_PROBE_MEDIATEST;
sc->tulip_probe_media = TULIP_MEDIA_10BASET;
sc->tulip_probe_timeout = TULIP_21041_PROBE_10BASET_TIMEOUT;
tulip_media_set(sc, TULIP_MEDIA_10BASET);
tulip_timeout(sc);
return;
}
if (sc->tulip_probe_state == TULIP_PROBE_INACTIVE)
return;
if (event == TULIP_MEDIAPOLL_TXPROBE_OK) {
#if defined(TULIP_DEBUG)
sc->tulip_dbg.dbg_txprobes_ok[sc->tulip_probe_media]++;
#endif
tulip_linkup(sc, sc->tulip_probe_media);
return;
}
sia_status = TULIP_CSR_READ(sc, csr_sia_status);
TULIP_CSR_WRITE(sc, csr_sia_status, sia_status);
if ((sia_status & TULIP_SIASTS_LINKFAIL) == 0) {
if (sc->tulip_revinfo >= 0x20) {
if (sia_status & (PHYSTS_10BASET_FD << (16 - 6)))
sc->tulip_probe_media = TULIP_MEDIA_10BASET_FD;
}
/*
* If the link has passed LinkPass, 10baseT is the
* proper media to use.
*/
tulip_linkup(sc, sc->tulip_probe_media);
return;
}
/*
* wait for up to 2.4 seconds for the link to reach pass state.
* Only then start scanning the other media for activity.
* choose media with receive activity over those without.
*/
if (sc->tulip_probe_media == TULIP_MEDIA_10BASET) {
if (event != TULIP_MEDIAPOLL_TIMER)
return;
if (sc->tulip_probe_timeout > 0
&& (sia_status & TULIP_SIASTS_OTHERRXACTIVITY) == 0) {
tulip_timeout(sc);
return;
}
sc->tulip_probe_timeout = TULIP_21041_PROBE_AUIBNC_TIMEOUT;
sc->tulip_flags |= TULIP_WANTRXACT;
if (sia_status & TULIP_SIASTS_OTHERRXACTIVITY) {
sc->tulip_probe_media = TULIP_MEDIA_BNC;
} else {
sc->tulip_probe_media = TULIP_MEDIA_AUI;
}
tulip_media_set(sc, sc->tulip_probe_media);
tulip_timeout(sc);
return;
}
/*
* If we failed, clear the txprobe active flag.
*/
if (event == TULIP_MEDIAPOLL_TXPROBE_FAILED)
sc->tulip_flags &= ~TULIP_TXPROBE_ACTIVE;
if (event == TULIP_MEDIAPOLL_TIMER) {
/*
* If we've received something, then that's our link!
*/
if (sc->tulip_flags & TULIP_RXACT) {
tulip_linkup(sc, sc->tulip_probe_media);
return;
}
/*
* if no txprobe active
*/
if ((sc->tulip_flags & TULIP_TXPROBE_ACTIVE) == 0
&& ((sc->tulip_flags & TULIP_WANTRXACT) == 0
|| (sia_status & TULIP_SIASTS_RXACTIVITY))) {
sc->tulip_probe_timeout = TULIP_21041_PROBE_AUIBNC_TIMEOUT;
tulip_txprobe(sc);
tulip_timeout(sc);
return;
}
/*
* Take 2 passes through before deciding to not
* wait for receive activity. Then take another
* two passes before spitting out a warning.
*/
if (sc->tulip_probe_timeout <= 0) {
if (sc->tulip_flags & TULIP_WANTRXACT) {
sc->tulip_flags &= ~TULIP_WANTRXACT;
sc->tulip_probe_timeout = TULIP_21041_PROBE_AUIBNC_TIMEOUT;
} else {
if_printf(sc->tulip_ifp,
"autosense failed: cable problem?\n");
if ((sc->tulip_ifp->if_flags & IFF_UP) == 0) {
sc->tulip_ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
sc->tulip_probe_state = TULIP_PROBE_INACTIVE;
return;
}
}
}
}
/*
* Since this media failed to probe, try the other one.
*/
sc->tulip_probe_timeout = TULIP_21041_PROBE_AUIBNC_TIMEOUT;
if (sc->tulip_probe_media == TULIP_MEDIA_AUI) {
sc->tulip_probe_media = TULIP_MEDIA_BNC;
} else {
sc->tulip_probe_media = TULIP_MEDIA_AUI;
}
tulip_media_set(sc, sc->tulip_probe_media);
sc->tulip_flags &= ~TULIP_TXPROBE_ACTIVE;
tulip_timeout(sc);
}
static const tulip_boardsw_t tulip_21041_boardsw = {
TULIP_21041_GENERIC,
tulip_21041_media_probe,
tulip_media_select,
tulip_21041_media_poll
};
static const tulip_phy_attr_t tulip_mii_phy_attrlist[] = {
{ 0x20005c00, 0, /* 08-00-17 */
{
{ 0x19, 0x0040, 0x0040 }, /* 10TX */
{ 0x19, 0x0040, 0x0000 }, /* 100TX */
},
#if defined(TULIP_DEBUG)
"NS DP83840",
#endif
},
{ 0x0281F400, 0, /* 00-A0-7D */
{
{ 0x12, 0x0010, 0x0000 }, /* 10T */
{ }, /* 100TX */
{ 0x12, 0x0010, 0x0010 }, /* 100T4 */
{ 0x12, 0x0008, 0x0008 }, /* FULL_DUPLEX */
},
#if defined(TULIP_DEBUG)
"Seeq 80C240"
#endif
},
#if 0
{ 0x0015F420, 0, /* 00-A0-7D */
{
{ 0x12, 0x0010, 0x0000 }, /* 10T */
{ }, /* 100TX */
{ 0x12, 0x0010, 0x0010 }, /* 100T4 */
{ 0x12, 0x0008, 0x0008 }, /* FULL_DUPLEX */
},
#if defined(TULIP_DEBUG)
"Broadcom BCM5000"
#endif
},
#endif
{ 0x0281F400, 0, /* 00-A0-BE */
{
{ 0x11, 0x8000, 0x0000 }, /* 10T */
{ 0x11, 0x8000, 0x8000 }, /* 100TX */
{ }, /* 100T4 */
{ 0x11, 0x4000, 0x4000 }, /* FULL_DUPLEX */
},
#if defined(TULIP_DEBUG)
"ICS 1890"
#endif
},
{ 0 }
};
static tulip_media_t
tulip_mii_phy_readspecific(tulip_softc_t * const sc)
{
const tulip_phy_attr_t *attr;
u_int16_t data;
u_int32_t id;
unsigned idx = 0;
static const tulip_media_t table[] = {
TULIP_MEDIA_UNKNOWN,
TULIP_MEDIA_10BASET,
TULIP_MEDIA_100BASETX,
TULIP_MEDIA_100BASET4,
TULIP_MEDIA_UNKNOWN,
TULIP_MEDIA_10BASET_FD,
TULIP_MEDIA_100BASETX_FD,
TULIP_MEDIA_UNKNOWN
};
TULIP_LOCK_ASSERT(sc);
/*
* Don't read phy specific registers if link is not up.
*/
data = tulip_mii_readreg(sc, sc->tulip_phyaddr, PHYREG_STATUS);
if ((data & (PHYSTS_LINK_UP|PHYSTS_EXTENDED_REGS)) != (PHYSTS_LINK_UP|PHYSTS_EXTENDED_REGS))
return TULIP_MEDIA_UNKNOWN;
id = (tulip_mii_readreg(sc, sc->tulip_phyaddr, PHYREG_IDLOW) << 16) |
tulip_mii_readreg(sc, sc->tulip_phyaddr, PHYREG_IDHIGH);
for (attr = tulip_mii_phy_attrlist;; attr++) {
if (attr->attr_id == 0)
return TULIP_MEDIA_UNKNOWN;
if ((id & ~0x0F) == attr->attr_id)
break;
}
if (attr->attr_modes[PHY_MODE_100TX].pm_regno) {
const tulip_phy_modedata_t * const pm = &attr->attr_modes[PHY_MODE_100TX];
data = tulip_mii_readreg(sc, sc->tulip_phyaddr, pm->pm_regno);
if ((data & pm->pm_mask) == pm->pm_value)
idx = 2;
}
if (idx == 0 && attr->attr_modes[PHY_MODE_100T4].pm_regno) {
const tulip_phy_modedata_t * const pm = &attr->attr_modes[PHY_MODE_100T4];
data = tulip_mii_readreg(sc, sc->tulip_phyaddr, pm->pm_regno);
if ((data & pm->pm_mask) == pm->pm_value)
idx = 3;
}
if (idx == 0 && attr->attr_modes[PHY_MODE_10T].pm_regno) {
const tulip_phy_modedata_t * const pm = &attr->attr_modes[PHY_MODE_10T];
data = tulip_mii_readreg(sc, sc->tulip_phyaddr, pm->pm_regno);
if ((data & pm->pm_mask) == pm->pm_value)
idx = 1;
}
if (idx != 0 && attr->attr_modes[PHY_MODE_FULLDUPLEX].pm_regno) {
const tulip_phy_modedata_t * const pm = &attr->attr_modes[PHY_MODE_FULLDUPLEX];
data = tulip_mii_readreg(sc, sc->tulip_phyaddr, pm->pm_regno);
idx += ((data & pm->pm_mask) == pm->pm_value ? 4 : 0);
}
return table[idx];
}
static unsigned
tulip_mii_get_phyaddr(tulip_softc_t * const sc, unsigned offset)
{
unsigned phyaddr;
TULIP_LOCK_ASSERT(sc);
for (phyaddr = 1; phyaddr < 32; phyaddr++) {
unsigned status = tulip_mii_readreg(sc, phyaddr, PHYREG_STATUS);
if (status == 0 || status == 0xFFFF || status < PHYSTS_10BASET)
continue;
if (offset == 0)
return phyaddr;
offset--;
}
if (offset == 0) {
unsigned status = tulip_mii_readreg(sc, 0, PHYREG_STATUS);
if (status == 0 || status == 0xFFFF || status < PHYSTS_10BASET)
return TULIP_MII_NOPHY;
return 0;
}
return TULIP_MII_NOPHY;
}
static int
tulip_mii_map_abilities(tulip_softc_t * const sc, unsigned abilities)
{
TULIP_LOCK_ASSERT(sc);
sc->tulip_abilities = abilities;
if (abilities & PHYSTS_100BASETX_FD) {
sc->tulip_probe_media = TULIP_MEDIA_100BASETX_FD;
} else if (abilities & PHYSTS_100BASET4) {
sc->tulip_probe_media = TULIP_MEDIA_100BASET4;
} else if (abilities & PHYSTS_100BASETX) {
sc->tulip_probe_media = TULIP_MEDIA_100BASETX;
} else if (abilities & PHYSTS_10BASET_FD) {
sc->tulip_probe_media = TULIP_MEDIA_10BASET_FD;
} else if (abilities & PHYSTS_10BASET) {
sc->tulip_probe_media = TULIP_MEDIA_10BASET;
} else {
sc->tulip_probe_state = TULIP_PROBE_MEDIATEST;
return 0;
}
sc->tulip_probe_state = TULIP_PROBE_INACTIVE;
return 1;
}
static void
tulip_mii_autonegotiate(tulip_softc_t * const sc, const unsigned phyaddr)
{
struct ifnet *ifp = sc->tulip_ifp;
TULIP_LOCK_ASSERT(sc);
switch (sc->tulip_probe_state) {
case TULIP_PROBE_MEDIATEST:
case TULIP_PROBE_INACTIVE: {
sc->tulip_flags |= TULIP_DIDNWAY;
tulip_mii_writereg(sc, phyaddr, PHYREG_CONTROL, PHYCTL_RESET);
sc->tulip_probe_timeout = 3000;
sc->tulip_intrmask |= TULIP_STS_ABNRMLINTR|TULIP_STS_NORMALINTR;
sc->tulip_probe_state = TULIP_PROBE_PHYRESET;
}
/* FALLTHROUGH */
case TULIP_PROBE_PHYRESET: {
u_int32_t status;
u_int32_t data = tulip_mii_readreg(sc, phyaddr, PHYREG_CONTROL);
if (data & PHYCTL_RESET) {
if (sc->tulip_probe_timeout > 0) {
tulip_timeout(sc);
return;
}
printf("%s(phy%d): error: reset of PHY never completed!\n",
ifp->if_xname, phyaddr);
sc->tulip_flags &= ~TULIP_TXPROBE_ACTIVE;
sc->tulip_probe_state = TULIP_PROBE_FAILED;
sc->tulip_ifp->if_flags &= ~IFF_UP;
sc->tulip_ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
return;
}
status = tulip_mii_readreg(sc, phyaddr, PHYREG_STATUS);
if ((status & PHYSTS_CAN_AUTONEG) == 0) {
#if defined(TULIP_DEBUG)
loudprintf("%s(phy%d): autonegotiation disabled\n",
ifp->if_xname, phyaddr);
#endif
sc->tulip_flags &= ~TULIP_DIDNWAY;
sc->tulip_probe_state = TULIP_PROBE_MEDIATEST;
return;
}
if (tulip_mii_readreg(sc, phyaddr, PHYREG_AUTONEG_ADVERTISEMENT) != ((status >> 6) | 0x01))
tulip_mii_writereg(sc, phyaddr, PHYREG_AUTONEG_ADVERTISEMENT, (status >> 6) | 0x01);
tulip_mii_writereg(sc, phyaddr, PHYREG_CONTROL, data|PHYCTL_AUTONEG_RESTART|PHYCTL_AUTONEG_ENABLE);
data = tulip_mii_readreg(sc, phyaddr, PHYREG_CONTROL);
#if defined(TULIP_DEBUG)
if ((data & PHYCTL_AUTONEG_ENABLE) == 0)
loudprintf("%s(phy%d): oops: enable autonegotiation failed: 0x%04x\n",
ifp->if_xname, phyaddr, data);
else
loudprintf("%s(phy%d): autonegotiation restarted: 0x%04x\n",
ifp->if_xname, phyaddr, data);
sc->tulip_dbg.dbg_nway_starts++;
#endif
sc->tulip_probe_state = TULIP_PROBE_PHYAUTONEG;
sc->tulip_probe_timeout = 3000;
}
/* FALLTHROUGH */
case TULIP_PROBE_PHYAUTONEG: {
u_int32_t status = tulip_mii_readreg(sc, phyaddr, PHYREG_STATUS);
u_int32_t data;
if ((status & PHYSTS_AUTONEG_DONE) == 0) {
if (sc->tulip_probe_timeout > 0) {
tulip_timeout(sc);
return;
}
#if defined(TULIP_DEBUG)
loudprintf("%s(phy%d): autonegotiation timeout: sts=0x%04x, ctl=0x%04x\n",
ifp->if_xname, phyaddr, status,
tulip_mii_readreg(sc, phyaddr, PHYREG_CONTROL));
#endif
sc->tulip_flags &= ~TULIP_DIDNWAY;
sc->tulip_probe_state = TULIP_PROBE_MEDIATEST;
return;
}
data = tulip_mii_readreg(sc, phyaddr, PHYREG_AUTONEG_ABILITIES);
#if defined(TULIP_DEBUG)
loudprintf("%s(phy%d): autonegotiation complete: 0x%04x\n",
ifp->if_xname, phyaddr, data);
#endif
data = (data << 6) & status;
if (!tulip_mii_map_abilities(sc, data))
sc->tulip_flags &= ~TULIP_DIDNWAY;
return;
}
default: {
#if defined(DIAGNOSTIC)
panic("tulip_media_poll: botch at line %d\n", __LINE__);
#endif
break;
}
}
#if defined(TULIP_DEBUG)
loudprintf("%s(phy%d): autonegotiation failure: state = %d\n",
ifp->if_xname, phyaddr, sc->tulip_probe_state);
sc->tulip_dbg.dbg_nway_failures++;
#endif
}
static void
tulip_2114x_media_preset(tulip_softc_t * const sc)
{
const tulip_media_info_t *mi = NULL;
tulip_media_t media = sc->tulip_media;
TULIP_LOCK_ASSERT(sc);
if (sc->tulip_probe_state == TULIP_PROBE_INACTIVE)
media = sc->tulip_media;
else
media = sc->tulip_probe_media;
sc->tulip_cmdmode &= ~TULIP_CMD_PORTSELECT;
sc->tulip_flags &= ~TULIP_SQETEST;
if (media != TULIP_MEDIA_UNKNOWN && media != TULIP_MEDIA_MAX) {
#if defined(TULIP_DEBUG)
if (media < TULIP_MEDIA_MAX && sc->tulip_mediums[media] != NULL) {
#endif
mi = sc->tulip_mediums[media];
if (mi->mi_type == TULIP_MEDIAINFO_MII) {
sc->tulip_cmdmode |= TULIP_CMD_PORTSELECT;
} else if (mi->mi_type == TULIP_MEDIAINFO_GPR
|| mi->mi_type == TULIP_MEDIAINFO_SYM) {
sc->tulip_cmdmode &= ~TULIP_GPR_CMDBITS;
sc->tulip_cmdmode |= mi->mi_cmdmode;
} else if (mi->mi_type == TULIP_MEDIAINFO_SIA) {
TULIP_CSR_WRITE(sc, csr_sia_connectivity, TULIP_SIACONN_RESET);
}
#if defined(TULIP_DEBUG)
} else {
if_printf(sc->tulip_ifp, "preset: bad media %d!\n", media);
}
#endif
}
switch (media) {
case TULIP_MEDIA_BNC:
case TULIP_MEDIA_AUI:
case TULIP_MEDIA_10BASET: {
sc->tulip_cmdmode &= ~TULIP_CMD_FULLDUPLEX;
sc->tulip_cmdmode |= TULIP_CMD_TXTHRSHLDCTL;
sc->tulip_ifp->if_baudrate = 10000000;
sc->tulip_flags |= TULIP_SQETEST;
break;
}
case TULIP_MEDIA_10BASET_FD: {
sc->tulip_cmdmode |= TULIP_CMD_FULLDUPLEX|TULIP_CMD_TXTHRSHLDCTL;
sc->tulip_ifp->if_baudrate = 10000000;
break;
}
case TULIP_MEDIA_100BASEFX:
case TULIP_MEDIA_100BASET4:
case TULIP_MEDIA_100BASETX: {
sc->tulip_cmdmode &= ~(TULIP_CMD_FULLDUPLEX|TULIP_CMD_TXTHRSHLDCTL);
sc->tulip_cmdmode |= TULIP_CMD_PORTSELECT;
sc->tulip_ifp->if_baudrate = 100000000;
break;
}
case TULIP_MEDIA_100BASEFX_FD:
case TULIP_MEDIA_100BASETX_FD: {
sc->tulip_cmdmode |= TULIP_CMD_FULLDUPLEX|TULIP_CMD_PORTSELECT;
sc->tulip_cmdmode &= ~TULIP_CMD_TXTHRSHLDCTL;
sc->tulip_ifp->if_baudrate = 100000000;
break;
}
default: {
break;
}
}
TULIP_CSR_WRITE(sc, csr_command, sc->tulip_cmdmode);
}
/*
********************************************************************
* Start of 21140/21140A support which does not use the MII interface
*/
static void
tulip_null_media_poll(tulip_softc_t * const sc, tulip_mediapoll_event_t event)
{
#if defined(TULIP_DEBUG)
sc->tulip_dbg.dbg_events[event]++;
#endif
#if defined(DIAGNOSTIC)
if_printf(sc->tulip_ifp, "botch(media_poll) at line %d\n", __LINE__);
#endif
}
__inline static void
tulip_21140_mediainit(tulip_softc_t * const sc, tulip_media_info_t * const mip,
tulip_media_t const media, unsigned gpdata, unsigned cmdmode)
{
TULIP_LOCK_ASSERT(sc);
sc->tulip_mediums[media] = mip;
mip->mi_type = TULIP_MEDIAINFO_GPR;
mip->mi_cmdmode = cmdmode;
mip->mi_gpdata = gpdata;
}
static void
tulip_21140_evalboard_media_probe(tulip_softc_t * const sc)
{
tulip_media_info_t *mip = sc->tulip_mediainfo;
TULIP_LOCK_ASSERT(sc);
sc->tulip_gpinit = TULIP_GP_EB_PINS;
sc->tulip_gpdata = TULIP_GP_EB_INIT;
TULIP_CSR_WRITE(sc, csr_gp, TULIP_GP_EB_PINS);
TULIP_CSR_WRITE(sc, csr_gp, TULIP_GP_EB_INIT);
TULIP_CSR_WRITE(sc, csr_command,
TULIP_CSR_READ(sc, csr_command) | TULIP_CMD_PORTSELECT |
TULIP_CMD_PCSFUNCTION | TULIP_CMD_SCRAMBLER | TULIP_CMD_MUSTBEONE);
TULIP_CSR_WRITE(sc, csr_command,
TULIP_CSR_READ(sc, csr_command) & ~TULIP_CMD_TXTHRSHLDCTL);
DELAY(1000000);
if ((TULIP_CSR_READ(sc, csr_gp) & TULIP_GP_EB_OK100) != 0) {
sc->tulip_media = TULIP_MEDIA_10BASET;
} else {
sc->tulip_media = TULIP_MEDIA_100BASETX;
}
tulip_21140_mediainit(sc, mip++, TULIP_MEDIA_10BASET,
TULIP_GP_EB_INIT,
TULIP_CMD_TXTHRSHLDCTL);
tulip_21140_mediainit(sc, mip++, TULIP_MEDIA_10BASET_FD,
TULIP_GP_EB_INIT,
TULIP_CMD_TXTHRSHLDCTL|TULIP_CMD_FULLDUPLEX);
tulip_21140_mediainit(sc, mip++, TULIP_MEDIA_100BASETX,
TULIP_GP_EB_INIT,
TULIP_CMD_PORTSELECT|TULIP_CMD_PCSFUNCTION
|TULIP_CMD_SCRAMBLER);
tulip_21140_mediainit(sc, mip++, TULIP_MEDIA_100BASETX_FD,
TULIP_GP_EB_INIT,
TULIP_CMD_PORTSELECT|TULIP_CMD_PCSFUNCTION
|TULIP_CMD_SCRAMBLER|TULIP_CMD_FULLDUPLEX);
}
static const tulip_boardsw_t tulip_21140_eb_boardsw = {
