freebsd-nq/sys/dev/bfe/if_bfe.c

1965 lines
50 KiB
C
Raw Normal View History

/*-
* Copyright (c) 2003 Stuart Walsh<stu@ipng.org.uk>
* and Duncan Barclay<dmlb@dmlb.org>
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS 'AS IS' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include <sys/param.h>
#include <sys/systm.h>
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
#include <sys/bus.h>
#include <sys/endian.h>
#include <sys/kernel.h>
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
#include <sys/malloc.h>
#include <sys/mbuf.h>
2004-05-30 20:08:47 +00:00
#include <sys/module.h>
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
#include <sys/rman.h>
#include <sys/socket.h>
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
#include <sys/sockio.h>
#include <sys/sysctl.h>
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
#include <net/bpf.h>
#include <net/if.h>
#include <net/if_var.h>
#include <net/ethernet.h>
#include <net/if_dl.h>
#include <net/if_media.h>
#include <net/if_types.h>
#include <net/if_vlan_var.h>
#include <dev/mii/mii.h>
#include <dev/mii/miivar.h>
#include <dev/pci/pcireg.h>
#include <dev/pci/pcivar.h>
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
#include <machine/bus.h>
#include <dev/bfe/if_bfereg.h>
MODULE_DEPEND(bfe, pci, 1, 1, 1);
MODULE_DEPEND(bfe, ether, 1, 1, 1);
MODULE_DEPEND(bfe, miibus, 1, 1, 1);
/* "device miibus" required. See GENERIC if you get errors here. */
#include "miibus_if.h"
#define BFE_DEVDESC_MAX 64 /* Maximum device description length */
static struct bfe_type bfe_devs[] = {
{ BCOM_VENDORID, BCOM_DEVICEID_BCM4401,
"Broadcom BCM4401 Fast Ethernet" },
{ BCOM_VENDORID, BCOM_DEVICEID_BCM4401B0,
"Broadcom BCM4401-B0 Fast Ethernet" },
{ 0, 0, NULL }
};
static int bfe_probe (device_t);
static int bfe_attach (device_t);
static int bfe_detach (device_t);
static int bfe_suspend (device_t);
static int bfe_resume (device_t);
static void bfe_release_resources (struct bfe_softc *);
static void bfe_intr (void *);
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
static int bfe_encap (struct bfe_softc *, struct mbuf **);
static void bfe_start (struct ifnet *);
static void bfe_start_locked (struct ifnet *);
static int bfe_ioctl (struct ifnet *, u_long, caddr_t);
static void bfe_init (void *);
static void bfe_init_locked (void *);
static void bfe_stop (struct bfe_softc *);
static void bfe_watchdog (struct bfe_softc *);
static int bfe_shutdown (device_t);
static void bfe_tick (void *);
static void bfe_txeof (struct bfe_softc *);
static void bfe_rxeof (struct bfe_softc *);
static void bfe_set_rx_mode (struct bfe_softc *);
static int bfe_list_rx_init (struct bfe_softc *);
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
static void bfe_list_tx_init (struct bfe_softc *);
static void bfe_discard_buf (struct bfe_softc *, int);
static int bfe_list_newbuf (struct bfe_softc *, int);
static void bfe_rx_ring_free (struct bfe_softc *);
static void bfe_pci_setup (struct bfe_softc *, u_int32_t);
static int bfe_ifmedia_upd (struct ifnet *);
static void bfe_ifmedia_sts (struct ifnet *, struct ifmediareq *);
static int bfe_miibus_readreg (device_t, int, int);
static int bfe_miibus_writereg (device_t, int, int, int);
static void bfe_miibus_statchg (device_t);
2004-08-07 20:55:53 +00:00
static int bfe_wait_bit (struct bfe_softc *, u_int32_t, u_int32_t,
u_long, const int);
static void bfe_get_config (struct bfe_softc *sc);
static void bfe_read_eeprom (struct bfe_softc *, u_int8_t *);
static void bfe_stats_update (struct bfe_softc *);
static void bfe_clear_stats (struct bfe_softc *);
static int bfe_readphy (struct bfe_softc *, u_int32_t, u_int32_t*);
static int bfe_writephy (struct bfe_softc *, u_int32_t, u_int32_t);
static int bfe_resetphy (struct bfe_softc *);
static int bfe_setupphy (struct bfe_softc *);
static void bfe_chip_reset (struct bfe_softc *);
static void bfe_chip_halt (struct bfe_softc *);
static void bfe_core_reset (struct bfe_softc *);
static void bfe_core_disable (struct bfe_softc *);
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
static int bfe_dma_alloc (struct bfe_softc *);
static void bfe_dma_free (struct bfe_softc *sc);
static void bfe_dma_map (void *, bus_dma_segment_t *, int, int);
static void bfe_cam_write (struct bfe_softc *, u_char *, int);
static int sysctl_bfe_stats (SYSCTL_HANDLER_ARGS);
static device_method_t bfe_methods[] = {
/* Device interface */
DEVMETHOD(device_probe, bfe_probe),
DEVMETHOD(device_attach, bfe_attach),
DEVMETHOD(device_detach, bfe_detach),
DEVMETHOD(device_shutdown, bfe_shutdown),
DEVMETHOD(device_suspend, bfe_suspend),
DEVMETHOD(device_resume, bfe_resume),
/* MII interface */
DEVMETHOD(miibus_readreg, bfe_miibus_readreg),
DEVMETHOD(miibus_writereg, bfe_miibus_writereg),
DEVMETHOD(miibus_statchg, bfe_miibus_statchg),
DEVMETHOD_END
};
static driver_t bfe_driver = {
"bfe",
bfe_methods,
sizeof(struct bfe_softc)
};
static devclass_t bfe_devclass;
DRIVER_MODULE(bfe, pci, bfe_driver, bfe_devclass, 0, 0);
DRIVER_MODULE(miibus, bfe, miibus_driver, miibus_devclass, 0, 0);
/*
2004-08-07 20:55:53 +00:00
* Probe for a Broadcom 4401 chip.
*/
static int
bfe_probe(device_t dev)
{
struct bfe_type *t;
t = bfe_devs;
while (t->bfe_name != NULL) {
if (pci_get_vendor(dev) == t->bfe_vid &&
pci_get_device(dev) == t->bfe_did) {
device_set_desc(dev, t->bfe_name);
return (BUS_PROBE_DEFAULT);
}
t++;
}
2004-08-07 20:55:53 +00:00
return (ENXIO);
}
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
struct bfe_dmamap_arg {
bus_addr_t bfe_busaddr;
};
static int
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
bfe_dma_alloc(struct bfe_softc *sc)
{
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
struct bfe_dmamap_arg ctx;
struct bfe_rx_data *rd;
struct bfe_tx_data *td;
int error, i;
/*
* parent tag. Apparently the chip cannot handle any DMA address
* greater than 1GB.