TULIP_21140_DEC_EB,
tulip_21140_evalboard_media_probe,
tulip_media_select,
tulip_null_media_poll,
tulip_2114x_media_preset,
};
static void
tulip_21140_accton_media_probe(tulip_softc_t * const sc)
{
tulip_media_info_t *mip = sc->tulip_mediainfo;
unsigned gpdata;
TULIP_LOCK_ASSERT(sc);
sc->tulip_gpinit = TULIP_GP_EB_PINS;
sc->tulip_gpdata = TULIP_GP_EB_INIT;
TULIP_CSR_WRITE(sc, csr_gp, TULIP_GP_EB_PINS);
TULIP_CSR_WRITE(sc, csr_gp, TULIP_GP_EB_INIT);
TULIP_CSR_WRITE(sc, csr_command,
TULIP_CSR_READ(sc, csr_command) | TULIP_CMD_PORTSELECT |
TULIP_CMD_PCSFUNCTION | TULIP_CMD_SCRAMBLER | TULIP_CMD_MUSTBEONE);
TULIP_CSR_WRITE(sc, csr_command,
TULIP_CSR_READ(sc, csr_command) & ~TULIP_CMD_TXTHRSHLDCTL);
DELAY(1000000);
gpdata = TULIP_CSR_READ(sc, csr_gp);
if ((gpdata & TULIP_GP_EN1207_UTP_INIT) == 0) {
sc->tulip_media = TULIP_MEDIA_10BASET;
} else {
if ((gpdata & TULIP_GP_EN1207_BNC_INIT) == 0) {
sc->tulip_media = TULIP_MEDIA_BNC;
} else {
sc->tulip_media = TULIP_MEDIA_100BASETX;
}
}
tulip_21140_mediainit(sc, mip++, TULIP_MEDIA_BNC,
TULIP_GP_EN1207_BNC_INIT,
TULIP_CMD_TXTHRSHLDCTL);
tulip_21140_mediainit(sc, mip++, TULIP_MEDIA_10BASET,
TULIP_GP_EN1207_UTP_INIT,
TULIP_CMD_TXTHRSHLDCTL);
tulip_21140_mediainit(sc, mip++, TULIP_MEDIA_10BASET_FD,
TULIP_GP_EN1207_UTP_INIT,
TULIP_CMD_TXTHRSHLDCTL|TULIP_CMD_FULLDUPLEX);
tulip_21140_mediainit(sc, mip++, TULIP_MEDIA_100BASETX,
TULIP_GP_EN1207_100_INIT,
TULIP_CMD_PORTSELECT|TULIP_CMD_PCSFUNCTION
|TULIP_CMD_SCRAMBLER);
tulip_21140_mediainit(sc, mip++, TULIP_MEDIA_100BASETX_FD,
TULIP_GP_EN1207_100_INIT,
TULIP_CMD_PORTSELECT|TULIP_CMD_PCSFUNCTION
|TULIP_CMD_SCRAMBLER|TULIP_CMD_FULLDUPLEX);
}
static const tulip_boardsw_t tulip_21140_accton_boardsw = {
TULIP_21140_EN1207,
tulip_21140_accton_media_probe,
tulip_media_select,
tulip_null_media_poll,
tulip_2114x_media_preset,
};
static void
tulip_21140_smc9332_media_probe(tulip_softc_t * const sc)
{
tulip_media_info_t *mip = sc->tulip_mediainfo;
int idx, cnt = 0;
TULIP_LOCK_ASSERT(sc);
TULIP_CSR_WRITE(sc, csr_command, TULIP_CMD_PORTSELECT|TULIP_CMD_MUSTBEONE);
TULIP_CSR_WRITE(sc, csr_busmode, TULIP_BUSMODE_SWRESET);
DELAY(10); /* Wait 10 microseconds (actually 50 PCI cycles but at
33MHz that comes to two microseconds but wait a
bit longer anyways) */
TULIP_CSR_WRITE(sc, csr_command, TULIP_CMD_PORTSELECT |
TULIP_CMD_PCSFUNCTION | TULIP_CMD_SCRAMBLER | TULIP_CMD_MUSTBEONE);
sc->tulip_gpinit = TULIP_GP_SMC_9332_PINS;
sc->tulip_gpdata = TULIP_GP_SMC_9332_INIT;
TULIP_CSR_WRITE(sc, csr_gp, TULIP_GP_SMC_9332_PINS|TULIP_GP_PINSET);
TULIP_CSR_WRITE(sc, csr_gp, TULIP_GP_SMC_9332_INIT);
DELAY(200000);
for (idx = 1000; idx > 0; idx--) {
u_int32_t csr = TULIP_CSR_READ(sc, csr_gp);
if ((csr & (TULIP_GP_SMC_9332_OK10|TULIP_GP_SMC_9332_OK100)) == (TULIP_GP_SMC_9332_OK10|TULIP_GP_SMC_9332_OK100)) {
if (++cnt > 100)
break;
} else if ((csr & TULIP_GP_SMC_9332_OK10) == 0) {
break;
} else {
cnt = 0;
}
DELAY(1000);
}
sc->tulip_media = cnt > 100 ? TULIP_MEDIA_100BASETX : TULIP_MEDIA_10BASET;
tulip_21140_mediainit(sc, mip++, TULIP_MEDIA_100BASETX,
TULIP_GP_SMC_9332_INIT,
TULIP_CMD_PORTSELECT|TULIP_CMD_PCSFUNCTION
|TULIP_CMD_SCRAMBLER);
tulip_21140_mediainit(sc, mip++, TULIP_MEDIA_100BASETX_FD,
TULIP_GP_SMC_9332_INIT,
TULIP_CMD_PORTSELECT|TULIP_CMD_PCSFUNCTION
|TULIP_CMD_SCRAMBLER|TULIP_CMD_FULLDUPLEX);
tulip_21140_mediainit(sc, mip++, TULIP_MEDIA_10BASET,
TULIP_GP_SMC_9332_INIT,
TULIP_CMD_TXTHRSHLDCTL);
tulip_21140_mediainit(sc, mip++, TULIP_MEDIA_10BASET_FD,
TULIP_GP_SMC_9332_INIT,
TULIP_CMD_TXTHRSHLDCTL|TULIP_CMD_FULLDUPLEX);
}
static const tulip_boardsw_t tulip_21140_smc9332_boardsw = {
TULIP_21140_SMC_9332,
tulip_21140_smc9332_media_probe,
tulip_media_select,
tulip_null_media_poll,
tulip_2114x_media_preset,
};
static void
tulip_21140_cogent_em100_media_probe(tulip_softc_t * const sc)
{
tulip_media_info_t *mip = sc->tulip_mediainfo;
1997-08-03 13:00:42 +00:00
u_int32_t cmdmode = TULIP_CSR_READ(sc, csr_command);
TULIP_LOCK_ASSERT(sc);
sc->tulip_gpinit = TULIP_GP_EM100_PINS;
sc->tulip_gpdata = TULIP_GP_EM100_INIT;
TULIP_CSR_WRITE(sc, csr_gp, TULIP_GP_EM100_PINS);
TULIP_CSR_WRITE(sc, csr_gp, TULIP_GP_EM100_INIT);
1997-08-03 13:00:42 +00:00
cmdmode = TULIP_CMD_PORTSELECT|TULIP_CMD_PCSFUNCTION|TULIP_CMD_MUSTBEONE;
cmdmode &= ~(TULIP_CMD_TXTHRSHLDCTL|TULIP_CMD_SCRAMBLER);
if (sc->tulip_rombuf[32] == TULIP_COGENT_EM100FX_ID) {
TULIP_CSR_WRITE(sc, csr_command, cmdmode);
sc->tulip_media = TULIP_MEDIA_100BASEFX;
tulip_21140_mediainit(sc, mip++, TULIP_MEDIA_100BASEFX,
TULIP_GP_EM100_INIT,
TULIP_CMD_PORTSELECT|TULIP_CMD_PCSFUNCTION);
tulip_21140_mediainit(sc, mip++, TULIP_MEDIA_100BASEFX_FD,
TULIP_GP_EM100_INIT,
TULIP_CMD_PORTSELECT|TULIP_CMD_PCSFUNCTION
|TULIP_CMD_FULLDUPLEX);
} else {
TULIP_CSR_WRITE(sc, csr_command, cmdmode|TULIP_CMD_SCRAMBLER);
sc->tulip_media = TULIP_MEDIA_100BASETX;
tulip_21140_mediainit(sc, mip++, TULIP_MEDIA_100BASETX,
TULIP_GP_EM100_INIT,
TULIP_CMD_PORTSELECT|TULIP_CMD_PCSFUNCTION
|TULIP_CMD_SCRAMBLER);
1997-08-03 13:00:42 +00:00
tulip_21140_mediainit(sc, mip++, TULIP_MEDIA_100BASETX_FD,
TULIP_GP_EM100_INIT,
TULIP_CMD_PORTSELECT|TULIP_CMD_PCSFUNCTION
|TULIP_CMD_SCRAMBLER|TULIP_CMD_FULLDUPLEX);
1997-08-03 13:00:42 +00:00
}
}
static const tulip_boardsw_t tulip_21140_cogent_em100_boardsw = {
TULIP_21140_COGENT_EM100,
tulip_21140_cogent_em100_media_probe,
tulip_media_select,
tulip_null_media_poll,
tulip_2114x_media_preset
};
static void
tulip_21140_znyx_zx34x_media_probe(tulip_softc_t * const sc)
{
tulip_media_info_t *mip = sc->tulip_mediainfo;
int cnt10 = 0, cnt100 = 0, idx;
TULIP_LOCK_ASSERT(sc);
sc->tulip_gpinit = TULIP_GP_ZX34X_PINS;
sc->tulip_gpdata = TULIP_GP_ZX34X_INIT;
TULIP_CSR_WRITE(sc, csr_gp, TULIP_GP_ZX34X_PINS);
TULIP_CSR_WRITE(sc, csr_gp, TULIP_GP_ZX34X_INIT);
TULIP_CSR_WRITE(sc, csr_command,
TULIP_CSR_READ(sc, csr_command) | TULIP_CMD_PORTSELECT |
TULIP_CMD_PCSFUNCTION | TULIP_CMD_SCRAMBLER | TULIP_CMD_MUSTBEONE);
TULIP_CSR_WRITE(sc, csr_command,
TULIP_CSR_READ(sc, csr_command) & ~TULIP_CMD_TXTHRSHLDCTL);
DELAY(200000);
for (idx = 1000; idx > 0; idx--) {
u_int32_t csr = TULIP_CSR_READ(sc, csr_gp);
if ((csr & (TULIP_GP_ZX34X_LNKFAIL|TULIP_GP_ZX34X_SYMDET|TULIP_GP_ZX34X_SIGDET)) == (TULIP_GP_ZX34X_LNKFAIL|TULIP_GP_ZX34X_SYMDET|TULIP_GP_ZX34X_SIGDET)) {
if (++cnt100 > 100)
break;
} else if ((csr & TULIP_GP_ZX34X_LNKFAIL) == 0) {
if (++cnt10 > 100)
break;
} else {
cnt10 = 0;
cnt100 = 0;
}
DELAY(1000);
}
sc->tulip_media = cnt100 > 100 ? TULIP_MEDIA_100BASETX : TULIP_MEDIA_10BASET;
tulip_21140_mediainit(sc, mip++, TULIP_MEDIA_10BASET,
TULIP_GP_ZX34X_INIT,
TULIP_CMD_TXTHRSHLDCTL);
tulip_21140_mediainit(sc, mip++, TULIP_MEDIA_10BASET_FD,
TULIP_GP_ZX34X_INIT,
TULIP_CMD_TXTHRSHLDCTL|TULIP_CMD_FULLDUPLEX);
tulip_21140_mediainit(sc, mip++, TULIP_MEDIA_100BASETX,
TULIP_GP_ZX34X_INIT,
TULIP_CMD_PORTSELECT|TULIP_CMD_PCSFUNCTION
|TULIP_CMD_SCRAMBLER);
tulip_21140_mediainit(sc, mip++, TULIP_MEDIA_100BASETX_FD,
TULIP_GP_ZX34X_INIT,
TULIP_CMD_PORTSELECT|TULIP_CMD_PCSFUNCTION
|TULIP_CMD_SCRAMBLER|TULIP_CMD_FULLDUPLEX);
}
static const tulip_boardsw_t tulip_21140_znyx_zx34x_boardsw = {
TULIP_21140_ZNYX_ZX34X,
tulip_21140_znyx_zx34x_media_probe,
tulip_media_select,
tulip_null_media_poll,
tulip_2114x_media_preset,
};
static void
tulip_2114x_media_probe(tulip_softc_t * const sc)
{
TULIP_LOCK_ASSERT(sc);
1997-08-03 13:00:42 +00:00
sc->tulip_cmdmode |= TULIP_CMD_MUSTBEONE
|TULIP_CMD_BACKOFFCTR|TULIP_CMD_THRSHLD72;
}
static const tulip_boardsw_t tulip_2114x_isv_boardsw = {
TULIP_21140_ISV,
tulip_2114x_media_probe,
tulip_media_select,
tulip_media_poll,
tulip_2114x_media_preset,
};
/*
* ******** END of chip-specific handlers. ***********
*/
/*
* Code the read the SROM and MII bit streams (I2C)
*/
#define EMIT do { TULIP_CSR_WRITE(sc, csr_srom_mii, csr); DELAY(1); } while (0)
static void
tulip_srom_idle(tulip_softc_t * const sc)
{
unsigned bit, csr;
csr = SROMSEL ; EMIT;
csr = SROMSEL | SROMRD; EMIT;
csr ^= SROMCS; EMIT;
csr ^= SROMCLKON; EMIT;
/*
* Write 25 cycles of 0 which will force the SROM to be idle.
*/
for (bit = 3 + SROM_BITWIDTH + 16; bit > 0; bit--) {
csr ^= SROMCLKOFF; EMIT; /* clock low; data not valid */
csr ^= SROMCLKON; EMIT; /* clock high; data valid */
}
csr ^= SROMCLKOFF; EMIT;
csr ^= SROMCS; EMIT;
csr = 0; EMIT;
}
static void
tulip_srom_read(tulip_softc_t * const sc)
{
1997-08-03 13:00:42 +00:00
unsigned idx;
const unsigned bitwidth = SROM_BITWIDTH;
const unsigned cmdmask = (SROMCMD_RD << bitwidth);
const unsigned msb = 1 << (bitwidth + 3 - 1);
unsigned lastidx = (1 << bitwidth) - 1;
tulip_srom_idle(sc);
for (idx = 0; idx <= lastidx; idx++) {
unsigned lastbit, data, bits, bit, csr;
csr = SROMSEL ; EMIT;
csr = SROMSEL | SROMRD; EMIT;
csr ^= SROMCSON; EMIT;
csr ^= SROMCLKON; EMIT;
lastbit = 0;
for (bits = idx|cmdmask, bit = bitwidth + 3; bit > 0; bit--, bits <<= 1) {
const unsigned thisbit = bits & msb;
csr ^= SROMCLKOFF; EMIT; /* clock low; data not valid */
if (thisbit != lastbit) {
csr ^= SROMDOUT; EMIT; /* clock low; invert data */
} else {
EMIT;
}
csr ^= SROMCLKON; EMIT; /* clock high; data valid */
lastbit = thisbit;
}
csr ^= SROMCLKOFF; EMIT;
for (data = 0, bits = 0; bits < 16; bits++) {
data <<= 1;
csr ^= SROMCLKON; EMIT; /* clock high; data valid */
data |= TULIP_CSR_READ(sc, csr_srom_mii) & SROMDIN ? 1 : 0;
csr ^= SROMCLKOFF; EMIT; /* clock low; data not valid */
}
sc->tulip_rombuf[idx*2] = data & 0xFF;
sc->tulip_rombuf[idx*2+1] = data >> 8;
csr = SROMSEL | SROMRD; EMIT;
csr = 0; EMIT;
}
tulip_srom_idle(sc);
}
#define MII_EMIT do { TULIP_CSR_WRITE(sc, csr_srom_mii, csr); DELAY(1); } while (0)
static void
tulip_mii_writebits(tulip_softc_t * const sc, unsigned data, unsigned bits)
{
unsigned msb = 1 << (bits - 1);
unsigned csr = TULIP_CSR_READ(sc, csr_srom_mii) & (MII_RD|MII_DOUT|MII_CLK);
unsigned lastbit = (csr & MII_DOUT) ? msb : 0;
TULIP_LOCK_ASSERT(sc);
csr |= MII_WR; MII_EMIT; /* clock low; assert write */
for (; bits > 0; bits--, data <<= 1) {
const unsigned thisbit = data & msb;
if (thisbit != lastbit) {
csr ^= MII_DOUT; MII_EMIT; /* clock low; invert data */
}
csr ^= MII_CLKON; MII_EMIT; /* clock high; data valid */
lastbit = thisbit;
csr ^= MII_CLKOFF; MII_EMIT; /* clock low; data not valid */
}
}
static void
tulip_mii_turnaround(tulip_softc_t * const sc, unsigned cmd)
{
unsigned csr = TULIP_CSR_READ(sc, csr_srom_mii) & (MII_RD|MII_DOUT|MII_CLK);
TULIP_LOCK_ASSERT(sc);
if (cmd == MII_WRCMD) {
csr |= MII_DOUT; MII_EMIT; /* clock low; change data */
csr ^= MII_CLKON; MII_EMIT; /* clock high; data valid */
csr ^= MII_CLKOFF; MII_EMIT; /* clock low; data not valid */
csr ^= MII_DOUT; MII_EMIT; /* clock low; change data */
} else {
csr |= MII_RD; MII_EMIT; /* clock low; switch to read */
}
csr ^= MII_CLKON; MII_EMIT; /* clock high; data valid */
csr ^= MII_CLKOFF; MII_EMIT; /* clock low; data not valid */
}
static unsigned
tulip_mii_readbits(tulip_softc_t * const sc)
{
unsigned data;
unsigned csr = TULIP_CSR_READ(sc, csr_srom_mii) & (MII_RD|MII_DOUT|MII_CLK);
int idx;
TULIP_LOCK_ASSERT(sc);
for (idx = 0, data = 0; idx < 16; idx++) {
data <<= 1; /* this is NOOP on the first pass through */
csr ^= MII_CLKON; MII_EMIT; /* clock high; data valid */
if (TULIP_CSR_READ(sc, csr_srom_mii) & MII_DIN)
data |= 1;
csr ^= MII_CLKOFF; MII_EMIT; /* clock low; data not valid */
}
csr ^= MII_RD; MII_EMIT; /* clock low; turn off read */
return data;
}
static unsigned
tulip_mii_readreg(tulip_softc_t * const sc, unsigned devaddr, unsigned regno)
{
unsigned csr = TULIP_CSR_READ(sc, csr_srom_mii) & (MII_RD|MII_DOUT|MII_CLK);
unsigned data;
TULIP_LOCK_ASSERT(sc);
csr &= ~(MII_RD|MII_CLK); MII_EMIT;
tulip_mii_writebits(sc, MII_PREAMBLE, 32);
tulip_mii_writebits(sc, MII_RDCMD, 8);
tulip_mii_writebits(sc, devaddr, 5);
tulip_mii_writebits(sc, regno, 5);
tulip_mii_turnaround(sc, MII_RDCMD);
data = tulip_mii_readbits(sc);
#if defined(TULIP_DEBUG)
sc->tulip_dbg.dbg_phyregs[regno][0] = data;
sc->tulip_dbg.dbg_phyregs[regno][1]++;
#endif
return data;
}
static void
tulip_mii_writereg(tulip_softc_t * const sc, unsigned devaddr, unsigned regno,
unsigned data)
{
unsigned csr = TULIP_CSR_READ(sc, csr_srom_mii) & (MII_RD|MII_DOUT|MII_CLK);
TULIP_LOCK_ASSERT(sc);
csr &= ~(MII_RD|MII_CLK); MII_EMIT;
tulip_mii_writebits(sc, MII_PREAMBLE, 32);
tulip_mii_writebits(sc, MII_WRCMD, 8);
tulip_mii_writebits(sc, devaddr, 5);
tulip_mii_writebits(sc, regno, 5);
tulip_mii_turnaround(sc, MII_WRCMD);
tulip_mii_writebits(sc, data, 16);
#if defined(TULIP_DEBUG)
sc->tulip_dbg.dbg_phyregs[regno][2] = data;
sc->tulip_dbg.dbg_phyregs[regno][3]++;
#endif
}
#define tulip_mchash(mca) (ether_crc32_le(mca, 6) & 0x1FF)
#define tulip_srom_crcok(databuf) ( \
((ether_crc32_le(databuf, 126) & 0xFFFFU) ^ 0xFFFFU) == \
((databuf)[126] | ((databuf)[127] << 8)))
static void
tulip_identify_dec_nic(tulip_softc_t * const sc)
{
TULIP_LOCK_ASSERT(sc);
strcpy(sc->tulip_boardid, "DEC ");
#define D0 4
1999-08-09 14:43:39 +00:00
if (sc->tulip_chipid <= TULIP_21040)
return;
if (bcmp(sc->tulip_rombuf + 29, "DE500", 5) == 0
|| bcmp(sc->tulip_rombuf + 29, "DE450", 5) == 0) {
bcopy(sc->tulip_rombuf + 29, &sc->tulip_boardid[D0], 8);
sc->tulip_boardid[D0+8] = ' ';
}
#undef D0
}
static void
tulip_identify_znyx_nic(tulip_softc_t * const sc)
{
unsigned id = 0;
TULIP_LOCK_ASSERT(sc);
strcpy(sc->tulip_boardid, "ZNYX ZX3XX ");
if (sc->tulip_chipid == TULIP_21140 || sc->tulip_chipid == TULIP_21140A) {
unsigned znyx_ptr;
sc->tulip_boardid[8] = '4';
znyx_ptr = sc->tulip_rombuf[124] + 256 * sc->tulip_rombuf[125];
if (znyx_ptr < 26 || znyx_ptr > 116) {
sc->tulip_boardsw = &tulip_21140_znyx_zx34x_boardsw;
return;
}
/* ZX344 = 0010 .. 0013FF
*/
if (sc->tulip_rombuf[znyx_ptr] == 0x4A
&& sc->tulip_rombuf[znyx_ptr + 1] == 0x52
&& sc->tulip_rombuf[znyx_ptr + 2] == 0x01) {
id = sc->tulip_rombuf[znyx_ptr + 5] + 256 * sc->tulip_rombuf[znyx_ptr + 4];
if ((id >> 8) == (TULIP_ZNYX_ID_ZX342 >> 8)) {
sc->tulip_boardid[9] = '2';
if (id == TULIP_ZNYX_ID_ZX342B) {
sc->tulip_boardid[10] = 'B';
sc->tulip_boardid[11] = ' ';
}
sc->tulip_boardsw = &tulip_21140_znyx_zx34x_boardsw;
} else if (id == TULIP_ZNYX_ID_ZX344) {
sc->tulip_boardid[10] = '4';
sc->tulip_boardsw = &tulip_21140_znyx_zx34x_boardsw;
} else if (id == TULIP_ZNYX_ID_ZX345) {
sc->tulip_boardid[9] = (sc->tulip_rombuf[19] > 1) ? '8' : '5';
} else if (id == TULIP_ZNYX_ID_ZX346) {
sc->tulip_boardid[9] = '6';
} else if (id == TULIP_ZNYX_ID_ZX351) {
sc->tulip_boardid[8] = '5';
sc->tulip_boardid[9] = '1';
}
}
if (id == 0) {
/*
* Assume it's a ZX342...