*/
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
error = bus_dma_tag_create(bus_get_dma_tag(sc->bfe_dev), /* parent */
1, 0, /* alignment, boundary */
BFE_DMA_MAXADDR, /* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
BUS_SPACE_MAXSIZE_32BIT, /* maxsize */
0, /* nsegments */
BUS_SPACE_MAXSIZE_32BIT, /* maxsegsize */
0, /* flags */
NULL, NULL, /* lockfunc, lockarg */
&sc->bfe_parent_tag);
if (error != 0) {
device_printf(sc->bfe_dev, "cannot create parent DMA tag.\n");
goto fail;
}
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
/* Create tag for Tx ring. */
error = bus_dma_tag_create(sc->bfe_parent_tag, /* parent */
BFE_TX_RING_ALIGN, 0, /* alignment, boundary */
BUS_SPACE_MAXADDR, /* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
BFE_TX_LIST_SIZE, /* maxsize */
1, /* nsegments */
BFE_TX_LIST_SIZE, /* maxsegsize */
0, /* flags */
NULL, NULL, /* lockfunc, lockarg */
&sc->bfe_tx_tag);
if (error != 0) {
device_printf(sc->bfe_dev, "cannot create Tx ring DMA tag.\n");
goto fail;
}
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
/* Create tag for Rx ring. */
error = bus_dma_tag_create(sc->bfe_parent_tag, /* parent */
BFE_RX_RING_ALIGN, 0, /* alignment, boundary */
BUS_SPACE_MAXADDR, /* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
BFE_RX_LIST_SIZE, /* maxsize */
1, /* nsegments */
BFE_RX_LIST_SIZE, /* maxsegsize */
0, /* flags */
NULL, NULL, /* lockfunc, lockarg */
&sc->bfe_rx_tag);
if (error != 0) {
device_printf(sc->bfe_dev, "cannot create Rx ring DMA tag.\n");
goto fail;
}
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
/* Create tag for Tx buffers. */
error = bus_dma_tag_create(sc->bfe_parent_tag, /* parent */
1, 0, /* alignment, boundary */
BUS_SPACE_MAXADDR, /* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
MCLBYTES * BFE_MAXTXSEGS, /* maxsize */
BFE_MAXTXSEGS, /* nsegments */
MCLBYTES, /* maxsegsize */
0, /* flags */
NULL, NULL, /* lockfunc, lockarg */
&sc->bfe_txmbuf_tag);
if (error != 0) {
device_printf(sc->bfe_dev,
"cannot create Tx buffer DMA tag.\n");
goto fail;
}
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
/* Create tag for Rx buffers. */
error = bus_dma_tag_create(sc->bfe_parent_tag, /* parent */
1, 0, /* alignment, boundary */
BUS_SPACE_MAXADDR, /* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
MCLBYTES, /* maxsize */
1, /* nsegments */
MCLBYTES, /* maxsegsize */
0, /* flags */
NULL, NULL, /* lockfunc, lockarg */
&sc->bfe_rxmbuf_tag);
if (error != 0) {
device_printf(sc->bfe_dev,
"cannot create Rx buffer DMA tag.\n");
goto fail;
}
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
/* Allocate DMA'able memory and load DMA map. */
error = bus_dmamem_alloc(sc->bfe_tx_tag, (void *)&sc->bfe_tx_list,
BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT, &sc->bfe_tx_map);
if (error != 0) {
device_printf(sc->bfe_dev,
"cannot allocate DMA'able memory for Tx ring.\n");
goto fail;
}
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
ctx.bfe_busaddr = 0;
error = bus_dmamap_load(sc->bfe_tx_tag, sc->bfe_tx_map,
sc->bfe_tx_list, BFE_TX_LIST_SIZE, bfe_dma_map, &ctx,
BUS_DMA_NOWAIT);
if (error != 0 || ctx.bfe_busaddr == 0) {
device_printf(sc->bfe_dev,
"cannot load DMA'able memory for Tx ring.\n");
goto fail;
}
sc->bfe_tx_dma = BFE_ADDR_LO(ctx.bfe_busaddr);
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
error = bus_dmamem_alloc(sc->bfe_rx_tag, (void *)&sc->bfe_rx_list,
BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT, &sc->bfe_rx_map);
if (error != 0) {
device_printf(sc->bfe_dev,
"cannot allocate DMA'able memory for Rx ring.\n");
goto fail;
}
ctx.bfe_busaddr = 0;
error = bus_dmamap_load(sc->bfe_rx_tag, sc->bfe_rx_map,
sc->bfe_rx_list, BFE_RX_LIST_SIZE, bfe_dma_map, &ctx,
BUS_DMA_NOWAIT);
if (error != 0 || ctx.bfe_busaddr == 0) {
device_printf(sc->bfe_dev,
"cannot load DMA'able memory for Rx ring.\n");
goto fail;
}
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
sc->bfe_rx_dma = BFE_ADDR_LO(ctx.bfe_busaddr);
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
/* Create DMA maps for Tx buffers. */
for (i = 0; i < BFE_TX_LIST_CNT; i++) {
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
td = &sc->bfe_tx_ring[i];
td->bfe_mbuf = NULL;
td->bfe_map = NULL;
error = bus_dmamap_create(sc->bfe_txmbuf_tag, 0, &td->bfe_map);
if (error != 0) {
device_printf(sc->bfe_dev,
"cannot create DMA map for Tx.\n");
goto fail;
}
}
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
/* Create spare DMA map for Rx buffers. */
error = bus_dmamap_create(sc->bfe_rxmbuf_tag, 0, &sc->bfe_rx_sparemap);
if (error != 0) {
device_printf(sc->bfe_dev, "cannot create spare DMA map for Rx.\n");
goto fail;
}
/* Create DMA maps for Rx buffers. */
for (i = 0; i < BFE_RX_LIST_CNT; i++) {
rd = &sc->bfe_rx_ring[i];
rd->bfe_mbuf = NULL;
rd->bfe_map = NULL;
rd->bfe_ctrl = 0;
error = bus_dmamap_create(sc->bfe_rxmbuf_tag, 0, &rd->bfe_map);
if (error != 0) {
device_printf(sc->bfe_dev,
"cannot create DMA map for Rx.\n");
goto fail;
}
}
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
fail:
return (error);
}
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
static void
bfe_dma_free(struct bfe_softc *sc)
{
struct bfe_tx_data *td;
struct bfe_rx_data *rd;
int i;