*/
sc->tulip_boardsw = &tulip_21140_znyx_zx34x_boardsw;
}
return;
}
sc->tulip_boardid[8] = '1';
if (sc->tulip_chipid == TULIP_21041) {
sc->tulip_boardid[10] = '1';
return;
}
if (sc->tulip_rombuf[32] == 0x4A && sc->tulip_rombuf[33] == 0x52) {
id = sc->tulip_rombuf[37] + 256 * sc->tulip_rombuf[36];
if (id == TULIP_ZNYX_ID_ZX312T) {
sc->tulip_boardid[9] = '2';
sc->tulip_boardid[10] = 'T';
sc->tulip_boardid[11] = ' ';
sc->tulip_boardsw = &tulip_21040_10baset_only_boardsw;
} else if (id == TULIP_ZNYX_ID_ZX314_INTA) {
sc->tulip_boardid[9] = '4';
sc->tulip_boardsw = &tulip_21040_10baset_only_boardsw;
sc->tulip_features |= TULIP_HAVE_SHAREDINTR|TULIP_HAVE_BASEROM;
} else if (id == TULIP_ZNYX_ID_ZX314) {
sc->tulip_boardid[9] = '4';
sc->tulip_boardsw = &tulip_21040_10baset_only_boardsw;
sc->tulip_features |= TULIP_HAVE_BASEROM;
} else if (id == TULIP_ZNYX_ID_ZX315_INTA) {
sc->tulip_boardid[9] = '5';
sc->tulip_features |= TULIP_HAVE_SHAREDINTR|TULIP_HAVE_BASEROM;
} else if (id == TULIP_ZNYX_ID_ZX315) {
sc->tulip_boardid[9] = '5';
sc->tulip_features |= TULIP_HAVE_BASEROM;
} else {
id = 0;
}
}
if (id == 0) {
if ((sc->tulip_enaddr[3] & ~3) == 0xF0 && (sc->tulip_enaddr[5] & 2) == 0) {
sc->tulip_boardid[9] = '4';
sc->tulip_boardsw = &tulip_21040_10baset_only_boardsw;
sc->tulip_features |= TULIP_HAVE_SHAREDINTR|TULIP_HAVE_BASEROM;
} else if ((sc->tulip_enaddr[3] & ~3) == 0xF4 && (sc->tulip_enaddr[5] & 1) == 0) {
sc->tulip_boardid[9] = '5';
sc->tulip_boardsw = &tulip_21040_boardsw;
sc->tulip_features |= TULIP_HAVE_SHAREDINTR|TULIP_HAVE_BASEROM;
} else if ((sc->tulip_enaddr[3] & ~3) == 0xEC) {
sc->tulip_boardid[9] = '2';
sc->tulip_boardsw = &tulip_21040_boardsw;
}
}
}
static void
tulip_identify_smc_nic(tulip_softc_t * const sc)
{
u_int32_t id1, id2, ei;
int auibnc = 0, utp = 0;
char *cp;
TULIP_LOCK_ASSERT(sc);
strcpy(sc->tulip_boardid, "SMC ");
if (sc->tulip_chipid == TULIP_21041)
return;
if (sc->tulip_chipid != TULIP_21040) {
if (sc->tulip_boardsw != &tulip_2114x_isv_boardsw) {
strcpy(&sc->tulip_boardid[4], "9332DST ");
sc->tulip_boardsw = &tulip_21140_smc9332_boardsw;
} else if (sc->tulip_features & (TULIP_HAVE_BASEROM|TULIP_HAVE_SLAVEDROM)) {
strcpy(&sc->tulip_boardid[4], "9334BDT ");
1997-08-03 13:00:42 +00:00
} else {
strcpy(&sc->tulip_boardid[4], "9332BDT ");
}
return;
}
id1 = sc->tulip_rombuf[0x60] | (sc->tulip_rombuf[0x61] << 8);
id2 = sc->tulip_rombuf[0x62] | (sc->tulip_rombuf[0x63] << 8);
ei = sc->tulip_rombuf[0x66] | (sc->tulip_rombuf[0x67] << 8);
strcpy(&sc->tulip_boardid[4], "8432");
cp = &sc->tulip_boardid[8];
if ((id1 & 1) == 0)
*cp++ = 'B', auibnc = 1;
if ((id1 & 0xFF) > 0x32)
*cp++ = 'T', utp = 1;
if ((id1 & 0x4000) == 0)
*cp++ = 'A', auibnc = 1;
if (id2 == 0x15) {
sc->tulip_boardid[7] = '4';
*cp++ = '-';
*cp++ = 'C';
*cp++ = 'H';
*cp++ = (ei ? '2' : '1');
}
*cp++ = ' ';
*cp = '\0';
if (utp && !auibnc)
sc->tulip_boardsw = &tulip_21040_10baset_only_boardsw;
else if (!utp && auibnc)
sc->tulip_boardsw = &tulip_21040_auibnc_only_boardsw;
}
static void
tulip_identify_cogent_nic(tulip_softc_t * const sc)
{
TULIP_LOCK_ASSERT(sc);
strcpy(sc->tulip_boardid, "Cogent ");
if (sc->tulip_chipid == TULIP_21140 || sc->tulip_chipid == TULIP_21140A) {
1997-08-03 13:00:42 +00:00
if (sc->tulip_rombuf[32] == TULIP_COGENT_EM100TX_ID) {
strcat(sc->tulip_boardid, "EM100TX ");
sc->tulip_boardsw = &tulip_21140_cogent_em100_boardsw;
#if defined(TULIP_COGENT_EM110TX_ID)
} else if (sc->tulip_rombuf[32] == TULIP_COGENT_EM110TX_ID) {
strcat(sc->tulip_boardid, "EM110TX ");
sc->tulip_boardsw = &tulip_21140_cogent_em100_boardsw;
#endif
1997-08-03 13:00:42 +00:00
} else if (sc->tulip_rombuf[32] == TULIP_COGENT_EM100FX_ID) {
strcat(sc->tulip_boardid, "EM100FX ");
sc->tulip_boardsw = &tulip_21140_cogent_em100_boardsw;
}
/*
* Magic number (0x24001109U) is the SubVendor (0x2400) and
* SubDevId (0x1109) for the ANA6944TX (EM440TX).
*/
if (*(u_int32_t *) sc->tulip_rombuf == 0x24001109U
&& (sc->tulip_features & TULIP_HAVE_BASEROM)) {
/*
* Cogent (Adaptec) is still mapping all INTs to INTA of
* first 21140. Dumb! Dumb!
*/
1997-08-03 13:00:42 +00:00
strcat(sc->tulip_boardid, "EM440TX ");
sc->tulip_features |= TULIP_HAVE_SHAREDINTR;
}
} else if (sc->tulip_chipid == TULIP_21040) {
sc->tulip_features |= TULIP_HAVE_SHAREDINTR|TULIP_HAVE_BASEROM;
}
}
static void
tulip_identify_accton_nic(tulip_softc_t * const sc)
{
TULIP_LOCK_ASSERT(sc);
strcpy(sc->tulip_boardid, "ACCTON ");
switch (sc->tulip_chipid) {
case TULIP_21140A:
strcat(sc->tulip_boardid, "EN1207 ");
if (sc->tulip_boardsw != &tulip_2114x_isv_boardsw)
sc->tulip_boardsw = &tulip_21140_accton_boardsw;
break;
case TULIP_21140:
strcat(sc->tulip_boardid, "EN1207TX ");
if (sc->tulip_boardsw != &tulip_2114x_isv_boardsw)
sc->tulip_boardsw = &tulip_21140_eb_boardsw;
break;
case TULIP_21040:
strcat(sc->tulip_boardid, "EN1203 ");
sc->tulip_boardsw = &tulip_21040_boardsw;
break;
case TULIP_21041:
strcat(sc->tulip_boardid, "EN1203 ");
sc->tulip_boardsw = &tulip_21041_boardsw;
break;
default:
sc->tulip_boardsw = &tulip_2114x_isv_boardsw;
break;
}
}
static void
tulip_identify_asante_nic(tulip_softc_t * const sc)
{
TULIP_LOCK_ASSERT(sc);
strcpy(sc->tulip_boardid, "Asante ");
if ((sc->tulip_chipid == TULIP_21140 || sc->tulip_chipid == TULIP_21140A)
&& sc->tulip_boardsw != &tulip_2114x_isv_boardsw) {
tulip_media_info_t *mi = sc->tulip_mediainfo;
int idx;
/*
* The Asante Fast Ethernet doesn't always ship with a valid
* new format SROM. So if isn't in the new format, we cheat
* set it up as if we had.
*/
sc->tulip_gpinit = TULIP_GP_ASANTE_PINS;
sc->tulip_gpdata = 0;
TULIP_CSR_WRITE(sc, csr_gp, TULIP_GP_ASANTE_PINS|TULIP_GP_PINSET);
TULIP_CSR_WRITE(sc, csr_gp, TULIP_GP_ASANTE_PHYRESET);
DELAY(100);
TULIP_CSR_WRITE(sc, csr_gp, 0);
mi->mi_type = TULIP_MEDIAINFO_MII;
mi->mi_gpr_length = 0;
mi->mi_gpr_offset = 0;
mi->mi_reset_length = 0;
mi->mi_reset_offset = 0;;
mi->mi_phyaddr = TULIP_MII_NOPHY;
for (idx = 20; idx > 0 && mi->mi_phyaddr == TULIP_MII_NOPHY; idx--) {
DELAY(10000);
mi->mi_phyaddr = tulip_mii_get_phyaddr(sc, 0);
}
if (mi->mi_phyaddr == TULIP_MII_NOPHY) {
if_printf(sc->tulip_ifp, "can't find phy 0\n");
return;
}
sc->tulip_features |= TULIP_HAVE_MII;
mi->mi_capabilities = PHYSTS_10BASET|PHYSTS_10BASET_FD|PHYSTS_100BASETX|PHYSTS_100BASETX_FD;
mi->mi_advertisement = PHYSTS_10BASET|PHYSTS_10BASET_FD|PHYSTS_100BASETX|PHYSTS_100BASETX_FD;
mi->mi_full_duplex = PHYSTS_10BASET_FD|PHYSTS_100BASETX_FD;
mi->mi_tx_threshold = PHYSTS_10BASET|PHYSTS_10BASET_FD;
TULIP_MEDIAINFO_ADD_CAPABILITY(sc, mi, 100BASETX_FD);
TULIP_MEDIAINFO_ADD_CAPABILITY(sc, mi, 100BASETX);
TULIP_MEDIAINFO_ADD_CAPABILITY(sc, mi, 100BASET4);
TULIP_MEDIAINFO_ADD_CAPABILITY(sc, mi, 10BASET_FD);
TULIP_MEDIAINFO_ADD_CAPABILITY(sc, mi, 10BASET);
mi->mi_phyid = (tulip_mii_readreg(sc, mi->mi_phyaddr, PHYREG_IDLOW) << 16) |
tulip_mii_readreg(sc, mi->mi_phyaddr, PHYREG_IDHIGH);
sc->tulip_boardsw = &tulip_2114x_isv_boardsw;
}
}
static void
tulip_identify_compex_nic(tulip_softc_t * const sc)
{
TULIP_LOCK_ASSERT(sc);
strcpy(sc->tulip_boardid, "COMPEX ");
if (sc->tulip_chipid == TULIP_21140A) {
int root_unit;
tulip_softc_t *root_sc = NULL;
strcat(sc->tulip_boardid, "400TX/PCI ");
/*
* All 4 chips on these boards share an interrupt. This code
* copied from tulip_read_macaddr.
*/
sc->tulip_features |= TULIP_HAVE_SHAREDINTR;
for (root_unit = sc->tulip_unit - 1; root_unit >= 0; root_unit--) {
1999-08-09 14:43:39 +00:00
root_sc = tulips[root_unit];
if (root_sc == NULL
|| !(root_sc->tulip_features & TULIP_HAVE_SLAVEDINTR))
break;
root_sc = NULL;
}
if (root_sc != NULL
&& root_sc->tulip_chipid == sc->tulip_chipid
&& root_sc->tulip_pci_busno == sc->tulip_pci_busno) {
sc->tulip_features |= TULIP_HAVE_SLAVEDINTR;
sc->tulip_slaves = root_sc->tulip_slaves;
root_sc->tulip_slaves = sc;
} else if(sc->tulip_features & TULIP_HAVE_SLAVEDINTR) {
printf("\nCannot find master device for %s interrupts",
sc->tulip_ifp->if_xname);
}
} else {
strcat(sc->tulip_boardid, "unknown ");
}
/* sc->tulip_boardsw = &tulip_21140_eb_boardsw; */
return;
}
static int
tulip_srom_decode(tulip_softc_t * const sc)
{
1997-08-03 13:00:42 +00:00
unsigned idx1, idx2, idx3;
const tulip_srom_header_t *shp = (const tulip_srom_header_t *) &sc->tulip_rombuf[0];
const tulip_srom_adapter_info_t *saip = (const tulip_srom_adapter_info_t *) (shp + 1);
tulip_srom_media_t srom_media;
tulip_media_info_t *mi = sc->tulip_mediainfo;
const u_int8_t *dp;
u_int32_t leaf_offset, blocks, data;
TULIP_LOCK_ASSERT(sc);
for (idx1 = 0; idx1 < shp->sh_adapter_count; idx1++, saip++) {
if (shp->sh_adapter_count == 1)
break;
if (saip->sai_device == sc->tulip_pci_devno)
break;
}
/*
* Didn't find the right media block for this card.
*/
if (idx1 == shp->sh_adapter_count)
return 0;
/*
* Save the hardware address.
*/
bcopy(shp->sh_ieee802_address, sc->tulip_enaddr, 6);
/*
* If this is a multiple port card, add the adapter index to the last
* byte of the hardware address. (if it isn't multiport, adding 0
* won't hurt.
*/
sc->tulip_enaddr[5] += idx1;
leaf_offset = saip->sai_leaf_offset_lowbyte
+ saip->sai_leaf_offset_highbyte * 256;
dp = sc->tulip_rombuf + leaf_offset;
sc->tulip_conntype = (tulip_srom_connection_t) (dp[0] + dp[1] * 256); dp += 2;
for (idx2 = 0;; idx2++) {
if (tulip_srom_conninfo[idx2].sc_type == sc->tulip_conntype
|| tulip_srom_conninfo[idx2].sc_type == TULIP_SROM_CONNTYPE_NOT_USED)
break;
}
sc->tulip_connidx = idx2;
if (sc->tulip_chipid == TULIP_21041) {
blocks = *dp++;
for (idx2 = 0; idx2 < blocks; idx2++) {
tulip_media_t media;
data = *dp++;
srom_media = (tulip_srom_media_t) (data & 0x3F);
for (idx3 = 0; tulip_srom_mediums[idx3].sm_type != TULIP_MEDIA_UNKNOWN; idx3++) {
if (tulip_srom_mediums[idx3].sm_srom_type == srom_media)
break;
}
media = tulip_srom_mediums[idx3].sm_type;
if (media != TULIP_MEDIA_UNKNOWN) {
if (data & TULIP_SROM_21041_EXTENDED) {
mi->mi_type = TULIP_MEDIAINFO_SIA;
sc->tulip_mediums[media] = mi;
mi->mi_sia_connectivity = dp[0] + dp[1] * 256;
mi->mi_sia_tx_rx = dp[2] + dp[3] * 256;
mi->mi_sia_general = dp[4] + dp[5] * 256;
mi++;
} else {
switch (media) {
case TULIP_MEDIA_BNC: {
TULIP_MEDIAINFO_SIA_INIT(sc, mi, 21041, BNC);
mi++;
break;
}
case TULIP_MEDIA_AUI: {
TULIP_MEDIAINFO_SIA_INIT(sc, mi, 21041, AUI);
mi++;
break;
}
case TULIP_MEDIA_10BASET: {
TULIP_MEDIAINFO_SIA_INIT(sc, mi, 21041, 10BASET);
mi++;
break;
}
case TULIP_MEDIA_10BASET_FD: {
TULIP_MEDIAINFO_SIA_INIT(sc, mi, 21041, 10BASET_FD);
mi++;
break;
}
default: {
break;
}
}
}
}
if (data & TULIP_SROM_21041_EXTENDED)
dp += 6;
}
#ifdef notdef
if (blocks == 0) {
TULIP_MEDIAINFO_SIA_INIT(sc, mi, 21041, BNC); mi++;
TULIP_MEDIAINFO_SIA_INIT(sc, mi, 21041, AUI); mi++;
TULIP_MEDIAINFO_SIA_INIT(sc, mi, 21041, 10BASET); mi++;
TULIP_MEDIAINFO_SIA_INIT(sc, mi, 21041, 10BASET_FD); mi++;
}
#endif
} else {
unsigned length, type;
tulip_media_t gp_media = TULIP_MEDIA_UNKNOWN;
if (sc->tulip_features & TULIP_HAVE_GPR)
sc->tulip_gpinit = *dp++;
blocks = *dp++;
for (idx2 = 0; idx2 < blocks; idx2++) {
const u_int8_t *ep;
if ((*dp & 0x80) == 0) {
length = 4;
type = 0;
} else {
length = (*dp++ & 0x7f) - 1;
type = *dp++ & 0x3f;
}
ep = dp + length;
switch (type & 0x3f) {
case 0: { /* 21140[A] GPR block */
tulip_media_t media;
srom_media = (tulip_srom_media_t)(dp[0] & 0x3f);
for (idx3 = 0; tulip_srom_mediums[idx3].sm_type != TULIP_MEDIA_UNKNOWN; idx3++) {
if (tulip_srom_mediums[idx3].sm_srom_type == srom_media)
break;
}
media = tulip_srom_mediums[idx3].sm_type;
if (media == TULIP_MEDIA_UNKNOWN)
break;
mi->mi_type = TULIP_MEDIAINFO_GPR;
sc->tulip_mediums[media] = mi;
mi->mi_gpdata = dp[1];
if (media > gp_media && !TULIP_IS_MEDIA_FD(media)) {
sc->tulip_gpdata = mi->mi_gpdata;
gp_media = media;
}
data = dp[2] + dp[3] * 256;
mi->mi_cmdmode = TULIP_SROM_2114X_CMDBITS(data);
if (data & TULIP_SROM_2114X_NOINDICATOR) {
mi->mi_actmask = 0;
} else {
#if 0
mi->mi_default = (data & TULIP_SROM_2114X_DEFAULT) != 0;
#endif
mi->mi_actmask = TULIP_SROM_2114X_BITPOS(data);
mi->mi_actdata = (data & TULIP_SROM_2114X_POLARITY) ? 0 : mi->mi_actmask;
}
mi++;
break;
}
case 1: { /* 21140[A] MII block */
const unsigned phyno = *dp++;
mi->mi_type = TULIP_MEDIAINFO_MII;
mi->mi_gpr_length = *dp++;
mi->mi_gpr_offset = dp - sc->tulip_rombuf;
dp += mi->mi_gpr_length;
mi->mi_reset_length = *dp++;
mi->mi_reset_offset = dp - sc->tulip_rombuf;
dp += mi->mi_reset_length;
/*
* Before we probe for a PHY, use the GPR information
* to select it. If we don't, it may be inaccessible.