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
/* Tx ring. */
if (sc->bfe_tx_tag != NULL) {
if (sc->bfe_tx_map != NULL)
bus_dmamap_unload(sc->bfe_tx_tag, sc->bfe_tx_map);
if (sc->bfe_tx_map != NULL && sc->bfe_tx_list != NULL)
bus_dmamem_free(sc->bfe_tx_tag, sc->bfe_tx_list,
sc->bfe_tx_map);
sc->bfe_tx_map = NULL;
sc->bfe_tx_list = NULL;
bus_dma_tag_destroy(sc->bfe_tx_tag);
sc->bfe_tx_tag = NULL;
}
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
/* Rx ring. */
if (sc->bfe_rx_tag != NULL) {
if (sc->bfe_rx_map != NULL)
bus_dmamap_unload(sc->bfe_rx_tag, sc->bfe_rx_map);
if (sc->bfe_rx_map != NULL && sc->bfe_rx_list != NULL)
bus_dmamem_free(sc->bfe_rx_tag, sc->bfe_rx_list,
sc->bfe_rx_map);
sc->bfe_rx_map = NULL;
sc->bfe_rx_list = NULL;
bus_dma_tag_destroy(sc->bfe_rx_tag);
sc->bfe_rx_tag = NULL;
}
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
/* Tx buffers. */
if (sc->bfe_txmbuf_tag != NULL) {
for (i = 0; i < BFE_TX_LIST_CNT; i++) {
td = &sc->bfe_tx_ring[i];
if (td->bfe_map != NULL) {
bus_dmamap_destroy(sc->bfe_txmbuf_tag,
td->bfe_map);
td->bfe_map = NULL;
}
}
bus_dma_tag_destroy(sc->bfe_txmbuf_tag);
sc->bfe_txmbuf_tag = NULL;
}
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
/* Rx buffers. */
if (sc->bfe_rxmbuf_tag != NULL) {
for (i = 0; i < BFE_RX_LIST_CNT; i++) {
rd = &sc->bfe_rx_ring[i];
if (rd->bfe_map != NULL) {
bus_dmamap_destroy(sc->bfe_rxmbuf_tag,
rd->bfe_map);
rd->bfe_map = NULL;
}
}
if (sc->bfe_rx_sparemap != NULL) {
bus_dmamap_destroy(sc->bfe_rxmbuf_tag,
sc->bfe_rx_sparemap);
sc->bfe_rx_sparemap = NULL;
}
bus_dma_tag_destroy(sc->bfe_rxmbuf_tag);
sc->bfe_rxmbuf_tag = NULL;
}
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
if (sc->bfe_parent_tag != NULL) {
bus_dma_tag_destroy(sc->bfe_parent_tag);
sc->bfe_parent_tag = NULL;
}
}
static int
bfe_attach(device_t dev)
{
struct ifnet *ifp = NULL;
struct bfe_softc *sc;
int error = 0, rid;
sc = device_get_softc(dev);
mtx_init(&sc->bfe_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
MTX_DEF);
callout_init_mtx(&sc->bfe_stat_co, &sc->bfe_mtx, 0);
sc->bfe_dev = dev;
/*
* Map control/status registers.
*/
pci_enable_busmaster(dev);
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
rid = PCIR_BAR(0);
sc->bfe_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid,
RF_ACTIVE);
if (sc->bfe_res == NULL) {
device_printf(dev, "couldn't map memory\n");
error = ENXIO;
goto fail;
}
/* Allocate interrupt */
rid = 0;
sc->bfe_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
RF_SHAREABLE | RF_ACTIVE);
if (sc->bfe_irq == NULL) {
device_printf(dev, "couldn't map interrupt\n");
error = ENXIO;
goto fail;
}
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
if (bfe_dma_alloc(sc) != 0) {
device_printf(dev, "failed to allocate DMA resources\n");
error = ENXIO;
goto fail;
}
SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev),
SYSCTL_CHILDREN(device_get_sysctl_tree(dev)), OID_AUTO,
"stats", CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_bfe_stats,
"I", "Statistics");
/* Set up ifnet structure */
ifp = sc->bfe_ifp = if_alloc(IFT_ETHER);
if (ifp == NULL) {
device_printf(dev, "failed to if_alloc()\n");
error = ENOSPC;
goto fail;
}
ifp->if_softc = sc;
if_initname(ifp, device_get_name(dev), device_get_unit(dev));
ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
ifp->if_ioctl = bfe_ioctl;
ifp->if_start = bfe_start;
ifp->if_init = bfe_init;
IFQ_SET_MAXLEN(&ifp->if_snd, BFE_TX_QLEN);
ifp->if_snd.ifq_drv_maxlen = BFE_TX_QLEN;
IFQ_SET_READY(&ifp->if_snd);
bfe_get_config(sc);
/* Reset the chip and turn on the PHY */
BFE_LOCK(sc);
bfe_chip_reset(sc);
BFE_UNLOCK(sc);
error = mii_attach(dev, &sc->bfe_miibus, ifp, bfe_ifmedia_upd,
bfe_ifmedia_sts, BMSR_DEFCAPMASK, sc->bfe_phyaddr, MII_OFFSET_ANY,
0);
if (error != 0) {
device_printf(dev, "attaching PHYs failed\n");
goto fail;
}
ether_ifattach(ifp, sc->bfe_enaddr);
/*
* Tell the upper layer(s) we support long frames.
*/
ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
ifp->if_capabilities |= IFCAP_VLAN_MTU;
ifp->if_capenable |= IFCAP_VLAN_MTU;
/*
* Hook interrupt last to avoid having to lock softc
*/
error = bus_setup_intr(dev, sc->bfe_irq, INTR_TYPE_NET | INTR_MPSAFE,
NULL, bfe_intr, sc, &sc->bfe_intrhand);
if (error) {
device_printf(dev, "couldn't set up irq\n");
goto fail;
}
fail:
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
if (error != 0)
bfe_detach(dev);
2004-08-07 20:55:53 +00:00
return (error);
}
static int
bfe_detach(device_t dev)
{
struct bfe_softc *sc;
struct ifnet *ifp;
sc = device_get_softc(dev);
ifp = sc->bfe_ifp;
if (device_is_attached(dev)) {
BFE_LOCK(sc);
sc->bfe_flags |= BFE_FLAG_DETACH;
bfe_stop(sc);
BFE_UNLOCK(sc);
callout_drain(&sc->bfe_stat_co);
if (ifp != NULL)
ether_ifdetach(ifp);
}
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
BFE_LOCK(sc);
bfe_chip_reset(sc);
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
BFE_UNLOCK(sc);
bus_generic_detach(dev);
if (sc->bfe_miibus != NULL)
device_delete_child(dev, sc->bfe_miibus);
bfe_release_resources(sc);
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
bfe_dma_free(sc);
mtx_destroy(&sc->bfe_mtx);
2004-08-07 20:55:53 +00:00
return (0);
}
/*
* Stop all chip I/O so that the kernel's probe routines don't
* get confused by errant DMAs when rebooting.