*/
TULIP_CSR_WRITE(sc, csr_gp, sc->tulip_gpinit|TULIP_GP_PINSET);
for (idx3 = 0; idx3 < mi->mi_reset_length; idx3++) {
DELAY(10);
TULIP_CSR_WRITE(sc, csr_gp, sc->tulip_rombuf[mi->mi_reset_offset + idx3]);
}
sc->tulip_phyaddr = mi->mi_phyaddr;
for (idx3 = 0; idx3 < mi->mi_gpr_length; idx3++) {
DELAY(10);
TULIP_CSR_WRITE(sc, csr_gp, sc->tulip_rombuf[mi->mi_gpr_offset + idx3]);
}
/*
* At least write something!
*/
if (mi->mi_reset_length == 0 && mi->mi_gpr_length == 0)
TULIP_CSR_WRITE(sc, csr_gp, 0);
mi->mi_phyaddr = TULIP_MII_NOPHY;
for (idx3 = 20; idx3 > 0 && mi->mi_phyaddr == TULIP_MII_NOPHY; idx3--) {
DELAY(10000);
mi->mi_phyaddr = tulip_mii_get_phyaddr(sc, phyno);
}
if (mi->mi_phyaddr == TULIP_MII_NOPHY) {
#if defined(TULIP_DEBUG)
if_printf(sc->tulip_ifp, "can't find phy %d\n",
phyno);
#endif
break;
}
sc->tulip_features |= TULIP_HAVE_MII;
mi->mi_capabilities = dp[0] + dp[1] * 256; dp += 2;
mi->mi_advertisement = dp[0] + dp[1] * 256; dp += 2;
mi->mi_full_duplex = dp[0] + dp[1] * 256; dp += 2;
mi->mi_tx_threshold = dp[0] + dp[1] * 256; dp += 2;
TULIP_MEDIAINFO_ADD_CAPABILITY(sc, mi, 100BASETX_FD);
TULIP_MEDIAINFO_ADD_CAPABILITY(sc, mi, 100BASETX);
TULIP_MEDIAINFO_ADD_CAPABILITY(sc, mi, 100BASET4);
TULIP_MEDIAINFO_ADD_CAPABILITY(sc, mi, 10BASET_FD);
TULIP_MEDIAINFO_ADD_CAPABILITY(sc, mi, 10BASET);
mi->mi_phyid = (tulip_mii_readreg(sc, mi->mi_phyaddr, PHYREG_IDLOW) << 16) |
tulip_mii_readreg(sc, mi->mi_phyaddr, PHYREG_IDHIGH);
mi++;
break;
}
case 2: { /* 2114[23] SIA block */
tulip_media_t media;
srom_media = (tulip_srom_media_t)(dp[0] & 0x3f);
for (idx3 = 0; tulip_srom_mediums[idx3].sm_type != TULIP_MEDIA_UNKNOWN; idx3++) {
if (tulip_srom_mediums[idx3].sm_srom_type == srom_media)
break;
}
media = tulip_srom_mediums[idx3].sm_type;
if (media == TULIP_MEDIA_UNKNOWN)
break;
mi->mi_type = TULIP_MEDIAINFO_SIA;
sc->tulip_mediums[media] = mi;
if (dp[0] & 0x40) {
mi->mi_sia_connectivity = dp[1] + dp[2] * 256;
mi->mi_sia_tx_rx = dp[3] + dp[4] * 256;
mi->mi_sia_general = dp[5] + dp[6] * 256;
dp += 6;
} else {
switch (media) {
case TULIP_MEDIA_BNC: {
TULIP_MEDIAINFO_SIA_INIT(sc, mi, 21142, BNC);
break;
}
case TULIP_MEDIA_AUI: {
TULIP_MEDIAINFO_SIA_INIT(sc, mi, 21142, AUI);
break;
}
case TULIP_MEDIA_10BASET: {
TULIP_MEDIAINFO_SIA_INIT(sc, mi, 21142, 10BASET);
sc->tulip_intrmask |= TULIP_STS_LINKPASS|TULIP_STS_LINKFAIL;
break;
}
case TULIP_MEDIA_10BASET_FD: {
TULIP_MEDIAINFO_SIA_INIT(sc, mi, 21142, 10BASET_FD);
sc->tulip_intrmask |= TULIP_STS_LINKPASS|TULIP_STS_LINKFAIL;
break;
}
default: {
goto bad_media;
}
}
}
mi->mi_sia_gp_control = (dp[1] + dp[2] * 256) << 16;
mi->mi_sia_gp_data = (dp[3] + dp[4] * 256) << 16;
mi++;
bad_media:
break;
}
case 3: { /* 2114[23] MII PHY block */
const unsigned phyno = *dp++;
const u_int8_t *dp0;
mi->mi_type = TULIP_MEDIAINFO_MII;
mi->mi_gpr_length = *dp++;
mi->mi_gpr_offset = dp - sc->tulip_rombuf;
dp += 2 * mi->mi_gpr_length;
mi->mi_reset_length = *dp++;
mi->mi_reset_offset = dp - sc->tulip_rombuf;
dp += 2 * mi->mi_reset_length;
dp0 = &sc->tulip_rombuf[mi->mi_reset_offset];
for (idx3 = 0; idx3 < mi->mi_reset_length; idx3++, dp0 += 2) {
DELAY(10);
TULIP_CSR_WRITE(sc, csr_sia_general, (dp0[0] + 256 * dp0[1]) << 16);
}
sc->tulip_phyaddr = mi->mi_phyaddr;
dp0 = &sc->tulip_rombuf[mi->mi_gpr_offset];
for (idx3 = 0; idx3 < mi->mi_gpr_length; idx3++, dp0 += 2) {
DELAY(10);
TULIP_CSR_WRITE(sc, csr_sia_general, (dp0[0] + 256 * dp0[1]) << 16);
}
if (mi->mi_reset_length == 0 && mi->mi_gpr_length == 0)
TULIP_CSR_WRITE(sc, csr_sia_general, 0);
mi->mi_phyaddr = TULIP_MII_NOPHY;
for (idx3 = 20; idx3 > 0 && mi->mi_phyaddr == TULIP_MII_NOPHY; idx3--) {
DELAY(10000);
mi->mi_phyaddr = tulip_mii_get_phyaddr(sc, phyno);
}
if (mi->mi_phyaddr == TULIP_MII_NOPHY) {
#if defined(TULIP_DEBUG)
if_printf(sc->tulip_ifp, "can't find phy %d\n",
phyno);
#endif
break;
}
sc->tulip_features |= TULIP_HAVE_MII;
mi->mi_capabilities = dp[0] + dp[1] * 256; dp += 2;
mi->mi_advertisement = dp[0] + dp[1] * 256; dp += 2;
mi->mi_full_duplex = dp[0] + dp[1] * 256; dp += 2;
mi->mi_tx_threshold = dp[0] + dp[1] * 256; dp += 2;
mi->mi_mii_interrupt = dp[0] + dp[1] * 256; dp += 2;
TULIP_MEDIAINFO_ADD_CAPABILITY(sc, mi, 100BASETX_FD);
TULIP_MEDIAINFO_ADD_CAPABILITY(sc, mi, 100BASETX);
TULIP_MEDIAINFO_ADD_CAPABILITY(sc, mi, 100BASET4);
TULIP_MEDIAINFO_ADD_CAPABILITY(sc, mi, 10BASET_FD);
TULIP_MEDIAINFO_ADD_CAPABILITY(sc, mi, 10BASET);
mi->mi_phyid = (tulip_mii_readreg(sc, mi->mi_phyaddr, PHYREG_IDLOW) << 16) |
tulip_mii_readreg(sc, mi->mi_phyaddr, PHYREG_IDHIGH);
mi++;
break;
}
case 4: { /* 21143 SYM block */
tulip_media_t media;
srom_media = (tulip_srom_media_t) dp[0];
for (idx3 = 0; tulip_srom_mediums[idx3].sm_type != TULIP_MEDIA_UNKNOWN; idx3++) {
if (tulip_srom_mediums[idx3].sm_srom_type == srom_media)
break;
}
media = tulip_srom_mediums[idx3].sm_type;
if (media == TULIP_MEDIA_UNKNOWN)
break;
mi->mi_type = TULIP_MEDIAINFO_SYM;
sc->tulip_mediums[media] = mi;
mi->mi_gpcontrol = (dp[1] + dp[2] * 256) << 16;
mi->mi_gpdata = (dp[3] + dp[4] * 256) << 16;
data = dp[5] + dp[6] * 256;
mi->mi_cmdmode = TULIP_SROM_2114X_CMDBITS(data);
if (data & TULIP_SROM_2114X_NOINDICATOR) {
mi->mi_actmask = 0;
} else {
mi->mi_default = (data & TULIP_SROM_2114X_DEFAULT) != 0;
mi->mi_actmask = TULIP_SROM_2114X_BITPOS(data);
mi->mi_actdata = (data & TULIP_SROM_2114X_POLARITY) ? 0 : mi->mi_actmask;
}
if (TULIP_IS_MEDIA_TP(media))
sc->tulip_intrmask |= TULIP_STS_LINKPASS|TULIP_STS_LINKFAIL;
mi++;
break;
}
#if 0
case 5: { /* 21143 Reset block */
mi->mi_type = TULIP_MEDIAINFO_RESET;
mi->mi_reset_length = *dp++;
mi->mi_reset_offset = dp - sc->tulip_rombuf;
dp += 2 * mi->mi_reset_length;
mi++;
break;
}
#endif
default: {
}
}
dp = ep;
}
}
return mi - sc->tulip_mediainfo;
}
static const struct {
void (*vendor_identify_nic)(tulip_softc_t * const sc);
unsigned char vendor_oui[3];
} tulip_vendors[] = {
{ tulip_identify_dec_nic, { 0x08, 0x00, 0x2B } },
{ tulip_identify_dec_nic, { 0x00, 0x00, 0xF8 } },
{ tulip_identify_smc_nic, { 0x00, 0x00, 0xC0 } },
{ tulip_identify_smc_nic, { 0x00, 0xE0, 0x29 } },
{ tulip_identify_znyx_nic, { 0x00, 0xC0, 0x95 } },
{ tulip_identify_cogent_nic, { 0x00, 0x00, 0x92 } },
{ tulip_identify_asante_nic, { 0x00, 0x00, 0x94 } },
{ tulip_identify_cogent_nic, { 0x00, 0x00, 0xD1 } },
{ tulip_identify_accton_nic, { 0x00, 0x00, 0xE8 } },
{ tulip_identify_compex_nic, { 0x00, 0x80, 0x48 } },
{ NULL }
};
/*
* This deals with the vagaries of the address roms and the
* brain-deadness that various vendors commit in using them.
*/
static int
tulip_read_macaddr(tulip_softc_t * const sc)
{
1997-08-03 13:00:42 +00:00
unsigned cksum, rom_cksum, idx;
u_int32_t csr;
unsigned char tmpbuf[8];
static const u_char testpat[] = { 0xFF, 0, 0x55, 0xAA, 0xFF, 0, 0x55, 0xAA };
sc->tulip_connidx = TULIP_SROM_LASTCONNIDX;
if (sc->tulip_chipid == TULIP_21040) {
TULIP_CSR_WRITE(sc, csr_enetrom, 1);
for (idx = 0; idx < sizeof(sc->tulip_rombuf); idx++) {
int cnt = 0;
while (((csr = TULIP_CSR_READ(sc, csr_enetrom)) & 0x80000000L) && cnt < 10000)
cnt++;
sc->tulip_rombuf[idx] = csr & 0xFF;
}
sc->tulip_boardsw = &tulip_21040_boardsw;
} else {
if (sc->tulip_chipid == TULIP_21041) {
/*
* Thankfully all 21041's act the same.
*/
sc->tulip_boardsw = &tulip_21041_boardsw;
} else {
/*
* Assume all 21140 board are compatible with the
* DEC 10/100 evaluation board. Not really valid but
* it's the best we can do until every one switches to
* the new SROM format.
*/
sc->tulip_boardsw = &tulip_21140_eb_boardsw;
}
tulip_srom_read(sc);
if (tulip_srom_crcok(sc->tulip_rombuf)) {
/*
* SROM CRC is valid therefore it must be in the
* new format.
*/
sc->tulip_features |= TULIP_HAVE_ISVSROM|TULIP_HAVE_OKSROM;
} else if (sc->tulip_rombuf[126] == 0xff && sc->tulip_rombuf[127] == 0xFF) {
/*
* No checksum is present. See if the SROM id checks out;
* the first 18 bytes should be 0 followed by a 1 followed
* by the number of adapters (which we don't deal with yet).
*/
for (idx = 0; idx < 18; idx++) {
if (sc->tulip_rombuf[idx] != 0)
break;
}
if (idx == 18 && sc->tulip_rombuf[18] == 1 && sc->tulip_rombuf[19] != 0)
sc->tulip_features |= TULIP_HAVE_ISVSROM;
} else if (sc->tulip_chipid >= TULIP_21142) {
sc->tulip_features |= TULIP_HAVE_ISVSROM;
sc->tulip_boardsw = &tulip_2114x_isv_boardsw;
}
if ((sc->tulip_features & TULIP_HAVE_ISVSROM) && tulip_srom_decode(sc)) {
if (sc->tulip_chipid != TULIP_21041)
sc->tulip_boardsw = &tulip_2114x_isv_boardsw;
/*
* If the SROM specifies more than one adapter, tag this as a
* BASE rom.
*/
if (sc->tulip_rombuf[19] > 1)
sc->tulip_features |= TULIP_HAVE_BASEROM;
if (sc->tulip_boardsw == NULL)
return -6;
goto check_oui;
}
}
if (bcmp(&sc->tulip_rombuf[0], &sc->tulip_rombuf[16], 8) != 0) {
/*
* Some folks don't use the standard ethernet rom format
* but instead just put the address in the first 6 bytes
* of the rom and let the rest be all 0xffs. (Can we say
* ZNYX?) (well sometimes they put in a checksum so we'll
* start at 8).
*/
for (idx = 8; idx < 32; idx++) {
if (sc->tulip_rombuf[idx] != 0xFF)
return -4;
}
/*
* Make sure the address is not multicast or locally assigned
* that the OUI is not 00-00-00.
*/
if ((sc->tulip_rombuf[0] & 3) != 0)
return -4;
if (sc->tulip_rombuf[0] == 0 && sc->tulip_rombuf[1] == 0
&& sc->tulip_rombuf[2] == 0)
return -4;
bcopy(sc->tulip_rombuf, sc->tulip_enaddr, 6);
sc->tulip_features |= TULIP_HAVE_OKROM;
goto check_oui;
} else {
/*
* A number of makers of multiport boards (ZNYX and Cogent)
* only put on one address ROM on their 21040 boards. So
* if the ROM is all zeros (or all 0xFFs), look at the
* previous configured boards (as long as they are on the same
* PCI bus and the bus number is non-zero) until we find the
* master board with address ROM. We then use its address ROM
* as the base for this board. (we add our relative board
* to the last byte of its address).
*/
for (idx = 0; idx < sizeof(sc->tulip_rombuf); idx++) {
if (sc->tulip_rombuf[idx] != 0 && sc->tulip_rombuf[idx] != 0xFF)
break;
}
if (idx == sizeof(sc->tulip_rombuf)) {
int root_unit;
tulip_softc_t *root_sc = NULL;
for (root_unit = sc->tulip_unit - 1; root_unit >= 0; root_unit--) {
1999-08-09 14:43:39 +00:00
root_sc = tulips[root_unit];
if (root_sc == NULL || (root_sc->tulip_features & (TULIP_HAVE_OKROM|TULIP_HAVE_SLAVEDROM)) == TULIP_HAVE_OKROM)
break;
root_sc = NULL;
}
if (root_sc != NULL && (root_sc->tulip_features & TULIP_HAVE_BASEROM)
&& root_sc->tulip_chipid == sc->tulip_chipid
&& root_sc->tulip_pci_busno == sc->tulip_pci_busno) {
sc->tulip_features |= TULIP_HAVE_SLAVEDROM;
sc->tulip_boardsw = root_sc->tulip_boardsw;
strcpy(sc->tulip_boardid, root_sc->tulip_boardid);
if (sc->tulip_boardsw->bd_type == TULIP_21140_ISV) {
bcopy(root_sc->tulip_rombuf, sc->tulip_rombuf,
sizeof(sc->tulip_rombuf));
if (!tulip_srom_decode(sc))
return -5;
} else {
bcopy(root_sc->tulip_enaddr, sc->tulip_enaddr, 6);
sc->tulip_enaddr[5] += sc->tulip_unit - root_sc->tulip_unit;
}
/*
* Now for a truly disgusting kludge: all 4 21040s on
* the ZX314 share the same INTA line so the mapping
* setup by the BIOS on the PCI bridge is worthless.
* Rather than reprogramming the value in the config
* register, we will handle this internally.
*/
if (root_sc->tulip_features & TULIP_HAVE_SHAREDINTR) {
sc->tulip_slaves = root_sc->tulip_slaves;
root_sc->tulip_slaves = sc;
sc->tulip_features |= TULIP_HAVE_SLAVEDINTR;
}
return 0;
}
}
}
/*
* This is the standard DEC address ROM test.
*/
if (bcmp(&sc->tulip_rombuf[24], testpat, 8) != 0)
return -3;
tmpbuf[0] = sc->tulip_rombuf[15]; tmpbuf[1] = sc->tulip_rombuf[14];
tmpbuf[2] = sc->tulip_rombuf[13]; tmpbuf[3] = sc->tulip_rombuf[12];
tmpbuf[4] = sc->tulip_rombuf[11]; tmpbuf[5] = sc->tulip_rombuf[10];
tmpbuf[6] = sc->tulip_rombuf[9]; tmpbuf[7] = sc->tulip_rombuf[8];
if (bcmp(&sc->tulip_rombuf[0], tmpbuf, 8) != 0)
return -2;
bcopy(sc->tulip_rombuf, sc->tulip_enaddr, 6);
cksum = *(u_int16_t *) &sc->tulip_enaddr[0];
cksum *= 2;
if (cksum > 65535) cksum -= 65535;
cksum += *(u_int16_t *) &sc->tulip_enaddr[2];
if (cksum > 65535) cksum -= 65535;
cksum *= 2;
if (cksum > 65535) cksum -= 65535;
cksum += *(u_int16_t *) &sc->tulip_enaddr[4];
if (cksum >= 65535) cksum -= 65535;
rom_cksum = *(u_int16_t *) &sc->tulip_rombuf[6];
if (cksum != rom_cksum)
return -1;
check_oui:
/*
* Check for various boards based on OUI. Did I say braindead?
*/
for (idx = 0; tulip_vendors[idx].vendor_identify_nic != NULL; idx++) {
if (bcmp(sc->tulip_enaddr, tulip_vendors[idx].vendor_oui, 3) == 0) {
(*tulip_vendors[idx].vendor_identify_nic)(sc);
break;
}
}
sc->tulip_features |= TULIP_HAVE_OKROM;
return 0;
}
static void
tulip_ifmedia_add(tulip_softc_t * const sc)
{
tulip_media_t media;
int medias = 0;
TULIP_LOCK_ASSERT(sc);
for (media = TULIP_MEDIA_UNKNOWN; media < TULIP_MEDIA_MAX; media++) {
if (sc->tulip_mediums[media] != NULL) {
ifmedia_add(&sc->tulip_ifmedia, tulip_media_to_ifmedia[media],
0, 0);
medias++;
}
}
if (medias == 0) {
sc->tulip_features |= TULIP_HAVE_NOMEDIA;
ifmedia_add(&sc->tulip_ifmedia, IFM_ETHER | IFM_NONE, 0, 0);
ifmedia_set(&sc->tulip_ifmedia, IFM_ETHER | IFM_NONE);
} else if (sc->tulip_media == TULIP_MEDIA_UNKNOWN) {
ifmedia_add(&sc->tulip_ifmedia, IFM_ETHER | IFM_AUTO, 0, 0);
ifmedia_set(&sc->tulip_ifmedia, IFM_ETHER | IFM_AUTO);
} else {
ifmedia_set(&sc->tulip_ifmedia, tulip_media_to_ifmedia[sc->tulip_media]);
sc->tulip_flags |= TULIP_PRINTMEDIA;
tulip_linkup(sc, sc->tulip_media);
}
}
static int
tulip_ifmedia_change(struct ifnet * const ifp)
{
tulip_softc_t * const sc = (tulip_softc_t *)ifp->if_softc;
TULIP_LOCK(sc);
sc->tulip_flags |= TULIP_NEEDRESET;
sc->tulip_probe_state = TULIP_PROBE_INACTIVE;
sc->tulip_media = TULIP_MEDIA_UNKNOWN;
if (IFM_SUBTYPE(sc->tulip_ifmedia.ifm_media) != IFM_AUTO) {
tulip_media_t media;
for (media = TULIP_MEDIA_UNKNOWN; media < TULIP_MEDIA_MAX; media++) {
if (sc->tulip_mediums[media] != NULL
&& sc->tulip_ifmedia.ifm_media == tulip_media_to_ifmedia[media]) {
sc->tulip_flags |= TULIP_PRINTMEDIA;
sc->tulip_flags &= ~TULIP_DIDNWAY;
tulip_linkup(sc, media);
TULIP_UNLOCK(sc);
return 0;
}
}
}
sc->tulip_flags &= ~(TULIP_TXPROBE_ACTIVE|TULIP_WANTRXACT);
tulip_reset(sc);
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
tulip_init_locked(sc);
TULIP_UNLOCK(sc);
return 0;
}
/*
* Media status callback
*/
static void
tulip_ifmedia_status(struct ifnet * const ifp, struct ifmediareq *req)
{
tulip_softc_t *sc = (tulip_softc_t *)ifp->if_softc;
TULIP_LOCK(sc);
if (sc->tulip_media == TULIP_MEDIA_UNKNOWN) {
TULIP_UNLOCK(sc);
return;
}
req->ifm_status = IFM_AVALID;
if (sc->tulip_flags & TULIP_LINKUP)
req->ifm_status |= IFM_ACTIVE;
req->ifm_active = tulip_media_to_ifmedia[sc->tulip_media];
TULIP_UNLOCK(sc);
}
static void
tulip_addr_filter(tulip_softc_t * const sc)
{
struct ifmultiaddr *ifma;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
struct ifnet *ifp;
u_char *addrp;
u_int16_t eaddr[ETHER_ADDR_LEN/2];
int multicnt;
TULIP_LOCK_ASSERT(sc);
sc->tulip_flags &= ~(TULIP_WANTHASHPERFECT|TULIP_WANTHASHONLY|TULIP_ALLMULTI);
1997-08-03 13:00:42 +00:00
sc->tulip_flags |= TULIP_WANTSETUP|TULIP_WANTTXSTART;
sc->tulip_cmdmode &= ~TULIP_CMD_RXRUN;
sc->tulip_intrmask &= ~TULIP_STS_RXSTOPPED;
#if defined(IFF_ALLMULTI)
if (sc->tulip_ifp->if_flags & IFF_ALLMULTI)
sc->tulip_flags |= TULIP_ALLMULTI ;
#endif
multicnt = 0;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
ifp = sc->tulip_ifp;
IF_ADDR_LOCK(ifp);
/* Copy MAC address on stack to align. */
bcopy(IF_LLADDR(ifp), eaddr, ETHER_ADDR_LEN);
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
if (ifma->ifma_addr->sa_family == AF_LINK)
multicnt++;
}
if (multicnt > 14) {
u_int32_t *sp = sc->tulip_setupdata;
unsigned hash;
/*
* Some early passes of the 21140 have broken implementations of
* hash-perfect mode. When we get too many multicasts for perfect
* filtering with these chips, we need to switch into hash-only
* mode (this is better than all-multicast on network with lots
* of multicast traffic).
*/
if (sc->tulip_features & TULIP_HAVE_BROKEN_HASH)
sc->tulip_flags |= TULIP_WANTHASHONLY;
else
sc->tulip_flags |= TULIP_WANTHASHPERFECT;
/*
* If we have more than 14 multicasts, we have
* go into hash perfect mode (512 bit multicast
* hash and one perfect hardware).
*/
bzero(sc->tulip_setupdata, sizeof(sc->tulip_setupdata));
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
if (ifma->ifma_addr->sa_family != AF_LINK)
continue;
hash = tulip_mchash(LLADDR((struct sockaddr_dl *)ifma->ifma_addr));
sp[hash >> 4] |= htole32(1 << (hash & 0xF));
}
/*
* No reason to use a hash if we are going to be
* receiving every multicast.
*/
if ((sc->tulip_flags & TULIP_ALLMULTI) == 0) {
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
hash = tulip_mchash(ifp->if_broadcastaddr);
sp[hash >> 4] |= htole32(1 << (hash & 0xF));
if (sc->tulip_flags & TULIP_WANTHASHONLY) {
hash = tulip_mchash((caddr_t)eaddr);
sp[hash >> 4] |= htole32(1 << (hash & 0xF));
} else {
sp[39] = TULIP_SP_MAC(eaddr[0]);
sp[40] = TULIP_SP_MAC(eaddr[1]);
sp[41] = TULIP_SP_MAC(eaddr[2]);
}
}
}
if ((sc->tulip_flags & (TULIP_WANTHASHPERFECT|TULIP_WANTHASHONLY)) == 0) {
u_int32_t *sp = sc->tulip_setupdata;
int idx = 0;
if ((sc->tulip_flags & TULIP_ALLMULTI) == 0) {
/*
* Else can get perfect filtering for 16 addresses.
*/
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
if (ifma->ifma_addr->sa_family != AF_LINK)
continue;
addrp = LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
*sp++ = TULIP_SP_MAC(((u_int16_t *)addrp)[0]);
*sp++ = TULIP_SP_MAC(((u_int16_t *)addrp)[1]);
*sp++ = TULIP_SP_MAC(((u_int16_t *)addrp)[2]);
idx++;
}
/*
* Add the broadcast address.
*/
idx++;
*sp++ = TULIP_SP_MAC(0xFFFF);
*sp++ = TULIP_SP_MAC(0xFFFF);
*sp++ = TULIP_SP_MAC(0xFFFF);
}
/*
* Pad the rest with our hardware address
*/
for (; idx < 16; idx++) {
*sp++ = TULIP_SP_MAC(eaddr[0]);
*sp++ = TULIP_SP_MAC(eaddr[1]);
*sp++ = TULIP_SP_MAC(eaddr[2]);
}
}
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
IF_ADDR_UNLOCK(ifp);
}
static void
tulip_reset(tulip_softc_t * const sc)
{
tulip_ringinfo_t *ri;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
tulip_descinfo_t *di;
struct mbuf *m;
u_int32_t inreset = (sc->tulip_flags & TULIP_INRESET);
TULIP_LOCK_ASSERT(sc);
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
CTR1(KTR_TULIP, "tulip_reset: inreset %d", inreset);
/*
* Brilliant. Simply brilliant. When switching modes/speeds
* on a 2114*, you need to set the appriopriate MII/PCS/SCL/PS
* bits in CSR6 and then do a software reset to get the 21140
* to properly reset its internal pathways to the right places.
* Grrrr.
*/
if ((sc->tulip_flags & TULIP_DEVICEPROBE) == 0
&& sc->tulip_boardsw->bd_media_preset != NULL)
(*sc->tulip_boardsw->bd_media_preset)(sc);
TULIP_CSR_WRITE(sc, csr_busmode, TULIP_BUSMODE_SWRESET);
DELAY(10); /* Wait 10 microseconds (actually 50 PCI cycles but at
33MHz that comes to two microseconds but wait a
bit longer anyways) */
if (!inreset) {
sc->tulip_flags |= TULIP_INRESET;
sc->tulip_flags &= ~(TULIP_NEEDRESET|TULIP_RXBUFSLOW);
sc->tulip_ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
}
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
TULIP_CSR_WRITE(sc, csr_txlist, sc->tulip_txinfo.ri_dma_addr);
TULIP_CSR_WRITE(sc, csr_rxlist, sc->tulip_rxinfo.ri_dma_addr);
TULIP_CSR_WRITE(sc, csr_busmode,
(1 << (3 /*pci_max_burst_len*/ + 8))
|TULIP_BUSMODE_CACHE_ALIGN8
|TULIP_BUSMODE_READMULTIPLE
|(BYTE_ORDER != LITTLE_ENDIAN ?
TULIP_BUSMODE_DESC_BIGENDIAN : 0));
sc->tulip_txtimer = 0;
/*
* Free all the mbufs that were on the transmit ring.
*/
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
CTR0(KTR_TULIP, "tulip_reset: drain transmit ring");
ri = &sc->tulip_txinfo;
for (di = ri->ri_first; di < ri->ri_last; di++) {
m = tulip_dequeue_mbuf(ri, di, SYNC_NONE);
if (m != NULL)
m_freem(m);
di->di_desc->d_status = 0;
}
ri->ri_nextin = ri->ri_nextout = ri->ri_first;
ri->ri_free = ri->ri_max;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
TULIP_TXDESC_PRESYNC(ri);
/*
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
* We need to collect all the mbufs that were on the
* receive ring before we reinit it either to put
* them back on or to know if we have to allocate
* more.
*/
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
CTR0(KTR_TULIP, "tulip_reset: drain receive ring");
ri = &sc->tulip_rxinfo;
ri->ri_nextin = ri->ri_nextout = ri->ri_first;
ri->ri_free = ri->ri_max;
for (di = ri->ri_first; di < ri->ri_last; di++) {
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
di->di_desc->d_status = 0;
di->di_desc->d_length1 = 0; di->di_desc->d_addr1 = 0;
di->di_desc->d_length2 = 0; di->di_desc->d_addr2 = 0;
}
TULIP_RXDESC_PRESYNC(ri);
for (di = ri->ri_first; di < ri->ri_last; di++) {
m = tulip_dequeue_mbuf(ri, di, SYNC_NONE);
if (m != NULL)
m_freem(m);
}
/*
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
* If tulip_reset is being called recursively, exit quickly knowing
* that when the outer tulip_reset returns all the right stuff will
* have happened.
*/
if (inreset)
return;
sc->tulip_intrmask |= TULIP_STS_NORMALINTR|TULIP_STS_RXINTR|TULIP_STS_TXINTR
|TULIP_STS_ABNRMLINTR|TULIP_STS_SYSERROR|TULIP_STS_TXSTOPPED
|TULIP_STS_TXUNDERFLOW|TULIP_STS_TXBABBLE
1997-08-03 13:00:42 +00:00
|TULIP_STS_RXSTOPPED;
if ((sc->tulip_flags & TULIP_DEVICEPROBE) == 0)
(*sc->tulip_boardsw->bd_media_select)(sc);
#if defined(TULIP_DEBUG)
if ((sc->tulip_flags & TULIP_NEEDRESET) == TULIP_NEEDRESET)
if_printf(sc->tulip_ifp,
"tulip_reset: additional reset needed?!?\n");
#endif
if (bootverbose)
tulip_media_print(sc);
if (sc->tulip_features & TULIP_HAVE_DUALSENSE)
TULIP_CSR_WRITE(sc, csr_sia_status, TULIP_CSR_READ(sc, csr_sia_status));
sc->tulip_flags &= ~(TULIP_DOINGSETUP|TULIP_WANTSETUP|TULIP_INRESET
|TULIP_RXACT);
}
static void
tulip_init(void *arg)
{
tulip_softc_t *sc = (tulip_softc_t *)arg;
TULIP_LOCK(sc);
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
tulip_init_locked(sc);
TULIP_UNLOCK(sc);
}
static void
tulip_init_locked(tulip_softc_t * const sc)
{
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
CTR0(KTR_TULIP, "tulip_init_locked");
if (sc->tulip_ifp->if_flags & IFF_UP) {
if ((sc->tulip_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
/* initialize the media */
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
CTR0(KTR_TULIP, "tulip_init_locked: up but not running, reset chip");
tulip_reset(sc);
}
tulip_addr_filter(sc);
sc->tulip_ifp->if_drv_flags |= IFF_DRV_RUNNING;
if (sc->tulip_ifp->if_flags & IFF_PROMISC) {
sc->tulip_flags |= TULIP_PROMISC;
sc->tulip_cmdmode |= TULIP_CMD_PROMISCUOUS;
1997-08-03 13:00:42 +00:00
sc->tulip_intrmask |= TULIP_STS_TXINTR;
} else {
sc->tulip_flags &= ~TULIP_PROMISC;
sc->tulip_cmdmode &= ~TULIP_CMD_PROMISCUOUS;
if (sc->tulip_flags & TULIP_ALLMULTI) {
sc->tulip_cmdmode |= TULIP_CMD_ALLMULTI;
} else {
sc->tulip_cmdmode &= ~TULIP_CMD_ALLMULTI;
}
}
sc->tulip_cmdmode |= TULIP_CMD_TXRUN;
if ((sc->tulip_flags & (TULIP_TXPROBE_ACTIVE|TULIP_WANTSETUP)) == 0) {
tulip_rx_intr(sc);
sc->tulip_cmdmode |= TULIP_CMD_RXRUN;
sc->tulip_intrmask |= TULIP_STS_RXSTOPPED;
} else {
sc->tulip_ifp->if_drv_flags |= IFF_DRV_OACTIVE;
sc->tulip_cmdmode &= ~TULIP_CMD_RXRUN;
sc->tulip_intrmask &= ~TULIP_STS_RXSTOPPED;
}
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
CTR2(KTR_TULIP, "tulip_init_locked: intr mask %08x cmdmode %08x",
sc->tulip_intrmask, sc->tulip_cmdmode);
TULIP_CSR_WRITE(sc, csr_intr, sc->tulip_intrmask);
TULIP_CSR_WRITE(sc, csr_command, sc->tulip_cmdmode);
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
CTR1(KTR_TULIP, "tulip_init_locked: status %08x\n",
TULIP_CSR_READ(sc, csr_status));
1997-08-03 13:00:42 +00:00
if ((sc->tulip_flags & (TULIP_WANTSETUP|TULIP_TXPROBE_ACTIVE)) == TULIP_WANTSETUP)
tulip_txput_setup(sc);
} else {
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
CTR0(KTR_TULIP, "tulip_init_locked: not up, reset chip");
sc->tulip_ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
tulip_reset(sc);
tulip_addr_filter(sc);
}
}
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
#define DESC_STATUS(di) (((volatile tulip_desc_t *)((di)->di_desc))->d_status)
#define DESC_FLAG(di) ((di)->di_desc->d_flag)
static void
tulip_rx_intr(tulip_softc_t * const sc)
{
1997-08-03 13:00:42 +00:00
TULIP_PERFSTART(rxintr)
tulip_ringinfo_t * const ri = &sc->tulip_rxinfo;
struct ifnet * const ifp = sc->tulip_ifp;
int fillok = 1;
#if defined(TULIP_DEBUG)
int cnt = 0;
#endif
TULIP_LOCK_ASSERT(sc);
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
CTR0(KTR_TULIP, "tulip_rx_intr: start");
for (;;) {
1997-08-03 13:00:42 +00:00
TULIP_PERFSTART(rxget)
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
tulip_descinfo_t *eop = ri->ri_nextin, *dip;
int total_len = 0, last_offset = 0;
struct mbuf *ms = NULL, *me = NULL;
int accept = 0;
int error;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
if (fillok && (ri->ri_max - ri->ri_free) < TULIP_RXQ_TARGET)
goto queue_mbuf;
#if defined(TULIP_DEBUG)
if (cnt == ri->ri_max)
break;
#endif
/*
* If the TULIP has no descriptors, there can't be any receive
* descriptors to process.
*/
if (eop == ri->ri_nextout)
break;
/*
* 90% of the packets will fit in one descriptor. So we optimize
* for that case.
*/
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
TULIP_RXDESC_POSTSYNC(ri);
if ((DESC_STATUS(eop) & (TULIP_DSTS_OWNER|TULIP_DSTS_RxFIRSTDESC|TULIP_DSTS_RxLASTDESC)) == (TULIP_DSTS_RxFIRSTDESC|TULIP_DSTS_RxLASTDESC)) {
ms = tulip_dequeue_mbuf(ri, eop, SYNC_RX);
CTR2(KTR_TULIP,
"tulip_rx_intr: single packet mbuf %p from descriptor %td", ms,
eop - ri->ri_first);
me = ms;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
ri->ri_free++;
} else {
/*
* If still owned by the TULIP, don't touch it.
*/
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
if (DESC_STATUS(eop) & TULIP_DSTS_OWNER)
break;
/*
* It is possible (though improbable unless MCLBYTES < 1518) for
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
* a received packet to cross more than one receive descriptor.
* We first loop through the descriptor ring making sure we have
* received a complete packet. If not, we bail until the next
* interrupt.
*/
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
dip = eop;
while ((DESC_STATUS(eop) & TULIP_DSTS_RxLASTDESC) == 0) {
if (++eop == ri->ri_last)
eop = ri->ri_first;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
TULIP_RXDESC_POSTSYNC(ri);
if (eop == ri->ri_nextout || DESC_STATUS(eop) & TULIP_DSTS_OWNER) {
#if defined(TULIP_DEBUG)
sc->tulip_dbg.dbg_rxintrs++;
sc->tulip_dbg.dbg_rxpktsperintr[cnt]++;
#endif
1997-08-03 13:00:42 +00:00
TULIP_PERFEND(rxget);
TULIP_PERFEND(rxintr);
return;
}
total_len++;
}
/*
* Dequeue the first buffer for the start of the packet. Hopefully
* this will be the only one we need to dequeue. However, if the
* packet consumed multiple descriptors, then we need to dequeue
* those buffers and chain to the starting mbuf. All buffers but
* the last buffer have the same length so we can set that now.
* (we add to last_offset instead of multiplying since we normally
* won't go into the loop and thereby saving ourselves from
* doing a multiplication by 0 in the normal case).
*/
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
ms = tulip_dequeue_mbuf(ri, dip, SYNC_RX);
CTR2(KTR_TULIP,
"tulip_rx_intr: start packet mbuf %p from descriptor %td", ms,
dip - ri->ri_first);
ri->ri_free++;
for (me = ms; total_len > 0; total_len--) {
me->m_len = TULIP_RX_BUFLEN;
last_offset += TULIP_RX_BUFLEN;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
if (++dip == ri->ri_last)
dip = ri->ri_first;
me->m_next = tulip_dequeue_mbuf(ri, dip, SYNC_RX);
ri->ri_free++;
me = me->m_next;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
CTR2(KTR_TULIP,
"tulip_rx_intr: cont packet mbuf %p from descriptor %td",
me, dip - ri->ri_first);
}
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
KASSERT(dip == eop, ("mismatched descinfo structs"));
}
/*
* Now get the size of received packet (minus the CRC).
*/
total_len = ((DESC_STATUS(eop) >> 16) & 0x7FFF) - ETHER_CRC_LEN;
if ((sc->tulip_flags & TULIP_RXIGNORE) == 0
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
&& ((DESC_STATUS(eop) & TULIP_DSTS_ERRSUM) == 0)) {
me->m_len = total_len - last_offset;
sc->tulip_flags |= TULIP_RXACT;
accept = 1;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
CTR1(KTR_TULIP, "tulip_rx_intr: good packet; length %d",
total_len);
} else {
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
CTR1(KTR_TULIP, "tulip_rx_intr: bad packet; status %08x",
DESC_STATUS(eop));
ifp->if_ierrors++;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
if (DESC_STATUS(eop) & (TULIP_DSTS_RxBADLENGTH|TULIP_DSTS_RxOVERFLOW|TULIP_DSTS_RxWATCHDOG)) {
sc->tulip_dot3stats.dot3StatsInternalMacReceiveErrors++;
} else {
#if defined(TULIP_VERBOSE)
const char *error = NULL;
#endif
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
if (DESC_STATUS(eop) & TULIP_DSTS_RxTOOLONG) {
sc->tulip_dot3stats.dot3StatsFrameTooLongs++;
#if defined(TULIP_VERBOSE)
error = "frame too long";
#endif
}
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
if (DESC_STATUS(eop) & TULIP_DSTS_RxBADCRC) {
if (DESC_STATUS(eop) & TULIP_DSTS_RxDRBBLBIT) {
sc->tulip_dot3stats.dot3StatsAlignmentErrors++;
#if defined(TULIP_VERBOSE)
error = "alignment error";
#endif
} else {
sc->tulip_dot3stats.dot3StatsFCSErrors++;
#if defined(TULIP_VERBOSE)
error = "bad crc";
#endif
}
}
#if defined(TULIP_VERBOSE)
if (error != NULL && (sc->tulip_flags & TULIP_NOMESSAGES) == 0) {
if_printf(sc->tulip_ifp, "receive: %6D: %s\n",
mtod(ms, u_char *) + 6, ":",
error);
sc->tulip_flags |= TULIP_NOMESSAGES;
}
#endif
}
}
#if defined(TULIP_DEBUG)
cnt++;
#endif
ifp->if_ipackets++;
if (++eop == ri->ri_last)
eop = ri->ri_first;
ri->ri_nextin = eop;
queue_mbuf:
/*
* We have received a good packet that needs to be passed up the
* stack.
*/
if (accept) {
struct mbuf *m0;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
KASSERT(ms != NULL, ("no packet to accept"));
#if defined(TULIP_COPY_RXDATA)
/*
* Copy the data into a new mbuf that is properly aligned. If
* we fail to allocate a new mbuf, then drop the packet. We will
* reuse the same rx buffer ('ms') below for another packet
* regardless.
*/
m0 = m_devget(mtod(ms, caddr_t), total_len, ETHER_ALIGN, ifp, NULL);
if (m0 == NULL) {
ifp->if_ierrors++;
goto skip_input;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
}
#else
/*
* Update the header for the mbuf referencing this receive
* buffer and pass it up the stack. Allocate a new mbuf cluster
* to replace the one we just passed up the stack.
*
* Note that if this packet crossed multiple descriptors
* we don't even try to reallocate all the mbufs here.
* Instead we rely on the test at the beginning of
* the loop to refill for the extra consumed mbufs.
*/
ms->m_pkthdr.len = total_len;
ms->m_pkthdr.rcvif = ifp;
m0 = ms;
ms = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
#endif
TULIP_UNLOCK(sc);
CTR1(KTR_TULIP, "tulip_rx_intr: passing %p to upper layer", m0);
(*ifp->if_input)(ifp, m0);
TULIP_LOCK(sc);
} else if (ms == NULL)
/*
* If we are priming the TULIP with mbufs, then allocate
* a new cluster for the next descriptor.
*/
ms = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
#if defined(TULIP_COPY_RXDATA)
skip_input:
#endif
if (ms == NULL) {
/*
* Couldn't allocate a new buffer. Don't bother
* trying to replenish the receive queue.
*/
fillok = 0;
sc->tulip_flags |= TULIP_RXBUFSLOW;
#if defined(TULIP_DEBUG)
sc->tulip_dbg.dbg_rxlowbufs++;
#endif
1997-08-03 13:00:42 +00:00
TULIP_PERFEND(rxget);
continue;
}
/*
* Now give the buffer(s) to the TULIP and save in our
* receive queue.