*/
static int
bfe_shutdown(device_t dev)
{
struct bfe_softc *sc;
sc = device_get_softc(dev);
BFE_LOCK(sc);
2004-08-07 20:55:53 +00:00
bfe_stop(sc);
BFE_UNLOCK(sc);
return (0);
}
static int
bfe_suspend(device_t dev)
{
struct bfe_softc *sc;
sc = device_get_softc(dev);
BFE_LOCK(sc);
bfe_stop(sc);
BFE_UNLOCK(sc);
return (0);
}
static int
bfe_resume(device_t dev)
{
struct bfe_softc *sc;
struct ifnet *ifp;
sc = device_get_softc(dev);
ifp = sc->bfe_ifp;
BFE_LOCK(sc);
bfe_chip_reset(sc);
if (ifp->if_flags & IFF_UP) {
bfe_init_locked(sc);
if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
bfe_start_locked(ifp);
}
BFE_UNLOCK(sc);
return (0);
}
static int
bfe_miibus_readreg(device_t dev, int phy, int reg)
{
struct bfe_softc *sc;
u_int32_t ret;
sc = device_get_softc(dev);
bfe_readphy(sc, reg, &ret);
2004-08-07 20:55:53 +00:00
return (ret);
}
static int
bfe_miibus_writereg(device_t dev, int phy, int reg, int val)
{
struct bfe_softc *sc;
sc = device_get_softc(dev);
2004-08-07 20:55:53 +00:00
bfe_writephy(sc, reg, val);
2004-08-07 20:55:53 +00:00
return (0);
}
static void
bfe_miibus_statchg(device_t dev)
{
struct bfe_softc *sc;
struct mii_data *mii;
u_int32_t val, flow;
sc = device_get_softc(dev);
mii = device_get_softc(sc->bfe_miibus);
sc->bfe_flags &= ~BFE_FLAG_LINK;
if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
(IFM_ACTIVE | IFM_AVALID)) {
switch (IFM_SUBTYPE(mii->mii_media_active)) {
case IFM_10_T:
case IFM_100_TX:
sc->bfe_flags |= BFE_FLAG_LINK;
break;
default:
break;
}
}
/* XXX Should stop Rx/Tx engine prior to touching MAC. */
val = CSR_READ_4(sc, BFE_TX_CTRL);
val &= ~BFE_TX_DUPLEX;
if ((IFM_OPTIONS(mii->mii_media_active) & IFM_FDX) != 0) {
val |= BFE_TX_DUPLEX;
flow = 0;
#ifdef notyet
flow = CSR_READ_4(sc, BFE_RXCONF);
flow &= ~BFE_RXCONF_FLOW;
if ((IFM_OPTIONS(sc->sc_mii->mii_media_active) &
IFM_ETH_RXPAUSE) != 0)
flow |= BFE_RXCONF_FLOW;
CSR_WRITE_4(sc, BFE_RXCONF, flow);
/*
* It seems that the hardware has Tx pause issues
* so enable only Rx pause.
*/
flow = CSR_READ_4(sc, BFE_MAC_FLOW);
flow &= ~BFE_FLOW_PAUSE_ENAB;
CSR_WRITE_4(sc, BFE_MAC_FLOW, flow);
#endif
}
CSR_WRITE_4(sc, BFE_TX_CTRL, val);
}
static void
bfe_tx_ring_free(struct bfe_softc *sc)
{
int i;
2004-08-07 20:55:53 +00:00
for(i = 0; i < BFE_TX_LIST_CNT; i++) {
if (sc->bfe_tx_ring[i].bfe_mbuf != NULL) {
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
bus_dmamap_sync(sc->bfe_txmbuf_tag,
sc->bfe_tx_ring[i].bfe_map, BUS_DMASYNC_POSTWRITE);
bus_dmamap_unload(sc->bfe_txmbuf_tag,
sc->bfe_tx_ring[i].bfe_map);
m_freem(sc->bfe_tx_ring[i].bfe_mbuf);
sc->bfe_tx_ring[i].bfe_mbuf = NULL;
}
}
bzero(sc->bfe_tx_list, BFE_TX_LIST_SIZE);
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
bus_dmamap_sync(sc->bfe_tx_tag, sc->bfe_tx_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
}
static void
bfe_rx_ring_free(struct bfe_softc *sc)
{
int i;
for (i = 0; i < BFE_RX_LIST_CNT; i++) {
if (sc->bfe_rx_ring[i].bfe_mbuf != NULL) {
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
bus_dmamap_sync(sc->bfe_rxmbuf_tag,
sc->bfe_rx_ring[i].bfe_map, BUS_DMASYNC_POSTREAD);
bus_dmamap_unload(sc->bfe_rxmbuf_tag,
sc->bfe_rx_ring[i].bfe_map);
m_freem(sc->bfe_rx_ring[i].bfe_mbuf);
sc->bfe_rx_ring[i].bfe_mbuf = NULL;
}
}
bzero(sc->bfe_rx_list, BFE_RX_LIST_SIZE);
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
bus_dmamap_sync(sc->bfe_rx_tag, sc->bfe_rx_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
}
2004-08-07 20:55:53 +00:00
static int
bfe_list_rx_init(struct bfe_softc *sc)
{
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
struct bfe_rx_data *rd;
int i;
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
sc->bfe_rx_prod = sc->bfe_rx_cons = 0;
bzero(sc->bfe_rx_list, BFE_RX_LIST_SIZE);
for (i = 0; i < BFE_RX_LIST_CNT; i++) {
rd = &sc->bfe_rx_ring[i];
rd->bfe_mbuf = NULL;
rd->bfe_ctrl = 0;
if (bfe_list_newbuf(sc, i) != 0)
2004-08-07 20:55:53 +00:00
return (ENOBUFS);
}
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
bus_dmamap_sync(sc->bfe_rx_tag, sc->bfe_rx_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
CSR_WRITE_4(sc, BFE_DMARX_PTR, (i * sizeof(struct bfe_desc)));
2004-08-07 20:55:53 +00:00
return (0);
}
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
static void
bfe_list_tx_init(struct bfe_softc *sc)
{
int i;
sc->bfe_tx_cnt = sc->bfe_tx_prod = sc->bfe_tx_cons = 0;
bzero(sc->bfe_tx_list, BFE_TX_LIST_SIZE);
for (i = 0; i < BFE_TX_LIST_CNT; i++)
sc->bfe_tx_ring[i].bfe_mbuf = NULL;
bus_dmamap_sync(sc->bfe_tx_tag, sc->bfe_tx_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
}
static void
bfe_discard_buf(struct bfe_softc *sc, int c)
{
struct bfe_rx_data *r;
struct bfe_desc *d;
r = &sc->bfe_rx_ring[c];
d = &sc->bfe_rx_list[c];
d->bfe_ctrl = htole32(r->bfe_ctrl);
}
static int