*/
do {
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
tulip_descinfo_t * const nextout = ri->ri_nextout;
M_ASSERTPKTHDR(ms);
KASSERT(ms->m_data == ms->m_ext.ext_buf,
("rx mbuf data doesn't point to cluster"));
ms->m_len = ms->m_pkthdr.len = MCLBYTES;
error = bus_dmamap_load_mbuf(ri->ri_data_tag, *nextout->di_map, ms,
tulip_dma_map_rxbuf, nextout->di_desc, BUS_DMA_NOWAIT);
if (error) {
if_printf(sc->tulip_ifp,
"unable to load rx map, error = %d\n", error);
panic("tulip_rx_intr"); /* XXX */
}
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
nextout->di_desc->d_status = TULIP_DSTS_OWNER;
KASSERT(nextout->di_mbuf == NULL, ("clobbering earlier rx mbuf"));
nextout->di_mbuf = ms;
CTR2(KTR_TULIP, "tulip_rx_intr: enqueued mbuf %p to descriptor %td",
ms, nextout - ri->ri_first);
TULIP_RXDESC_POSTSYNC(ri);
if (++ri->ri_nextout == ri->ri_last)
ri->ri_nextout = ri->ri_first;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
ri->ri_free--;
me = ms->m_next;
ms->m_next = NULL;
} while ((ms = me) != NULL);
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
if ((ri->ri_max - ri->ri_free) >= TULIP_RXQ_TARGET)
sc->tulip_flags &= ~TULIP_RXBUFSLOW;
1997-08-03 13:00:42 +00:00
TULIP_PERFEND(rxget);
}
#if defined(TULIP_DEBUG)
sc->tulip_dbg.dbg_rxintrs++;
sc->tulip_dbg.dbg_rxpktsperintr[cnt]++;
#endif
1997-08-03 13:00:42 +00:00
TULIP_PERFEND(rxintr);
}
static int
tulip_tx_intr(tulip_softc_t * const sc)
{
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
TULIP_PERFSTART(txintr)
tulip_ringinfo_t * const ri = &sc->tulip_txinfo;
struct mbuf *m;
int xmits = 0;
1997-08-03 13:00:42 +00:00
int descs = 0;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
CTR0(KTR_TULIP, "tulip_tx_intr: start");
TULIP_LOCK_ASSERT(sc);
while (ri->ri_free < ri->ri_max) {
1997-08-03 13:00:42 +00:00
u_int32_t d_flag;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
TULIP_TXDESC_POSTSYNC(ri);
if (DESC_STATUS(ri->ri_nextin) & TULIP_DSTS_OWNER)
break;
ri->ri_free++;
descs++;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
d_flag = DESC_FLAG(ri->ri_nextin);
1997-08-03 13:00:42 +00:00
if (d_flag & TULIP_DFLAG_TxLASTSEG) {
if (d_flag & TULIP_DFLAG_TxSETUPPKT) {
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
CTR2(KTR_TULIP,
"tulip_tx_intr: setup packet from descriptor %td: %08x",
ri->ri_nextin - ri->ri_first, DESC_STATUS(ri->ri_nextin));
/*
* We've just finished processing a setup packet.
* Mark that we finished it. If there's not
* another pending, startup the TULIP receiver.
* Make sure we ack the RXSTOPPED so we won't get
* an abormal interrupt indication.
*/
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
bus_dmamap_sync(sc->tulip_setup_tag, sc->tulip_setup_map,
BUS_DMASYNC_POSTWRITE);
sc->tulip_flags &= ~(TULIP_DOINGSETUP|TULIP_HASHONLY);
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
if (DESC_FLAG(ri->ri_nextin) & TULIP_DFLAG_TxINVRSFILT)
sc->tulip_flags |= TULIP_HASHONLY;
if ((sc->tulip_flags & (TULIP_WANTSETUP|TULIP_TXPROBE_ACTIVE)) == 0) {
tulip_rx_intr(sc);
sc->tulip_cmdmode |= TULIP_CMD_RXRUN;
sc->tulip_intrmask |= TULIP_STS_RXSTOPPED;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
CTR2(KTR_TULIP,
"tulip_tx_intr: intr mask %08x cmdmode %08x",
sc->tulip_intrmask, sc->tulip_cmdmode);
TULIP_CSR_WRITE(sc, csr_status, TULIP_STS_RXSTOPPED);
TULIP_CSR_WRITE(sc, csr_intr, sc->tulip_intrmask);
TULIP_CSR_WRITE(sc, csr_command, sc->tulip_cmdmode);
}
} else {
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
const u_int32_t d_status = DESC_STATUS(ri->ri_nextin);
m = tulip_dequeue_mbuf(ri, ri->ri_nextin, SYNC_TX);
CTR2(KTR_TULIP,
"tulip_tx_intr: data packet %p from descriptor %td", m,
ri->ri_nextin - ri->ri_first);
if (m != NULL) {
m_freem(m);
#if defined(TULIP_DEBUG)
} else {
if_printf(sc->tulip_ifp,
"tx_intr: failed to dequeue mbuf?!?\n");
#endif
}
if (sc->tulip_flags & TULIP_TXPROBE_ACTIVE) {
tulip_mediapoll_event_t event = TULIP_MEDIAPOLL_TXPROBE_OK;
1997-08-03 13:00:42 +00:00
if (d_status & (TULIP_DSTS_TxNOCARR|TULIP_DSTS_TxEXCCOLL)) {
#if defined(TULIP_DEBUG)
1997-08-03 13:00:42 +00:00
if (d_status & TULIP_DSTS_TxNOCARR)
sc->tulip_dbg.dbg_txprobe_nocarr++;
1997-08-03 13:00:42 +00:00
if (d_status & TULIP_DSTS_TxEXCCOLL)
sc->tulip_dbg.dbg_txprobe_exccoll++;
#endif
event = TULIP_MEDIAPOLL_TXPROBE_FAILED;
}
(*sc->tulip_boardsw->bd_media_poll)(sc, event);
/*
* Escape from the loop before media poll has reset the TULIP!
*/
break;
} else {
xmits++;
1997-08-03 13:00:42 +00:00
if (d_status & TULIP_DSTS_ERRSUM) {
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
CTR1(KTR_TULIP, "tulip_tx_intr: output error: %08x",
d_status);
sc->tulip_ifp->if_oerrors++;
1997-08-03 13:00:42 +00:00
if (d_status & TULIP_DSTS_TxEXCCOLL)
sc->tulip_dot3stats.dot3StatsExcessiveCollisions++;
1997-08-03 13:00:42 +00:00
if (d_status & TULIP_DSTS_TxLATECOLL)
sc->tulip_dot3stats.dot3StatsLateCollisions++;
1997-08-03 13:00:42 +00:00
if (d_status & (TULIP_DSTS_TxNOCARR|TULIP_DSTS_TxCARRLOSS))
sc->tulip_dot3stats.dot3StatsCarrierSenseErrors++;
1997-08-03 13:00:42 +00:00
if (d_status & (TULIP_DSTS_TxUNDERFLOW|TULIP_DSTS_TxBABBLE))
sc->tulip_dot3stats.dot3StatsInternalMacTransmitErrors++;
1997-08-03 13:00:42 +00:00
if (d_status & TULIP_DSTS_TxUNDERFLOW)
sc->tulip_dot3stats.dot3StatsInternalTransmitUnderflows++;
if (d_status & TULIP_DSTS_TxBABBLE)
sc->tulip_dot3stats.dot3StatsInternalTransmitBabbles++;
} else {
u_int32_t collisions =
1997-08-03 13:00:42 +00:00
(d_status & TULIP_DSTS_TxCOLLMASK)
>> TULIP_DSTS_V_TxCOLLCNT;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
CTR2(KTR_TULIP,
"tulip_tx_intr: output ok, collisions %d, status %08x",
collisions, d_status);
sc->tulip_ifp->if_collisions += collisions;
if (collisions == 1)
sc->tulip_dot3stats.dot3StatsSingleCollisionFrames++;
else if (collisions > 1)
sc->tulip_dot3stats.dot3StatsMultipleCollisionFrames++;
1997-08-03 13:00:42 +00:00
else if (d_status & TULIP_DSTS_TxDEFERRED)
sc->tulip_dot3stats.dot3StatsDeferredTransmissions++;
/*
* SQE is only valid for 10baseT/BNC/AUI when not
* running in full-duplex. In order to speed up the
* test, the corresponding bit in tulip_flags needs to
* set as well to get us to count SQE Test Errors.
*/
1997-08-03 13:00:42 +00:00
if (d_status & TULIP_DSTS_TxNOHRTBT & sc->tulip_flags)
sc->tulip_dot3stats.dot3StatsSQETestErrors++;
}
}
}
}
if (++ri->ri_nextin == ri->ri_last)
ri->ri_nextin = ri->ri_first;
if ((sc->tulip_flags & TULIP_TXPROBE_ACTIVE) == 0)
sc->tulip_ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
}
/*
* If nothing left to transmit, disable the timer.
* Else if progress, reset the timer back to 2 ticks.
*/
if (ri->ri_free == ri->ri_max || (sc->tulip_flags & TULIP_TXPROBE_ACTIVE))
sc->tulip_txtimer = 0;
else if (xmits > 0)
sc->tulip_txtimer = TULIP_TXTIMER;
sc->tulip_ifp->if_opackets += xmits;
1997-08-03 13:00:42 +00:00
TULIP_PERFEND(txintr);
return descs;
}
static void
tulip_print_abnormal_interrupt(tulip_softc_t * const sc, u_int32_t csr)
{
const char * const *msgp = tulip_status_bits;
const char *sep;
1997-08-03 13:00:42 +00:00
u_int32_t mask;
const char thrsh[] = "72|128\0\0\0" "96|256\0\0\0" "128|512\0\0" "160|1024";
TULIP_LOCK_ASSERT(sc);
csr &= (1 << (sizeof(tulip_status_bits)/sizeof(tulip_status_bits[0]))) - 1;
if_printf(sc->tulip_ifp, "abnormal interrupt:");
1997-08-03 13:00:42 +00:00
for (sep = " ", mask = 1; mask <= csr; mask <<= 1, msgp++) {
if ((csr & mask) && *msgp != NULL) {
printf("%s%s", sep, *msgp);
1997-08-03 13:00:42 +00:00
if (mask == TULIP_STS_TXUNDERFLOW && (sc->tulip_flags & TULIP_NEWTXTHRESH)) {
sc->tulip_flags &= ~TULIP_NEWTXTHRESH;
if (sc->tulip_cmdmode & TULIP_CMD_STOREFWD) {
printf(" (switching to store-and-forward mode)");
} else {
printf(" (raising TX threshold to %s)",
&thrsh[9 * ((sc->tulip_cmdmode & TULIP_CMD_THRESHOLDCTL) >> 14)]);
}
}
sep = ", ";
}
}
printf("\n");
}
static void
tulip_intr_handler(tulip_softc_t * const sc)
{
1997-08-03 13:00:42 +00:00
TULIP_PERFSTART(intr)
u_int32_t csr;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
CTR0(KTR_TULIP, "tulip_intr_handler invoked");
TULIP_LOCK_ASSERT(sc);
while ((csr = TULIP_CSR_READ(sc, csr_status)) & sc->tulip_intrmask) {
TULIP_CSR_WRITE(sc, csr_status, csr);
if (csr & TULIP_STS_SYSERROR) {
sc->tulip_last_system_error = (csr & TULIP_STS_ERRORMASK) >> TULIP_STS_ERR_SHIFT;
if (sc->tulip_flags & TULIP_NOMESSAGES) {
sc->tulip_flags |= TULIP_SYSTEMERROR;
} else {
if_printf(sc->tulip_ifp, "system error: %s\n",
tulip_system_errors[sc->tulip_last_system_error]);
}
sc->tulip_flags |= TULIP_NEEDRESET;
sc->tulip_system_errors++;
break;
}
if (csr & (TULIP_STS_LINKPASS|TULIP_STS_LINKFAIL) & sc->tulip_intrmask) {
#if defined(TULIP_DEBUG)
sc->tulip_dbg.dbg_link_intrs++;
#endif
if (sc->tulip_boardsw->bd_media_poll != NULL) {
(*sc->tulip_boardsw->bd_media_poll)(sc, csr & TULIP_STS_LINKFAIL
? TULIP_MEDIAPOLL_LINKFAIL
: TULIP_MEDIAPOLL_LINKPASS);
csr &= ~TULIP_STS_ABNRMLINTR;
}
tulip_media_print(sc);
}
if (csr & (TULIP_STS_RXINTR|TULIP_STS_RXNOBUF)) {
u_int32_t misses = TULIP_CSR_READ(sc, csr_missed_frames);
if (csr & TULIP_STS_RXNOBUF)
sc->tulip_dot3stats.dot3StatsMissedFrames += misses & 0xFFFF;
/*
* Pass 2.[012] of the 21140A-A[CDE] may hang and/or corrupt data
* on receive overflows.
*/
if ((misses & 0x0FFE0000) && (sc->tulip_features & TULIP_HAVE_RXBADOVRFLW)) {
sc->tulip_dot3stats.dot3StatsInternalMacReceiveErrors++;
/*
* Stop the receiver process and spin until it's stopped.
* Tell rx_intr to drop the packets it dequeues.
*/
TULIP_CSR_WRITE(sc, csr_command, sc->tulip_cmdmode & ~TULIP_CMD_RXRUN);
while ((TULIP_CSR_READ(sc, csr_status) & TULIP_STS_RXSTOPPED) == 0)
;
TULIP_CSR_WRITE(sc, csr_status, TULIP_STS_RXSTOPPED);
sc->tulip_flags |= TULIP_RXIGNORE;
}
tulip_rx_intr(sc);
if (sc->tulip_flags & TULIP_RXIGNORE) {
/*
* Restart the receiver.
*/
sc->tulip_flags &= ~TULIP_RXIGNORE;
TULIP_CSR_WRITE(sc, csr_command, sc->tulip_cmdmode);
}
}
if (csr & TULIP_STS_ABNRMLINTR) {
u_int32_t tmp = csr & sc->tulip_intrmask
& ~(TULIP_STS_NORMALINTR|TULIP_STS_ABNRMLINTR);
1997-08-03 13:00:42 +00:00
if (csr & TULIP_STS_TXUNDERFLOW) {
if ((sc->tulip_cmdmode & TULIP_CMD_THRESHOLDCTL) != TULIP_CMD_THRSHLD160) {
sc->tulip_cmdmode += TULIP_CMD_THRSHLD96;
sc->tulip_flags |= TULIP_NEWTXTHRESH;
} else if (sc->tulip_features & TULIP_HAVE_STOREFWD) {
sc->tulip_cmdmode |= TULIP_CMD_STOREFWD;
sc->tulip_flags |= TULIP_NEWTXTHRESH;
}
}
if (sc->tulip_flags & TULIP_NOMESSAGES) {
sc->tulip_statusbits |= tmp;
} else {
tulip_print_abnormal_interrupt(sc, tmp);
sc->tulip_flags |= TULIP_NOMESSAGES;
}
TULIP_CSR_WRITE(sc, csr_command, sc->tulip_cmdmode);
}
1997-08-03 13:00:42 +00:00
if (sc->tulip_flags & (TULIP_WANTTXSTART|TULIP_TXPROBE_ACTIVE|TULIP_DOINGSETUP|TULIP_PROMISC)) {
tulip_tx_intr(sc);
if ((sc->tulip_flags & TULIP_TXPROBE_ACTIVE) == 0)
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
tulip_start_locked(sc);
}
}
if (sc->tulip_flags & TULIP_NEEDRESET) {
tulip_reset(sc);
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
tulip_init_locked(sc);
}
1997-08-03 13:00:42 +00:00
TULIP_PERFEND(intr);
}
static void
tulip_intr_shared(void *arg)
{
tulip_softc_t * sc = arg;
for (; sc != NULL; sc = sc->tulip_slaves) {
TULIP_LOCK(sc);
#if defined(TULIP_DEBUG)
sc->tulip_dbg.dbg_intrs++;
#endif
tulip_intr_handler(sc);
TULIP_UNLOCK(sc);
}
}
static void
tulip_intr_normal(void *arg)
{
tulip_softc_t * sc = (tulip_softc_t *) arg;
TULIP_LOCK(sc);
#if defined(TULIP_DEBUG)
sc->tulip_dbg.dbg_intrs++;
#endif
tulip_intr_handler(sc);
TULIP_UNLOCK(sc);
}
1997-08-03 13:00:42 +00:00
static struct mbuf *
tulip_txput(tulip_softc_t * const sc, struct mbuf *m)
1997-08-03 13:00:42 +00:00
{
TULIP_PERFSTART(txput)
tulip_ringinfo_t * const ri = &sc->tulip_txinfo;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
tulip_descinfo_t *eop, *nextout;
1997-08-03 13:00:42 +00:00
int segcnt, free;
u_int32_t d_status;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
bus_dma_segment_t segs[TULIP_MAX_TXSEG];
bus_dmamap_t *map;
int error, nsegs;
struct mbuf *m0;
1997-08-03 13:00:42 +00:00
TULIP_LOCK_ASSERT(sc);
1997-08-03 13:00:42 +00:00
#if defined(TULIP_DEBUG)
if ((sc->tulip_cmdmode & TULIP_CMD_TXRUN) == 0) {
if_printf(sc->tulip_ifp, "txput%s: tx not running\n",
1997-08-03 13:00:42 +00:00
(sc->tulip_flags & TULIP_TXPROBE_ACTIVE) ? "(probe)" : "");
sc->tulip_flags |= TULIP_WANTTXSTART;
sc->tulip_dbg.dbg_txput_finishes[0]++;
1997-08-03 13:00:42 +00:00
goto finish;
}
#endif
/*
* Now we try to fill in our transmit descriptors. This is
* a bit reminiscent of going on the Ark two by two
* since each descriptor for the TULIP can describe
* two buffers. So we advance through packet filling
* each of the two entries at a time to to fill each
* descriptor. Clear the first and last segment bits
* in each descriptor (actually just clear everything
* but the end-of-ring or chain bits) to make sure
* we don't get messed up by previously sent packets.
*
* We may fail to put the entire packet on the ring if
* there is either not enough ring entries free or if the
* packet has more than MAX_TXSEG segments. In the former
* case we will just wait for the ring to empty. In the
* latter case we have to recopy.
*/
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
#if defined(KTR) && KTR_TULIP
segcnt = 1;
m0 = m;
while (m0->m_next != NULL) {
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
segcnt++;
m0 = m0->m_next;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
}
#endif
CTR2(KTR_TULIP, "tulip_txput: sending packet %p (%d chunks)", m, segcnt);
1997-08-03 13:00:42 +00:00
d_status = 0;
eop = nextout = ri->ri_nextout;
segcnt = 0;
free = ri->ri_free;
/*
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
* Reclaim some tx descriptors if we are out since we need at least one
* free descriptor so that we have a dma_map to load the mbuf.
*/
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
if (free == 0) {
#if defined(TULIP_DEBUG)
sc->tulip_dbg.dbg_no_txmaps++;
#endif
free += tulip_tx_intr(sc);
}
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
if (free == 0) {
sc->tulip_flags |= TULIP_WANTTXSTART;
#if defined(TULIP_DEBUG)
sc->tulip_dbg.dbg_txput_finishes[1]++;
#endif
goto finish;
}
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
error = bus_dmamap_load_mbuf_sg(ri->ri_data_tag, *eop->di_map, m, segs,
&nsegs, BUS_DMA_NOWAIT);
if (error != 0) {
if (error == EFBIG) {
/*
* The packet exceeds the number of transmit buffer
* entries that we can use for one packet, so we have
* to recopy it into one mbuf and then try again. If
* we can't recopy it, try again later.
*/
m0 = m_defrag(m, M_DONTWAIT);
if (m0 == NULL) {
sc->tulip_flags |= TULIP_WANTTXSTART;
#if defined(TULIP_DEBUG)
sc->tulip_dbg.dbg_txput_finishes[2]++;
#endif
goto finish;
}
m = m0;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
error = bus_dmamap_load_mbuf_sg(ri->ri_data_tag, *eop->di_map, m,
segs, &nsegs, BUS_DMA_NOWAIT);
}
if (error != 0) {
if_printf(sc->tulip_ifp,
"unable to load tx map, error = %d\n", error);
#if defined(TULIP_DEBUG)
sc->tulip_dbg.dbg_txput_finishes[3]++;
#endif
goto finish;
}
}
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
CTR1(KTR_TULIP, "tulip_txput: nsegs %d", nsegs);
/*
* Each descriptor allows for up to 2 fragments since we don't use
* the descriptor chaining mode in this driver.
*/
if ((free -= (nsegs + 1) / 2) <= 0
/*
* See if there's any unclaimed space in the transmit ring.
*/
&& (free += tulip_tx_intr(sc)) <= 0) {
/*
* There's no more room but since nothing
* has been committed at this point, just
* show output is active, put back the
* mbuf and return.
*/
sc->tulip_flags |= TULIP_WANTTXSTART;
#if defined(TULIP_DEBUG)
sc->tulip_dbg.dbg_txput_finishes[4]++;
#endif
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
bus_dmamap_unload(ri->ri_data_tag, *eop->di_map);
goto finish;
}
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
for (; nsegs - segcnt > 1; segcnt += 2) {
eop = nextout;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
eop->di_desc->d_flag &= TULIP_DFLAG_ENDRING|TULIP_DFLAG_CHAIN;
eop->di_desc->d_status = d_status;
eop->di_desc->d_addr1 = segs[segcnt].ds_addr;
eop->di_desc->d_length1 = segs[segcnt].ds_len;
eop->di_desc->d_addr2 = segs[segcnt+1].ds_addr;
eop->di_desc->d_length2 = segs[segcnt+1].ds_len;
d_status = TULIP_DSTS_OWNER;
if (++nextout == ri->ri_last)
nextout = ri->ri_first;
}
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
if (segcnt < nsegs) {
eop = nextout;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
eop->di_desc->d_flag &= TULIP_DFLAG_ENDRING|TULIP_DFLAG_CHAIN;
eop->di_desc->d_status = d_status;
eop->di_desc->d_addr1 = segs[segcnt].ds_addr;
eop->di_desc->d_length1 = segs[segcnt].ds_len;
eop->di_desc->d_addr2 = 0;
eop->di_desc->d_length2 = 0;
if (++nextout == ri->ri_last)
nextout = ri->ri_first;
}
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
/*
* tulip_tx_intr() harvests the mbuf from the last descriptor in the
* frame. We just used the dmamap in the first descriptor for the
* load operation however. Thus, to let the tulip_dequeue_mbuf() call
* in tulip_tx_intr() unload the correct dmamap, we swap the dmamap
* pointers in the two descriptors if this is a multiple-descriptor
* packet.
*/
if (eop != ri->ri_nextout) {
map = eop->di_map;
eop->di_map = ri->ri_nextout->di_map;
ri->ri_nextout->di_map = map;
}
1997-08-03 13:00:42 +00:00
/*
* bounce a copy to the bpf listener, if any.
*/
BPF_MTAP(sc->tulip_ifp, m);
1997-08-03 13:00:42 +00:00
/*
* The descriptors have been filled in. Now get ready
* to transmit.
*/
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
CTR3(KTR_TULIP, "tulip_txput: enqueued mbuf %p to descriptors %td - %td",
m, ri->ri_nextout - ri->ri_first, eop - ri->ri_first);
KASSERT(eop->di_mbuf == NULL, ("clobbering earlier tx mbuf"));
eop->di_mbuf = m;
TULIP_TXMAP_PRESYNC(ri, ri->ri_nextout);
1997-08-03 13:00:42 +00:00
m = NULL;
/*
* Make sure the next descriptor after this packet is owned
* by us since it may have been set up above if we ran out
* of room in the ring.
*/
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
nextout->di_desc->d_status = 0;
TULIP_TXDESC_PRESYNC(ri);
1997-08-03 13:00:42 +00:00
/*
* Mark the last and first segments, indicate we want a transmit
* complete interrupt, and tell it to transmit!
*/
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
eop->di_desc->d_flag |= TULIP_DFLAG_TxLASTSEG|TULIP_DFLAG_TxWANTINTR;
1997-08-03 13:00:42 +00:00
/*
* Note that ri->ri_nextout is still the start of the packet
* and until we set the OWNER bit, we can still back out of
* everything we have done.
*/
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
ri->ri_nextout->di_desc->d_flag |= TULIP_DFLAG_TxFIRSTSEG;
TULIP_TXDESC_PRESYNC(ri);
ri->ri_nextout->di_desc->d_status = TULIP_DSTS_OWNER;
TULIP_TXDESC_PRESYNC(ri);
1997-08-03 13:00:42 +00:00
/*
* This advances the ring for us.
*/
ri->ri_nextout = nextout;
ri->ri_free = free;
TULIP_PERFEND(txput);
if (sc->tulip_flags & TULIP_TXPROBE_ACTIVE) {
TULIP_CSR_WRITE(sc, csr_txpoll, 1);
sc->tulip_ifp->if_drv_flags |= IFF_DRV_OACTIVE;
1997-08-03 13:00:42 +00:00
TULIP_PERFEND(txput);
return NULL;
}
/*
* switch back to the single queueing ifstart.
*/
sc->tulip_flags &= ~TULIP_WANTTXSTART;
if (sc->tulip_txtimer == 0)
sc->tulip_txtimer = TULIP_TXTIMER;
#if defined(TULIP_DEBUG)
sc->tulip_dbg.dbg_txput_finishes[5]++;
#endif
1997-08-03 13:00:42 +00:00
/*
* If we want a txstart, there must be not enough space in the
* transmit ring. So we want to enable transmit done interrupts
* so we can immediately reclaim some space. When the transmit
* interrupt is posted, the interrupt handler will call tx_intr
* to reclaim space and then txstart (since WANTTXSTART is set).
* txstart will move the packet into the transmit ring and clear
* WANTTXSTART thereby causing TXINTR to be cleared.
*/
finish:
#if defined(TULIP_DEBUG)
sc->tulip_dbg.dbg_txput_finishes[6]++;
#endif
1997-08-03 13:00:42 +00:00
if (sc->tulip_flags & (TULIP_WANTTXSTART|TULIP_DOINGSETUP)) {
sc->tulip_ifp->if_drv_flags |= IFF_DRV_OACTIVE;
1997-08-03 13:00:42 +00:00
if ((sc->tulip_intrmask & TULIP_STS_TXINTR) == 0) {
sc->tulip_intrmask |= TULIP_STS_TXINTR;
TULIP_CSR_WRITE(sc, csr_intr, sc->tulip_intrmask);
}
} else if ((sc->tulip_flags & TULIP_PROMISC) == 0) {
if (sc->tulip_intrmask & TULIP_STS_TXINTR) {
sc->tulip_intrmask &= ~TULIP_STS_TXINTR;
TULIP_CSR_WRITE(sc, csr_intr, sc->tulip_intrmask);
}
}
TULIP_CSR_WRITE(sc, csr_txpoll, 1);
1997-08-03 13:00:42 +00:00
TULIP_PERFEND(txput);
return m;
}
1997-08-03 13:00:42 +00:00
static void
tulip_txput_setup(tulip_softc_t * const sc)
1997-08-03 13:00:42 +00:00
{
tulip_ringinfo_t * const ri = &sc->tulip_txinfo;
tulip_desc_t *nextout;
TULIP_LOCK_ASSERT(sc);
1997-08-03 13:00:42 +00:00
/*
* We will transmit, at most, one setup packet per call to ifstart.
*/
#if defined(TULIP_DEBUG)
if ((sc->tulip_cmdmode & TULIP_CMD_TXRUN) == 0) {
if_printf(sc->tulip_ifp, "txput_setup: tx not running\n");
1997-08-03 13:00:42 +00:00
sc->tulip_flags |= TULIP_WANTTXSTART;
return;
}
#endif
/*
* Try to reclaim some free descriptors..
*/
if (ri->ri_free < 2)
tulip_tx_intr(sc);
if ((sc->tulip_flags & TULIP_DOINGSETUP) || ri->ri_free == 1) {
sc->tulip_flags |= TULIP_WANTTXSTART;
return;
}
bcopy(sc->tulip_setupdata, sc->tulip_setupbuf,
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
sizeof(sc->tulip_setupdata));
1997-08-03 13:00:42 +00:00
/*
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
* Clear WANTSETUP and set DOINGSETUP. Since we know that WANTSETUP is
1997-08-03 13:00:42 +00:00
* set and DOINGSETUP is clear doing an XOR of the two will DTRT.
*/
sc->tulip_flags ^= TULIP_WANTSETUP|TULIP_DOINGSETUP;
ri->ri_free--;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
nextout = ri->ri_nextout->di_desc;
1997-08-03 13:00:42 +00:00
nextout->d_flag &= TULIP_DFLAG_ENDRING|TULIP_DFLAG_CHAIN;
nextout->d_flag |= TULIP_DFLAG_TxFIRSTSEG|TULIP_DFLAG_TxLASTSEG
|TULIP_DFLAG_TxSETUPPKT|TULIP_DFLAG_TxWANTINTR;
if (sc->tulip_flags & TULIP_WANTHASHPERFECT)
nextout->d_flag |= TULIP_DFLAG_TxHASHFILT;
else if (sc->tulip_flags & TULIP_WANTHASHONLY)
nextout->d_flag |= TULIP_DFLAG_TxHASHFILT|TULIP_DFLAG_TxINVRSFILT;
nextout->d_length2 = 0;
nextout->d_addr2 = 0;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
nextout->d_length1 = sizeof(sc->tulip_setupdata);
nextout->d_addr1 = sc->tulip_setup_dma_addr;
bus_dmamap_sync(sc->tulip_setup_tag, sc->tulip_setup_map,
BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
TULIP_TXDESC_PRESYNC(ri);
CTR1(KTR_TULIP, "tulip_txput_setup: using descriptor %td",
ri->ri_nextout - ri->ri_first);
1997-08-03 13:00:42 +00:00
/*
* Advance the ring for the next transmit packet.
*/
if (++ri->ri_nextout == ri->ri_last)
ri->ri_nextout = ri->ri_first;
/*
* Make sure the next descriptor is owned by us since it
* may have been set up above if we ran out of room in the
* ring.
*/
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
ri->ri_nextout->di_desc->d_status = 0;
TULIP_TXDESC_PRESYNC(ri);
1997-08-03 13:00:42 +00:00
nextout->d_status = TULIP_DSTS_OWNER;
/*
* Flush the ownwership of the current descriptor
*/
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
TULIP_TXDESC_PRESYNC(ri);
1997-08-03 13:00:42 +00:00
TULIP_CSR_WRITE(sc, csr_txpoll, 1);
if ((sc->tulip_intrmask & TULIP_STS_TXINTR) == 0) {
sc->tulip_intrmask |= TULIP_STS_TXINTR;
TULIP_CSR_WRITE(sc, csr_intr, sc->tulip_intrmask);
}
}
static int
tulip_ifioctl(struct ifnet * ifp, u_long cmd, caddr_t data)
{
1997-08-03 13:00:42 +00:00
TULIP_PERFSTART(ifioctl)
tulip_softc_t * const sc = (tulip_softc_t *)ifp->if_softc;
struct ifreq *ifr = (struct ifreq *) data;
int error = 0;
switch (cmd) {
case SIOCSIFFLAGS: {
TULIP_LOCK(sc);
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
tulip_init_locked(sc);
TULIP_UNLOCK(sc);
break;
}
case SIOCSIFMEDIA:
case SIOCGIFMEDIA: {
error = ifmedia_ioctl(ifp, ifr, &sc->tulip_ifmedia, cmd);
break;
}
case SIOCADDMULTI:
case SIOCDELMULTI: {
/*
* Update multicast listeners
*/
TULIP_LOCK(sc);
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
tulip_init_locked(sc);
TULIP_UNLOCK(sc);
error = 0;
break;
}
#ifdef SIOCGADDRROM
case SIOCGADDRROM: {
error = copyout(sc->tulip_rombuf, ifr->ifr_data, sizeof(sc->tulip_rombuf));
break;
}
#endif
#ifdef SIOCGCHIPID
case SIOCGCHIPID: {
ifr->ifr_metric = (int) sc->tulip_chipid;
break;
}
#endif
default: {
error = ether_ioctl(ifp, cmd, data);
break;
}
}
1997-08-03 13:00:42 +00:00
TULIP_PERFEND(ifioctl);
return error;
}
1999-08-09 14:43:39 +00:00
static void
tulip_start(struct ifnet * const ifp)
{
1997-08-03 13:00:42 +00:00
TULIP_PERFSTART(ifstart)
tulip_softc_t * const sc = (tulip_softc_t *)ifp->if_softc;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
TULIP_LOCK(sc);
tulip_start_locked(sc);
TULIP_UNLOCK(sc);
TULIP_PERFEND(ifstart);
}
static void
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
tulip_start_locked(tulip_softc_t * const sc)
{
struct mbuf *m;
TULIP_LOCK_ASSERT(sc);
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
CTR0(KTR_TULIP, "tulip_start_locked invoked");
if ((sc->tulip_flags & (TULIP_WANTSETUP|TULIP_TXPROBE_ACTIVE)) == TULIP_WANTSETUP)
tulip_txput_setup(sc);
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
CTR1(KTR_TULIP, "tulip_start_locked: %d tx packets pending",
sc->tulip_ifp->if_snd.ifq_len);
while (!IFQ_DRV_IS_EMPTY(&sc->tulip_ifp->if_snd)) {
IFQ_DRV_DEQUEUE(&sc->tulip_ifp->if_snd, m);
if(m == NULL)
break;
if ((m = tulip_txput(sc, m)) != NULL) {
IFQ_DRV_PREPEND(&sc->tulip_ifp->if_snd, m);
break;
}
1997-08-03 13:00:42 +00:00
}
}
/*
* Even though this routine runs at device spl, it does not break
* our use of splnet (splsoftnet under NetBSD) for the majority
* of this driver since
* if_watcbog is called from if_watchdog which is called from
* splsoftclock which is below spl[soft]net.
*/
static void
tulip_ifwatchdog(struct ifnet *ifp)
{
1997-08-03 13:00:42 +00:00
TULIP_PERFSTART(ifwatchdog)
tulip_softc_t * const sc = (tulip_softc_t *)ifp->if_softc;
#if defined(TULIP_DEBUG)
u_int32_t rxintrs;
#endif
TULIP_LOCK(sc);
#if defined(TULIP_DEBUG)
rxintrs = sc->tulip_dbg.dbg_rxintrs - sc->tulip_dbg.dbg_last_rxintrs;
if (rxintrs > sc->tulip_dbg.dbg_high_rxintrs_hz)
sc->tulip_dbg.dbg_high_rxintrs_hz = rxintrs;
sc->tulip_dbg.dbg_last_rxintrs = sc->tulip_dbg.dbg_rxintrs;
#endif /* TULIP_DEBUG */
sc->tulip_ifp->if_timer = 1;
/*
* These should be rare so do a bulk test up front so we can just skip
* them if needed.
*/
if (sc->tulip_flags & (TULIP_SYSTEMERROR|TULIP_RXBUFSLOW|TULIP_NOMESSAGES)) {
/*
* If the number of receive buffer is low, try to refill
*/
if (sc->tulip_flags & TULIP_RXBUFSLOW)
tulip_rx_intr(sc);
if (sc->tulip_flags & TULIP_SYSTEMERROR) {
if_printf(sc->tulip_ifp, "%d system errors: last was %s\n",
sc->tulip_system_errors,
tulip_system_errors[sc->tulip_last_system_error]);
}
if (sc->tulip_statusbits) {
tulip_print_abnormal_interrupt(sc, sc->tulip_statusbits);
sc->tulip_statusbits = 0;
}
sc->tulip_flags &= ~(TULIP_NOMESSAGES|TULIP_SYSTEMERROR);
}
1997-08-03 13:00:42 +00:00
if (sc->tulip_txtimer)
tulip_tx_intr(sc);
if (sc->tulip_txtimer && --sc->tulip_txtimer == 0) {
if_printf(sc->tulip_ifp, "transmission timeout\n");
if (TULIP_DO_AUTOSENSE(sc)) {
sc->tulip_media = TULIP_MEDIA_UNKNOWN;
sc->tulip_probe_state = TULIP_PROBE_INACTIVE;
sc->tulip_flags &= ~(TULIP_WANTRXACT|TULIP_LINKUP);
}
tulip_reset(sc);
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
tulip_init_locked(sc);
}
1997-08-03 13:00:42 +00:00
TULIP_PERFEND(ifwatchdog);
TULIP_PERFMERGE(sc, perf_intr_cycles);
TULIP_PERFMERGE(sc, perf_ifstart_cycles);
TULIP_PERFMERGE(sc, perf_ifioctl_cycles);
TULIP_PERFMERGE(sc, perf_ifwatchdog_cycles);
TULIP_PERFMERGE(sc, perf_timeout_cycles);
TULIP_PERFMERGE(sc, perf_ifstart_one_cycles);
TULIP_PERFMERGE(sc, perf_txput_cycles);
TULIP_PERFMERGE(sc, perf_txintr_cycles);
TULIP_PERFMERGE(sc, perf_rxintr_cycles);
TULIP_PERFMERGE(sc, perf_rxget_cycles);
TULIP_PERFMERGE(sc, perf_intr);
TULIP_PERFMERGE(sc, perf_ifstart);
TULIP_PERFMERGE(sc, perf_ifioctl);
TULIP_PERFMERGE(sc, perf_ifwatchdog);
TULIP_PERFMERGE(sc, perf_timeout);
TULIP_PERFMERGE(sc, perf_ifstart_one);
TULIP_PERFMERGE(sc, perf_txput);
TULIP_PERFMERGE(sc, perf_txintr);
TULIP_PERFMERGE(sc, perf_rxintr);
TULIP_PERFMERGE(sc, perf_rxget);
TULIP_UNLOCK(sc);
}
static void
tulip_attach(tulip_softc_t * const sc)
{
struct ifnet *ifp;
ifp = sc->tulip_ifp = if_alloc(IFT_ETHER);
/* XXX: driver name/unit should be set some other way */
if_initname(ifp, "de", sc->tulip_unit);
ifp->if_softc = sc;
ifp->if_flags = IFF_BROADCAST|IFF_SIMPLEX|IFF_MULTICAST;
ifp->if_ioctl = tulip_ifioctl;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
ifp->if_start = tulip_start;
ifp->if_watchdog = tulip_ifwatchdog;
ifp->if_timer = 1;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
ifp->if_init = tulip_init;
IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
ifp->if_snd.ifq_drv_maxlen = ifqmaxlen;
IFQ_SET_READY(&ifp->if_snd);
if_printf(ifp, "%s%s pass %d.%d%s\n",
sc->tulip_boardid,
tulip_chipdescs[sc->tulip_chipid],
(sc->tulip_revinfo & 0xF0) >> 4,
sc->tulip_revinfo & 0x0F,
(sc->tulip_features & (TULIP_HAVE_ISVSROM|TULIP_HAVE_OKSROM))
== TULIP_HAVE_ISVSROM ? " (invalid EESPROM checksum)" : "");
TULIP_LOCK(sc);
#if defined(__alpha__)
/*
* In case the SRM console told us about a bogus media,
* we need to check to be safe.
*/
if (sc->tulip_mediums[sc->tulip_media] == NULL)
sc->tulip_media = TULIP_MEDIA_UNKNOWN;
#endif
(*sc->tulip_boardsw->bd_media_probe)(sc);
ifmedia_init(&sc->tulip_ifmedia, 0,
tulip_ifmedia_change,
tulip_ifmedia_status);
sc->tulip_flags &= ~TULIP_DEVICEPROBE;
tulip_ifmedia_add(sc);
tulip_reset(sc);
TULIP_UNLOCK(sc);
ether_ifattach(sc->tulip_ifp, sc->tulip_enaddr);
}
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
/* Release memory for a single descriptor ring. */
static void
tulip_busdma_freering(tulip_ringinfo_t *ri)
{
int i;
/* Release the DMA maps and tag for data buffers. */
if (ri->ri_data_maps != NULL) {
for (i = 0; i < ri->ri_max; i++) {
if (ri->ri_data_maps[i] != NULL) {
bus_dmamap_destroy(ri->ri_data_tag, ri->ri_data_maps[i]);
ri->ri_data_maps[i] = NULL;
}
}
free(ri->ri_data_maps, M_DEVBUF);
ri->ri_data_maps = NULL;
}
if (ri->ri_data_tag != NULL) {
bus_dma_tag_destroy(ri->ri_data_tag);
ri->ri_data_tag = NULL;
}
/* Release the DMA memory and tag for the ring descriptors. */
if (ri->ri_dma_addr != 0) {
bus_dmamap_unload(ri->ri_ring_tag, ri->ri_ring_map);
ri->ri_dma_addr = 0;
}
if (ri->ri_descs != NULL) {
bus_dmamem_free(ri->ri_ring_tag, ri->ri_descs, ri->ri_ring_map);
ri->ri_ring_map = NULL;
ri->ri_descs = NULL;
}
if (ri->ri_ring_tag != NULL) {
bus_dma_tag_destroy(ri->ri_ring_tag);
ri->ri_ring_tag = NULL;
}
}
/* Allocate memory for a single descriptor ring. */
static int
tulip_busdma_allocring(device_t dev, tulip_softc_t * const sc, size_t count,
bus_size_t maxsize, int nsegs, tulip_ringinfo_t *ri, const char *name)
{
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
size_t size;
int error, i;
/* First, setup a tag. */
ri->ri_max = count;
size = count * sizeof(tulip_desc_t);
error = bus_dma_tag_create(NULL, PAGE_SIZE, 0, BUS_SPACE_MAXADDR_32BIT,
BUS_SPACE_MAXADDR, NULL, NULL, size, 1, size, 0, NULL, NULL,
&ri->ri_ring_tag);
if (error) {
device_printf(dev, "failed to allocate %s descriptor ring dma tag\n",
name);
return (error);
}
/* Next, allocate memory for the descriptors. */
error = bus_dmamem_alloc(ri->ri_ring_tag, (void **)&ri->ri_descs,
BUS_DMA_NOWAIT | BUS_DMA_ZERO, &ri->ri_ring_map);
if (error) {
device_printf(dev, "failed to allocate memory for %s descriptor ring\n",
name);
return (error);
}
/* Map the descriptors. */
error = bus_dmamap_load(ri->ri_ring_tag, ri->ri_ring_map, ri->ri_descs,
size, tulip_dma_map_addr, &ri->ri_dma_addr, BUS_DMA_NOWAIT);
if (error) {
device_printf(dev, "failed to get dma address for %s descriptor ring\n",
name);
return (error);
}
/* Allocate a tag for the data buffers. */
error = bus_dma_tag_create(NULL, 4, 0,
BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL, NULL,
maxsize, nsegs, TULIP_DATA_PER_DESC, 0, NULL, NULL, &ri->ri_data_tag);
if (error) {
device_printf(dev, "failed to allocate %s buffer dma tag\n", name);
return (error);
}
/* Allocate maps for the data buffers. */
ri->ri_data_maps = malloc(sizeof(bus_dmamap_t) * count, M_DEVBUF,
M_WAITOK | M_ZERO);
for (i = 0; i < count; i++) {
error = bus_dmamap_create(ri->ri_data_tag, 0, &ri->ri_data_maps[i]);
if (error) {
device_printf(dev, "failed to create map for %s buffer %d\n",
name, i);
return (error);
}
}
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
return (0);
}
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
/* Release busdma maps, tags, and memory. */
static void
tulip_busdma_cleanup(tulip_softc_t * const sc)
{
/* Release resources for the setup descriptor. */
if (sc->tulip_setup_dma_addr != 0) {
bus_dmamap_unload(sc->tulip_setup_tag, sc->tulip_setup_map);
sc->tulip_setup_dma_addr = 0;
}
if (sc->tulip_setupbuf != NULL) {
bus_dmamem_free(sc->tulip_setup_tag, sc->tulip_setupdata,
sc->tulip_setup_map);
sc->tulip_setup_map = NULL;
sc->tulip_setupbuf = NULL;
}
if (sc->tulip_setup_tag != NULL) {
bus_dma_tag_destroy(sc->tulip_setup_tag);
sc->tulip_setup_tag = NULL;
}
/* Release the transmit ring. */
tulip_busdma_freering(&sc->tulip_txinfo);
/* Release the receive ring. */
tulip_busdma_freering(&sc->tulip_rxinfo);
}
static int
tulip_busdma_init(device_t dev, tulip_softc_t * const sc)
{
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
int error;
/*
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
* Allocate space and dmamap for transmit ring.