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
bfe_list_newbuf(struct bfe_softc *sc, int c)
{
struct bfe_rxheader *rx_header;
struct bfe_desc *d;
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
struct bfe_rx_data *r;
struct mbuf *m;
bus_dma_segment_t segs[1];
bus_dmamap_t map;
u_int32_t ctrl;
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
int nsegs;
m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
m->m_len = m->m_pkthdr.len = MCLBYTES;
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
if (bus_dmamap_load_mbuf_sg(sc->bfe_rxmbuf_tag, sc->bfe_rx_sparemap,
m, segs, &nsegs, 0) != 0) {
m_freem(m);
return (ENOBUFS);
}
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
KASSERT(nsegs == 1, ("%s: %d segments returned!", __func__, nsegs));
r = &sc->bfe_rx_ring[c];
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
if (r->bfe_mbuf != NULL) {
bus_dmamap_sync(sc->bfe_rxmbuf_tag, r->bfe_map,
BUS_DMASYNC_POSTREAD);
bus_dmamap_unload(sc->bfe_rxmbuf_tag, r->bfe_map);
}
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
map = r->bfe_map;
r->bfe_map = sc->bfe_rx_sparemap;
sc->bfe_rx_sparemap = map;
r->bfe_mbuf = m;
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
rx_header = mtod(m, struct bfe_rxheader *);
rx_header->len = 0;
rx_header->flags = 0;
bus_dmamap_sync(sc->bfe_rxmbuf_tag, r->bfe_map, BUS_DMASYNC_PREREAD);
ctrl = segs[0].ds_len & BFE_DESC_LEN;
KASSERT(ctrl > ETHER_MAX_LEN + 32, ("%s: buffer size too small(%d)!",
__func__, ctrl));
if (c == BFE_RX_LIST_CNT - 1)
ctrl |= BFE_DESC_EOT;
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
r->bfe_ctrl = ctrl;
d = &sc->bfe_rx_list[c];
d->bfe_ctrl = htole32(ctrl);
/* The chip needs all addresses to be added to BFE_PCI_DMA. */
d->bfe_addr = htole32(BFE_ADDR_LO(segs[0].ds_addr) + BFE_PCI_DMA);
2004-08-07 20:55:53 +00:00
return (0);
}
static void
bfe_get_config(struct bfe_softc *sc)
{
u_int8_t eeprom[128];
bfe_read_eeprom(sc, eeprom);
sc->bfe_enaddr[0] = eeprom[79];
sc->bfe_enaddr[1] = eeprom[78];
sc->bfe_enaddr[2] = eeprom[81];
sc->bfe_enaddr[3] = eeprom[80];
sc->bfe_enaddr[4] = eeprom[83];
sc->bfe_enaddr[5] = eeprom[82];
sc->bfe_phyaddr = eeprom[90] & 0x1f;
sc->bfe_mdc_port = (eeprom[90] >> 14) & 0x1;
2004-08-07 20:55:53 +00:00
sc->bfe_core_unit = 0;
sc->bfe_dma_offset = BFE_PCI_DMA;
}
static void
bfe_pci_setup(struct bfe_softc *sc, u_int32_t cores)
{
u_int32_t bar_orig, pci_rev, val;
bar_orig = pci_read_config(sc->bfe_dev, BFE_BAR0_WIN, 4);
pci_write_config(sc->bfe_dev, BFE_BAR0_WIN, BFE_REG_PCI, 4);
pci_rev = CSR_READ_4(sc, BFE_SBIDHIGH) & BFE_RC_MASK;
val = CSR_READ_4(sc, BFE_SBINTVEC);
val |= cores;
CSR_WRITE_4(sc, BFE_SBINTVEC, val);
val = CSR_READ_4(sc, BFE_SSB_PCI_TRANS_2);
val |= BFE_SSB_PCI_PREF | BFE_SSB_PCI_BURST;
CSR_WRITE_4(sc, BFE_SSB_PCI_TRANS_2, val);
pci_write_config(sc->bfe_dev, BFE_BAR0_WIN, bar_orig, 4);
}
2004-08-07 20:55:53 +00:00
static void
bfe_clear_stats(struct bfe_softc *sc)
{
uint32_t reg;
BFE_LOCK_ASSERT(sc);
CSR_WRITE_4(sc, BFE_MIB_CTRL, BFE_MIB_CLR_ON_READ);
for (reg = BFE_TX_GOOD_O; reg <= BFE_TX_PAUSE; reg += 4)
CSR_READ_4(sc, reg);
for (reg = BFE_RX_GOOD_O; reg <= BFE_RX_NPAUSE; reg += 4)
CSR_READ_4(sc, reg);
}
2004-08-07 20:55:53 +00:00
static int
bfe_resetphy(struct bfe_softc *sc)
{
u_int32_t val;
bfe_writephy(sc, 0, BMCR_RESET);
DELAY(100);
bfe_readphy(sc, 0, &val);
if (val & BMCR_RESET) {
device_printf(sc->bfe_dev, "PHY Reset would not complete.\n");
2004-08-07 20:55:53 +00:00
return (ENXIO);
}
2004-08-07 20:55:53 +00:00
return (0);
}
static void
bfe_chip_halt(struct bfe_softc *sc)
{
BFE_LOCK_ASSERT(sc);
/* disable interrupts - not that it actually does..*/
CSR_WRITE_4(sc, BFE_IMASK, 0);
CSR_READ_4(sc, BFE_IMASK);
CSR_WRITE_4(sc, BFE_ENET_CTRL, BFE_ENET_DISABLE);
bfe_wait_bit(sc, BFE_ENET_CTRL, BFE_ENET_DISABLE, 200, 1);
CSR_WRITE_4(sc, BFE_DMARX_CTRL, 0);
CSR_WRITE_4(sc, BFE_DMATX_CTRL, 0);
DELAY(10);
}
static void
bfe_chip_reset(struct bfe_softc *sc)
{
2004-08-07 20:55:53 +00:00
u_int32_t val;
BFE_LOCK_ASSERT(sc);
/* Set the interrupt vector for the enet core */
bfe_pci_setup(sc, BFE_INTVEC_ENET0);
/* is core up? */
val = CSR_READ_4(sc, BFE_SBTMSLOW) &
(BFE_RESET | BFE_REJECT | BFE_CLOCK);
if (val == BFE_CLOCK) {
/* It is, so shut it down */
CSR_WRITE_4(sc, BFE_RCV_LAZY, 0);
CSR_WRITE_4(sc, BFE_ENET_CTRL, BFE_ENET_DISABLE);
bfe_wait_bit(sc, BFE_ENET_CTRL, BFE_ENET_DISABLE, 100, 1);
CSR_WRITE_4(sc, BFE_DMATX_CTRL, 0);
2004-08-07 20:55:53 +00:00
if (CSR_READ_4(sc, BFE_DMARX_STAT) & BFE_STAT_EMASK)
bfe_wait_bit(sc, BFE_DMARX_STAT, BFE_STAT_SIDLE,
100, 0);
CSR_WRITE_4(sc, BFE_DMARX_CTRL, 0);
}
bfe_core_reset(sc);
bfe_clear_stats(sc);
/*
* We want the phy registers to be accessible even when
* the driver is "downed" so initialize MDC preamble, frequency,
* and whether internal or external phy here.
*/
/* 4402 has 62.5Mhz SB clock and internal phy */
CSR_WRITE_4(sc, BFE_MDIO_CTRL, 0x8d);
/* Internal or external PHY? */
val = CSR_READ_4(sc, BFE_DEVCTRL);
if (!(val & BFE_IPP))
CSR_WRITE_4(sc, BFE_ENET_CTRL, BFE_ENET_EPSEL);
else if (CSR_READ_4(sc, BFE_DEVCTRL) & BFE_EPR) {
BFE_AND(sc, BFE_DEVCTRL, ~BFE_EPR);
DELAY(100);
}
2004-08-07 20:55:53 +00:00
/* Enable CRC32 generation and set proper LED modes */
BFE_OR(sc, BFE_MAC_CTRL, BFE_CTRL_CRC32_ENAB | BFE_CTRL_LED);
2004-08-07 20:55:53 +00:00
/* Reset or clear powerdown control bit */
BFE_AND(sc, BFE_MAC_CTRL, ~BFE_CTRL_PDOWN);
2004-08-07 20:55:53 +00:00
CSR_WRITE_4(sc, BFE_RCV_LAZY, ((1 << BFE_LAZY_FC_SHIFT) &
BFE_LAZY_FC_MASK));
2004-08-07 20:55:53 +00:00
/*
* We don't want lazy interrupts, so just send them at
2004-08-07 20:55:53 +00:00
* the end of a frame, please
*/
BFE_OR(sc, BFE_RCV_LAZY, 0);
/* Set max lengths, accounting for VLAN tags */
CSR_WRITE_4(sc, BFE_RXMAXLEN, ETHER_MAX_LEN+32);
CSR_WRITE_4(sc, BFE_TXMAXLEN, ETHER_MAX_LEN+32);
/* Set watermark XXX - magic */
CSR_WRITE_4(sc, BFE_TX_WMARK, 56);
2004-08-07 20:55:53 +00:00
/*
* Initialise DMA channels
2004-08-07 20:55:53 +00:00
* - not forgetting dma addresses need to be added to BFE_PCI_DMA
*/
CSR_WRITE_4(sc, BFE_DMATX_CTRL, BFE_TX_CTRL_ENABLE);
CSR_WRITE_4(sc, BFE_DMATX_ADDR, sc->bfe_tx_dma + BFE_PCI_DMA);
2004-08-07 20:55:53 +00:00
CSR_WRITE_4(sc, BFE_DMARX_CTRL, (BFE_RX_OFFSET << BFE_RX_CTRL_ROSHIFT) |
BFE_RX_CTRL_ENABLE);
CSR_WRITE_4(sc, BFE_DMARX_ADDR, sc->bfe_rx_dma + BFE_PCI_DMA);
bfe_resetphy(sc);
bfe_setupphy(sc);
}
static void
bfe_core_disable(struct bfe_softc *sc)
{
if ((CSR_READ_4(sc, BFE_SBTMSLOW)) & BFE_RESET)
return;
2004-08-07 20:55:53 +00:00
/*
* Set reject, wait for it set, then wait for the core to stop
* being busy, then set reset and reject and enable the clocks.
*/
CSR_WRITE_4(sc, BFE_SBTMSLOW, (BFE_REJECT | BFE_CLOCK));
bfe_wait_bit(sc, BFE_SBTMSLOW, BFE_REJECT, 1000, 0);
bfe_wait_bit(sc, BFE_SBTMSHIGH, BFE_BUSY, 1000, 1);
CSR_WRITE_4(sc, BFE_SBTMSLOW, (BFE_FGC | BFE_CLOCK | BFE_REJECT |
BFE_RESET));
CSR_READ_4(sc, BFE_SBTMSLOW);
DELAY(10);
/* Leave reset and reject set */
CSR_WRITE_4(sc, BFE_SBTMSLOW, (BFE_REJECT | BFE_RESET));
DELAY(10);
}
static void
bfe_core_reset(struct bfe_softc *sc)
{
u_int32_t val;
/* Disable the core */
bfe_core_disable(sc);
/* and bring it back up */
CSR_WRITE_4(sc, BFE_SBTMSLOW, (BFE_RESET | BFE_CLOCK | BFE_FGC));
CSR_READ_4(sc, BFE_SBTMSLOW);
DELAY(10);
/* Chip bug, clear SERR, IB and TO if they are set. */
if (CSR_READ_4(sc, BFE_SBTMSHIGH) & BFE_SERR)
CSR_WRITE_4(sc, BFE_SBTMSHIGH, 0);
val = CSR_READ_4(sc, BFE_SBIMSTATE);
if (val & (BFE_IBE | BFE_TO))
CSR_WRITE_4(sc, BFE_SBIMSTATE, val & ~(BFE_IBE | BFE_TO));
/* Clear reset and allow it to move through the core */
CSR_WRITE_4(sc, BFE_SBTMSLOW, (BFE_CLOCK | BFE_FGC));
CSR_READ_4(sc, BFE_SBTMSLOW);
DELAY(10);
/* Leave the clock set */
CSR_WRITE_4(sc, BFE_SBTMSLOW, BFE_CLOCK);
CSR_READ_4(sc, BFE_SBTMSLOW);
DELAY(10);
}
2004-08-07 20:55:53 +00:00
static void
bfe_cam_write(struct bfe_softc *sc, u_char *data, int index)
{
u_int32_t val;
val = ((u_int32_t) data[2]) << 24;
val |= ((u_int32_t) data[3]) << 16;
val |= ((u_int32_t) data[4]) << 8;
val |= ((u_int32_t) data[5]);
CSR_WRITE_4(sc, BFE_CAM_DATA_LO, val);
val = (BFE_CAM_HI_VALID |
(((u_int32_t) data[0]) << 8) |
(((u_int32_t) data[1])));
CSR_WRITE_4(sc, BFE_CAM_DATA_HI, val);
CSR_WRITE_4(sc, BFE_CAM_CTRL, (BFE_CAM_WRITE |
((u_int32_t) index << BFE_CAM_INDEX_SHIFT)));
bfe_wait_bit(sc, BFE_CAM_CTRL, BFE_CAM_BUSY, 10000, 1);
}
2004-08-07 20:55:53 +00:00
static void
bfe_set_rx_mode(struct bfe_softc *sc)
{
struct ifnet *ifp = sc->bfe_ifp;
struct ifmultiaddr *ifma;
u_int32_t val;
int i = 0;
BFE_LOCK_ASSERT(sc);
val = CSR_READ_4(sc, BFE_RXCONF);
if (ifp->if_flags & IFF_PROMISC)
val |= BFE_RXCONF_PROMISC;
else
val &= ~BFE_RXCONF_PROMISC;
if (ifp->if_flags & IFF_BROADCAST)
val &= ~BFE_RXCONF_DBCAST;
else
val |= BFE_RXCONF_DBCAST;
CSR_WRITE_4(sc, BFE_CAM_CTRL, 0);
bfe_cam_write(sc, IF_LLADDR(sc->bfe_ifp), i++);
if (ifp->if_flags & IFF_ALLMULTI)
val |= BFE_RXCONF_ALLMULTI;
else {
val &= ~BFE_RXCONF_ALLMULTI;
if_maddr_rlock(ifp);
TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
if (ifma->ifma_addr->sa_family != AF_LINK)
continue;
bfe_cam_write(sc,
LLADDR((struct sockaddr_dl *)ifma->ifma_addr), i++);
}
if_maddr_runlock(ifp);
}
CSR_WRITE_4(sc, BFE_RXCONF, val);
BFE_OR(sc, BFE_CAM_CTRL, BFE_CAM_ENABLE);
}
static void
bfe_dma_map(void *arg, bus_dma_segment_t *segs, int nseg, int error)
{
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
struct bfe_dmamap_arg *ctx;
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