*/
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
error = tulip_busdma_allocring(dev, sc, TULIP_TXDESCS, TULIP_DATA_PER_DESC,
TULIP_MAX_TXSEG, &sc->tulip_txinfo, "transmit");
if (error)
return (error);
/*
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
* Allocate space and dmamap for receive ring. We tell bus_dma that
* we can map MCLBYTES so that it will accept a full MCLBYTES cluster,
* but we will only map the first TULIP_RX_BUFLEN bytes. This is not
* a waste in practice though as an ethernet frame can easily fit
* in TULIP_RX_BUFLEN bytes.
*/
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
error = tulip_busdma_allocring(dev, sc, TULIP_RXDESCS, MCLBYTES, 1,
&sc->tulip_rxinfo, "receive");
if (error)
return (error);
/*
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
* Allocate a DMA tag, memory, and map for setup descriptor
*/
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
error = bus_dma_tag_create(NULL, 4, 0,
BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL, NULL,
sizeof(sc->tulip_setupdata), 1, sizeof(sc->tulip_setupdata), 0,
NULL, NULL, &sc->tulip_setup_tag);
if (error) {
device_printf(dev, "failed to allocate setup descriptor dma tag\n");
return (error);
}
error = bus_dmamem_alloc(sc->tulip_setup_tag, (void **)&sc->tulip_setupbuf,
BUS_DMA_NOWAIT | BUS_DMA_ZERO, &sc->tulip_setup_map);
if (error) {
device_printf(dev, "failed to allocate memory for setup descriptor\n");
return (error);
}
error = bus_dmamap_load(sc->tulip_setup_tag, sc->tulip_setup_map,
sc->tulip_setupbuf, sizeof(sc->tulip_setupdata),
tulip_dma_map_addr, &sc->tulip_setup_dma_addr, BUS_DMA_NOWAIT);
if (error) {
device_printf(dev, "failed to get dma address for setup descriptor\n");
return (error);
}
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
return error;
}
static void
tulip_initcsrs(tulip_softc_t * const sc, tulip_csrptr_t csr_base,
size_t csr_size)
{
sc->tulip_csrs.csr_busmode = csr_base + 0 * csr_size;
sc->tulip_csrs.csr_txpoll = csr_base + 1 * csr_size;
sc->tulip_csrs.csr_rxpoll = csr_base + 2 * csr_size;
sc->tulip_csrs.csr_rxlist = csr_base + 3 * csr_size;
sc->tulip_csrs.csr_txlist = csr_base + 4 * csr_size;
sc->tulip_csrs.csr_status = csr_base + 5 * csr_size;
sc->tulip_csrs.csr_command = csr_base + 6 * csr_size;
sc->tulip_csrs.csr_intr = csr_base + 7 * csr_size;
sc->tulip_csrs.csr_missed_frames = csr_base + 8 * csr_size;
sc->tulip_csrs.csr_9 = csr_base + 9 * csr_size;
sc->tulip_csrs.csr_10 = csr_base + 10 * csr_size;
sc->tulip_csrs.csr_11 = csr_base + 11 * csr_size;
sc->tulip_csrs.csr_12 = csr_base + 12 * csr_size;
sc->tulip_csrs.csr_13 = csr_base + 13 * csr_size;
sc->tulip_csrs.csr_14 = csr_base + 14 * csr_size;
sc->tulip_csrs.csr_15 = csr_base + 15 * csr_size;
}
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
static int
tulip_initring(
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
device_t dev,
tulip_softc_t * const sc,
tulip_ringinfo_t * const ri,
int ndescs)
{
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
int i;
ri->ri_descinfo = malloc(sizeof(tulip_descinfo_t) * ndescs, M_DEVBUF,
M_WAITOK | M_ZERO);
for (i = 0; i < ndescs; i++) {
ri->ri_descinfo[i].di_desc = &ri->ri_descs[i];
ri->ri_descinfo[i].di_map = &ri->ri_data_maps[i];
}
ri->ri_first = ri->ri_descinfo;
ri->ri_max = ndescs;
ri->ri_last = ri->ri_first + ri->ri_max;
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
bzero(ri->ri_descs, sizeof(tulip_desc_t) * ri->ri_max);
ri->ri_last[-1].di_desc->d_flag = TULIP_DFLAG_ENDRING;
return (0);
}
/*
1999-08-09 14:43:39 +00:00
* This is the PCI configuration support.
*/
#define PCI_CBIO PCIR_BAR(0) /* Configuration Base IO Address */
#define PCI_CBMA PCIR_BAR(1) /* Configuration Base Memory Address */
#define PCI_CFDA 0x40 /* Configuration Driver Area */
static int
tulip_pci_probe(device_t dev)
{
const char *name = NULL;
if (pci_get_vendor(dev) != DEC_VENDORID)
return ENXIO;
/*
* Some LanMedia WAN cards use the Tulip chip, but they have
* their own driver, and we should not recognize them
*/
if (pci_get_subvendor(dev) == 0x1376)
return ENXIO;
switch (pci_get_device(dev)) {
case CHIPID_21040:
name = "Digital 21040 Ethernet";
break;
case CHIPID_21041:
name = "Digital 21041 Ethernet";
break;
case CHIPID_21140:
if (pci_get_revid(dev) >= 0x20)
name = "Digital 21140A Fast Ethernet";
else
name = "Digital 21140 Fast Ethernet";
break;
case CHIPID_21142:
if (pci_get_revid(dev) >= 0x20)
name = "Digital 21143 Fast Ethernet";
else
name = "Digital 21142 Fast Ethernet";
break;
}
if (name) {
device_set_desc(dev, name);
2005-02-24 21:32:56 +00:00
return BUS_PROBE_LOW_PRIORITY;
}
return ENXIO;
}
static int
tulip_shutdown(device_t dev)
{
tulip_softc_t * const sc = device_get_softc(dev);
TULIP_CSR_WRITE(sc, csr_busmode, TULIP_BUSMODE_SWRESET);
DELAY(10); /* Wait 10 microseconds (actually 50 PCI cycles but at
33MHz that comes to two microseconds but wait a
bit longer anyways) */
return 0;
}
static int
tulip_pci_attach(device_t dev)
{
tulip_softc_t *sc;
#if defined(__alpha__)
tulip_media_t media = TULIP_MEDIA_UNKNOWN;
#endif
int retval, idx;
u_int32_t revinfo, cfdainfo;
unsigned csroffset = TULIP_PCI_CSROFFSET;
unsigned csrsize = TULIP_PCI_CSRSIZE;
tulip_csrptr_t csr_base;
tulip_chipid_t chipid = TULIP_CHIPID_UNKNOWN;
struct resource *res;
int rid, unit;
unit = device_get_unit(dev);
if (unit >= TULIP_MAX_DEVICES) {
device_printf(dev, "not configured; limit of %d reached or exceeded\n",
TULIP_MAX_DEVICES);
return ENXIO;
}
revinfo = pci_get_revid(dev);
cfdainfo = pci_read_config(dev, PCI_CFDA, 4);
/* turn busmaster on in case BIOS doesn't set it */
pci_enable_busmaster(dev);
if (pci_get_vendor(dev) == DEC_VENDORID) {
if (pci_get_device(dev) == CHIPID_21040)
chipid = TULIP_21040;
else if (pci_get_device(dev) == CHIPID_21041)
chipid = TULIP_21041;
else if (pci_get_device(dev) == CHIPID_21140)
chipid = (revinfo >= 0x20) ? TULIP_21140A : TULIP_21140;
else if (pci_get_device(dev) == CHIPID_21142)
chipid = (revinfo >= 0x20) ? TULIP_21143 : TULIP_21142;
}
if (chipid == TULIP_CHIPID_UNKNOWN)
return ENXIO;
1999-08-09 14:43:39 +00:00
if (chipid == TULIP_21040 && revinfo < 0x20) {
device_printf(dev,
"not configured; 21040 pass 2.0 required (%d.%d found)\n",
revinfo >> 4, revinfo & 0x0f);
return ENXIO;
} else if (chipid == TULIP_21140 && revinfo < 0x11) {
device_printf(dev,
"not configured; 21140 pass 1.1 required (%d.%d found)\n",
revinfo >> 4, revinfo & 0x0f);
return ENXIO;
}
sc = device_get_softc(dev);
sc->tulip_pci_busno = pci_get_bus(dev);
sc->tulip_pci_devno = pci_get_slot(dev);
sc->tulip_chipid = chipid;
sc->tulip_flags |= TULIP_DEVICEPROBE;
if (chipid == TULIP_21140 || chipid == TULIP_21140A)
1997-08-03 13:00:42 +00:00
sc->tulip_features |= TULIP_HAVE_GPR|TULIP_HAVE_STOREFWD;
if (chipid == TULIP_21140A && revinfo <= 0x22)
sc->tulip_features |= TULIP_HAVE_RXBADOVRFLW;
if (chipid == TULIP_21140)
sc->tulip_features |= TULIP_HAVE_BROKEN_HASH;
1999-08-09 14:43:39 +00:00
if (chipid != TULIP_21040 && chipid != TULIP_21140)
sc->tulip_features |= TULIP_HAVE_POWERMGMT;
if (chipid == TULIP_21041 || chipid == TULIP_21142 || chipid == TULIP_21143) {
sc->tulip_features |= TULIP_HAVE_DUALSENSE;
if (chipid != TULIP_21041 || revinfo >= 0x20)
sc->tulip_features |= TULIP_HAVE_SIANWAY;
if (chipid != TULIP_21041)
1997-08-03 13:00:42 +00:00
sc->tulip_features |= TULIP_HAVE_SIAGP|TULIP_HAVE_RXBADOVRFLW|TULIP_HAVE_STOREFWD;
if (chipid != TULIP_21041 && revinfo >= 0x20)
sc->tulip_features |= TULIP_HAVE_SIA100;
}
if (sc->tulip_features & TULIP_HAVE_POWERMGMT
&& (cfdainfo & (TULIP_CFDA_SLEEP|TULIP_CFDA_SNOOZE))) {
cfdainfo &= ~(TULIP_CFDA_SLEEP|TULIP_CFDA_SNOOZE);
pci_write_config(dev, PCI_CFDA, cfdainfo, 4);
DELAY(11*1000);
}
#if defined(__alpha__)
/*
* The Alpha SRM console encodes a console set media in the driver
* part of the CFDA register. Note that the Multia presents a
* problem in that its BNC mode is really EXTSIA. So in that case
* force a probe.
*/
switch ((cfdainfo >> 8) & 0xff) {
1999-08-09 14:43:39 +00:00
case 1: media = chipid > TULIP_21040 ? TULIP_MEDIA_AUI : TULIP_MEDIA_AUIBNC; break;
case 2: media = chipid > TULIP_21040 ? TULIP_MEDIA_BNC : TULIP_MEDIA_UNKNOWN; break;
case 3: media = TULIP_MEDIA_10BASET; break;
case 4: media = TULIP_MEDIA_10BASET_FD; break;
case 5: media = TULIP_MEDIA_100BASETX; break;
case 6: media = TULIP_MEDIA_100BASETX_FD; break;
default: media = TULIP_MEDIA_UNKNOWN; break;
}
#endif
sc->tulip_unit = unit;
sc->tulip_revinfo = revinfo;
#if defined(TULIP_IOMAPPED)
rid = PCI_CBIO;
res = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid, RF_ACTIVE);
#else
rid = PCI_CBMA;
res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid, RF_ACTIVE);
#endif
if (!res)
return ENXIO;
sc->tulip_csrs_bst = rman_get_bustag(res);
sc->tulip_csrs_bsh = rman_get_bushandle(res);
csr_base = 0;
mtx_init(TULIP_MUTEX(sc), MTX_NETWORK_LOCK, device_get_nameunit(dev),
MTX_DEF);
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
callout_init_mtx(&sc->tulip_callout, TULIP_MUTEX(sc), 0);
tulips[unit] = sc;
tulip_initcsrs(sc, csr_base + csroffset, csrsize);
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
if ((retval = tulip_busdma_init(dev, sc)) != 0) {
device_printf(dev, "error initing bus_dma: %d\n", retval);
tulip_busdma_cleanup(sc);
mtx_destroy(TULIP_MUTEX(sc));
return ENXIO;
}
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
retval = tulip_initring(dev, sc, &sc->tulip_rxinfo, TULIP_RXDESCS);
if (retval == 0)
retval = tulip_initring(dev, sc, &sc->tulip_txinfo, TULIP_TXDESCS);
if (retval) {
tulip_busdma_cleanup(sc);
mtx_destroy(TULIP_MUTEX(sc));
return retval;
}
/*
* Make sure there won't be any interrupts or such...
*/
TULIP_CSR_WRITE(sc, csr_busmode, TULIP_BUSMODE_SWRESET);
DELAY(100); /* Wait 10 microseconds (actually 50 PCI cycles but at
33MHz that comes to two microseconds but wait a
bit longer anyways) */
TULIP_LOCK(sc);
retval = tulip_read_macaddr(sc);
TULIP_UNLOCK(sc);
if (retval < 0) {
device_printf(dev, "can't read ENET ROM (why=%d) (", retval);
for (idx = 0; idx < 32; idx++)
printf("%02x", sc->tulip_rombuf[idx]);
printf("\n");
device_printf(dev, "%s%s pass %d.%d\n",
sc->tulip_boardid, tulip_chipdescs[sc->tulip_chipid],
(sc->tulip_revinfo & 0xF0) >> 4, sc->tulip_revinfo & 0x0F);
device_printf(dev, "address unknown\n");
} else {
void (*intr_rtn)(void *) = tulip_intr_normal;
if (sc->tulip_features & TULIP_HAVE_SHAREDINTR)
intr_rtn = tulip_intr_shared;
#if defined(__alpha__)
sc->tulip_media = media;
#endif
tulip_attach(sc);
/* Setup interrupt last. */
if ((sc->tulip_features & TULIP_HAVE_SLAVEDINTR) == 0) {
void *ih;
rid = 0;
res = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
RF_SHAREABLE | RF_ACTIVE);
if (res == 0 || bus_setup_intr(dev, res, INTR_TYPE_NET |
INTR_MPSAFE, intr_rtn, sc, &ih)) {
device_printf(dev, "couldn't map interrupt\n");
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
tulip_busdma_cleanup(sc);
ether_ifdetach(sc->tulip_ifp);
if_free(sc->tulip_ifp);
mtx_destroy(TULIP_MUTEX(sc));
return ENXIO;
}
}
#if defined(__alpha__)
TULIP_LOCK(sc);
if (sc->tulip_media != TULIP_MEDIA_UNKNOWN)
tulip_linkup(sc, media);
TULIP_UNLOCK(sc);
#endif
}
return 0;
}
static device_method_t tulip_pci_methods[] = {
/* Device interface */
DEVMETHOD(device_probe, tulip_pci_probe),
DEVMETHOD(device_attach, tulip_pci_attach),
DEVMETHOD(device_shutdown, tulip_shutdown),
{ 0, 0 }
};
static driver_t tulip_pci_driver = {
"de",
tulip_pci_methods,
sizeof(tulip_softc_t),
};
static devclass_t tulip_devclass;
DRIVER_MODULE(de, pci, tulip_pci_driver, tulip_devclass, 0, 0);
Major rototill of this driver to add FreeBSD bus-dma support: - Allocate arrays of metadata for the descriptors in the rx and tx rings and change the ring pointers to walk the metadata array rather than the actual descriptor rings. Each metadata object contains a pointer to its descriptor, a pointer to any associated mbuf, and a pointer to the associated bus_dmamap_t in the bus_dma case. The mbuf pointers replace the tulip_txq and tulip_rxq local ifqueue's in the softc. - Add lots of KTR trace entries using a local KTR_TULIP level which defaults to 0, but can be changed to KTR_DEV at the top of the file when debugging. - Rename tulip_init(), tulip_start(), tulip_ifinit(), and tulip_ifstart() to tulip_init_locked(), tulip_start_locked(), tulip_init(), and tulip_start(), respectively, to match the convention in other drivers. - Add a TULIP_SP_MAC() macro to encode two bytes of the MAC address into the setup buffer and use that in place of lots of BYTE_ORDER #ifdef's. Also, remove an incorrect XXX comment I added earlier, the driver was correct (at least it does the same thing dc(4) does). TULIP_SP_MAC was shamelessly copied from DC_SP_MAC() in dc(4). - Remove the #ifdef'd NetBSD bus-dma code and replace it with FreeBSD bus-dma code that not only compiles but even works at runtime. - Use callout_init_mtx() instead of just callout_init(). - Correct the various wrapper macros for bus_dmamap_sync() for the rx and tx buffers to only ask for the sync ops that they actually need. - Tidy the #ifdef TULIP_COPY_RXDATA code by expanding an #ifdef a bit so it becomes easier to read at the expense of a couple of duplicated lines of code. Also, use m_getcl() to get an mbuf cluster rather than MGETHDR() followed by MCLGET(). - Maintain the ring free (ri_free) count for the rx ring metadata since we no longer have tulip_rxq.ifq_len around to indicate how many mbuf's are currently in the rx ring. - Add code to teardown bus_dma resources when attach fails and generally fixup attach to do a better job of cleaning up when it fails. This gets us a good bit closer to possibly having a detach method someday and making this driver an unloadable module. - Add some functions that can be called from ddb to dump the state of a descriptor ring and to dump the state of an individual descriptor. - Various comment grammer and spelling fixes. I have bus-dma turned on by default, but I've left the non-bus-dma code around so that it can be turned off to aid in debugging should any problems turn up later on. I'll be removing the non-bus-dma code in a subsequent commit.
2005-08-25 20:12:43 +00:00
#ifdef DDB
void tulip_dumpring(int unit, int ring);
void tulip_dumpdesc(int unit, int ring, int desc);
void tulip_status(int unit);
void
tulip_dumpring(int unit, int ring)
{
tulip_softc_t *sc;
tulip_ringinfo_t *ri;
tulip_descinfo_t *di;
if (unit < 0 || unit >= TULIP_MAX_DEVICES) {
db_printf("invalid unit %d\n", unit);
return;
}
sc = tulips[unit];
if (sc == NULL) {
db_printf("unit %d not present\n", unit);
return;
}
switch (ring) {
case 0:
db_printf("receive ring:\n");
ri = &sc->tulip_rxinfo;
break;
case 1:
db_printf("transmit ring:\n");
ri = &sc->tulip_txinfo;
break;
default:
db_printf("invalid ring %d\n", ring);
return;
}
db_printf(" nextin: %td, nextout: %td, max: %d, free: %d\n",
ri->ri_nextin - ri->ri_first, ri->ri_nextout - ri->ri_first,
ri->ri_max, ri->ri_free);
for (di = ri->ri_first; di != ri->ri_last; di++) {
if (di->di_mbuf != NULL)
db_printf(" descriptor %td: mbuf %p\n", di - ri->ri_first,
di->di_mbuf);
else if (di->di_desc->d_flag & TULIP_DFLAG_TxSETUPPKT)
db_printf(" descriptor %td: setup packet\n", di - ri->ri_first);
}
}
void
tulip_dumpdesc(int unit, int ring, int desc)
{
tulip_softc_t *sc;
tulip_ringinfo_t *ri;
tulip_descinfo_t *di;
char *s;
if (unit < 0 || unit >= TULIP_MAX_DEVICES) {
db_printf("invalid unit %d\n", unit);
return;
}
sc = tulips[unit];
if (sc == NULL) {
db_printf("unit %d not present\n", unit);
return;
}
switch (ring) {
case 0:
s = "receive";
ri = &sc->tulip_rxinfo;
break;
case 1:
s = "transmit";
ri = &sc->tulip_txinfo;
break;
default:
db_printf("invalid ring %d\n", ring);
return;
}
if (desc < 0 || desc >= ri->ri_max) {
db_printf("invalid descriptor %d\n", desc);
return;
}
db_printf("%s descriptor %d:\n", s, desc);
di = &ri->ri_first[desc];
db_printf(" mbuf: %p\n", di->di_mbuf);
db_printf(" status: %08x flag: %03x\n", di->di_desc->d_status,
di->di_desc->d_flag);
db_printf(" addr1: %08x len1: %03x\n", di->di_desc->d_addr1,
di->di_desc->d_length1);
db_printf(" addr2: %08x len2: %03x\n", di->di_desc->d_addr2,
di->di_desc->d_length2);
}
#endif