if (error != 0)
return;
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
KASSERT(nseg == 1, ("%s : %d segments returned!", __func__, nseg));
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
ctx = (struct bfe_dmamap_arg *)arg;
ctx->bfe_busaddr = segs[0].ds_addr;
}
static void
bfe_release_resources(struct bfe_softc *sc)
{
if (sc->bfe_intrhand != NULL)
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
bus_teardown_intr(sc->bfe_dev, sc->bfe_irq, sc->bfe_intrhand);
if (sc->bfe_irq != NULL)
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
bus_release_resource(sc->bfe_dev, SYS_RES_IRQ, 0, sc->bfe_irq);
if (sc->bfe_res != NULL)
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
bus_release_resource(sc->bfe_dev, SYS_RES_MEMORY, PCIR_BAR(0),
sc->bfe_res);
if (sc->bfe_ifp != NULL)
if_free(sc->bfe_ifp);
}
static void
bfe_read_eeprom(struct bfe_softc *sc, u_int8_t *data)
{
long i;
u_int16_t *ptr = (u_int16_t *)data;
for(i = 0; i < 128; i += 2)
ptr[i/2] = CSR_READ_4(sc, 4096 + i);
}
static int
2004-08-07 20:55:53 +00:00
bfe_wait_bit(struct bfe_softc *sc, u_int32_t reg, u_int32_t bit,
u_long timeout, const int clear)
{
u_long i;
for (i = 0; i < timeout; i++) {
u_int32_t val = CSR_READ_4(sc, reg);
if (clear && !(val & bit))
break;
if (!clear && (val & bit))
break;
DELAY(10);
}
if (i == timeout) {
device_printf(sc->bfe_dev,
"BUG! Timeout waiting for bit %08x of register "
"%x to %s.\n", bit, reg, (clear ? "clear" : "set"));
2004-08-07 20:55:53 +00:00
return (-1);
}
2004-08-07 20:55:53 +00:00
return (0);
}
static int
bfe_readphy(struct bfe_softc *sc, u_int32_t reg, u_int32_t *val)
{
2004-08-07 20:55:53 +00:00
int err;
/* Clear MII ISR */
CSR_WRITE_4(sc, BFE_EMAC_ISTAT, BFE_EMAC_INT_MII);
CSR_WRITE_4(sc, BFE_MDIO_DATA, (BFE_MDIO_SB_START |
(BFE_MDIO_OP_READ << BFE_MDIO_OP_SHIFT) |
(sc->bfe_phyaddr << BFE_MDIO_PMD_SHIFT) |
(reg << BFE_MDIO_RA_SHIFT) |
(BFE_MDIO_TA_VALID << BFE_MDIO_TA_SHIFT)));
err = bfe_wait_bit(sc, BFE_EMAC_ISTAT, BFE_EMAC_INT_MII, 100, 0);
*val = CSR_READ_4(sc, BFE_MDIO_DATA) & BFE_MDIO_DATA_DATA;
2004-08-07 20:55:53 +00:00
return (err);
}
static int
bfe_writephy(struct bfe_softc *sc, u_int32_t reg, u_int32_t val)
{
int status;
CSR_WRITE_4(sc, BFE_EMAC_ISTAT, BFE_EMAC_INT_MII);
CSR_WRITE_4(sc, BFE_MDIO_DATA, (BFE_MDIO_SB_START |
(BFE_MDIO_OP_WRITE << BFE_MDIO_OP_SHIFT) |
(sc->bfe_phyaddr << BFE_MDIO_PMD_SHIFT) |
(reg << BFE_MDIO_RA_SHIFT) |
(BFE_MDIO_TA_VALID << BFE_MDIO_TA_SHIFT) |
(val & BFE_MDIO_DATA_DATA)));
status = bfe_wait_bit(sc, BFE_EMAC_ISTAT, BFE_EMAC_INT_MII, 100, 0);
2004-08-07 20:55:53 +00:00
return (status);
}
2004-08-07 20:55:53 +00:00
/*
* XXX - I think this is handled by the PHY driver, but it can't hurt to do it
* twice
*/
static int
bfe_setupphy(struct bfe_softc *sc)
{
u_int32_t val;
/* Enable activity LED */
bfe_readphy(sc, 26, &val);
2004-08-07 20:55:53 +00:00
bfe_writephy(sc, 26, val & 0x7fff);
bfe_readphy(sc, 26, &val);
/* Enable traffic meter LED mode */
bfe_readphy(sc, 27, &val);
bfe_writephy(sc, 27, val | (1 << 6));
2004-08-07 20:55:53 +00:00
return (0);
}
2004-08-07 20:55:53 +00:00
static void
bfe_stats_update(struct bfe_softc *sc)
{
struct bfe_hw_stats *stats;
struct ifnet *ifp;
uint32_t mib[BFE_MIB_CNT];
uint32_t reg, *val;
BFE_LOCK_ASSERT(sc);
val = mib;
CSR_WRITE_4(sc, BFE_MIB_CTRL, BFE_MIB_CLR_ON_READ);
for (reg = BFE_TX_GOOD_O; reg <= BFE_TX_PAUSE; reg += 4)
*val++ = CSR_READ_4(sc, reg);
for (reg = BFE_RX_GOOD_O; reg <= BFE_RX_NPAUSE; reg += 4)
*val++ = CSR_READ_4(sc, reg);
ifp = sc->bfe_ifp;
stats = &sc->bfe_stats;
/* Tx stat. */
stats->tx_good_octets += mib[MIB_TX_GOOD_O];
stats->tx_good_frames += mib[MIB_TX_GOOD_P];
stats->tx_octets += mib[MIB_TX_O];
stats->tx_frames += mib[MIB_TX_P];
stats->tx_bcast_frames += mib[MIB_TX_BCAST];
stats->tx_mcast_frames += mib[MIB_TX_MCAST];
stats->tx_pkts_64 += mib[MIB_TX_64];
stats->tx_pkts_65_127 += mib[MIB_TX_65_127];
stats->tx_pkts_128_255 += mib[MIB_TX_128_255];
stats->tx_pkts_256_511 += mib[MIB_TX_256_511];
stats->tx_pkts_512_1023 += mib[MIB_TX_512_1023];
stats->tx_pkts_1024_max += mib[MIB_TX_1024_MAX];
stats->tx_jabbers += mib[MIB_TX_JABBER];
stats->tx_oversize_frames += mib[MIB_TX_OSIZE];
stats->tx_frag_frames += mib[MIB_TX_FRAG];
stats->tx_underruns += mib[MIB_TX_URUNS];
stats->tx_colls += mib[MIB_TX_TCOLS];
stats->tx_single_colls += mib[MIB_TX_SCOLS];
stats->tx_multi_colls += mib[MIB_TX_MCOLS];
stats->tx_excess_colls += mib[MIB_TX_ECOLS];
stats->tx_late_colls += mib[MIB_TX_LCOLS];
stats->tx_deferrals += mib[MIB_TX_DEFERED];
stats->tx_carrier_losts += mib[MIB_TX_CLOST];
stats->tx_pause_frames += mib[MIB_TX_PAUSE];
/* Rx stat. */
stats->rx_good_octets += mib[MIB_RX_GOOD_O];
stats->rx_good_frames += mib[MIB_RX_GOOD_P];
stats->rx_octets += mib[MIB_RX_O];
stats->rx_frames += mib[MIB_RX_P];
stats->rx_bcast_frames += mib[MIB_RX_BCAST];
stats->rx_mcast_frames += mib[MIB_RX_MCAST];
stats->rx_pkts_64 += mib[MIB_RX_64];
stats->rx_pkts_65_127 += mib[MIB_RX_65_127];
stats->rx_pkts_128_255 += mib[MIB_RX_128_255];
stats->rx_pkts_256_511 += mib[MIB_RX_256_511];
stats->rx_pkts_512_1023 += mib[MIB_RX_512_1023];
stats->rx_pkts_1024_max += mib[MIB_RX_1024_MAX];
stats->rx_jabbers += mib[MIB_RX_JABBER];
stats->rx_oversize_frames += mib[MIB_RX_OSIZE];
stats->rx_frag_frames += mib[MIB_RX_FRAG];
stats->rx_missed_frames += mib[MIB_RX_MISS];
stats->rx_crc_align_errs += mib[MIB_RX_CRCA];
stats->rx_runts += mib[MIB_RX_USIZE];
stats->rx_crc_errs += mib[MIB_RX_CRC];
stats->rx_align_errs += mib[MIB_RX_ALIGN];
stats->rx_symbol_errs += mib[MIB_RX_SYM];
stats->rx_pause_frames += mib[MIB_RX_PAUSE];
stats->rx_control_frames += mib[MIB_RX_NPAUSE];
/* Update counters in ifnet. */
ifp->if_opackets += (u_long)mib[MIB_TX_GOOD_P];
ifp->if_collisions += (u_long)mib[MIB_TX_TCOLS];
ifp->if_oerrors += (u_long)mib[MIB_TX_URUNS] +
(u_long)mib[MIB_TX_ECOLS] +
(u_long)mib[MIB_TX_DEFERED] +
(u_long)mib[MIB_TX_CLOST];
ifp->if_ipackets += (u_long)mib[MIB_RX_GOOD_P];
ifp->if_ierrors += mib[MIB_RX_JABBER] +
mib[MIB_RX_MISS] +
mib[MIB_RX_CRCA] +
mib[MIB_RX_USIZE] +
mib[MIB_RX_CRC] +
mib[MIB_RX_ALIGN] +
mib[MIB_RX_SYM];
}
static void
bfe_txeof(struct bfe_softc *sc)
{
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
struct bfe_tx_data *r;
struct ifnet *ifp;
int i, chipidx;
BFE_LOCK_ASSERT(sc);
ifp = sc->bfe_ifp;
chipidx = CSR_READ_4(sc, BFE_DMATX_STAT) & BFE_STAT_CDMASK;
chipidx /= sizeof(struct bfe_desc);
i = sc->bfe_tx_cons;
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
if (i == chipidx)
return;
bus_dmamap_sync(sc->bfe_tx_tag, sc->bfe_tx_map,
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
/* Go through the mbufs and free those that have been transmitted */
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
for (; i != chipidx; BFE_INC(i, BFE_TX_LIST_CNT)) {
r = &sc->bfe_tx_ring[i];
sc->bfe_tx_cnt--;
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
if (r->bfe_mbuf == NULL)
continue;
bus_dmamap_sync(sc->bfe_txmbuf_tag, r->bfe_map,
BUS_DMASYNC_POSTWRITE);
bus_dmamap_unload(sc->bfe_txmbuf_tag, r->bfe_map);
m_freem(r->bfe_mbuf);
r->bfe_mbuf = NULL;
}
if (i != sc->bfe_tx_cons) {
/* we freed up some mbufs */
sc->bfe_tx_cons = i;
ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
}
if (sc->bfe_tx_cnt == 0)
sc->bfe_watchdog_timer = 0;
}
/* Pass a received packet up the stack */
static void
bfe_rxeof(struct bfe_softc *sc)
{
struct mbuf *m;
struct ifnet *ifp;
struct bfe_rxheader *rxheader;
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
struct bfe_rx_data *r;
int cons, prog;
u_int32_t status, current, len, flags;
BFE_LOCK_ASSERT(sc);
cons = sc->bfe_rx_cons;
status = CSR_READ_4(sc, BFE_DMARX_STAT);
current = (status & BFE_STAT_CDMASK) / sizeof(struct bfe_desc);
ifp = sc->bfe_ifp;
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
bus_dmamap_sync(sc->bfe_rx_tag, sc->bfe_rx_map,
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
for (prog = 0; current != cons; prog++,
BFE_INC(cons, BFE_RX_LIST_CNT)) {
r = &sc->bfe_rx_ring[cons];
m = r->bfe_mbuf;
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
/*
* Rx status should be read from mbuf such that we can't
* delay bus_dmamap_sync(9). This hardware limiation
* results in inefficent mbuf usage as bfe(4) couldn't
* reuse mapped buffer from errored frame.
*/
if (bfe_list_newbuf(sc, cons) != 0) {
ifp->if_iqdrops++;
bfe_discard_buf(sc, cons);
continue;
}
rxheader = mtod(m, struct bfe_rxheader*);
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
len = le16toh(rxheader->len);
flags = le16toh(rxheader->flags);
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
/* Remove CRC bytes. */
len -= ETHER_CRC_LEN;
/* flag an error and try again */
if ((len > ETHER_MAX_LEN+32) || (flags & BFE_RX_FLAG_ERRORS)) {
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
m_freem(m);
continue;
}
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
/* Make sure to skip header bytes written by hardware. */
m_adj(m, BFE_RX_OFFSET);
m->m_len = m->m_pkthdr.len = len;
m->m_pkthdr.rcvif = ifp;
BFE_UNLOCK(sc);
(*ifp->if_input)(ifp, m);
BFE_LOCK(sc);
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
}
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
if (prog > 0) {
sc->bfe_rx_cons = cons;
bus_dmamap_sync(sc->bfe_rx_tag, sc->bfe_rx_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
}
}
static void
bfe_intr(void *xsc)
{
struct bfe_softc *sc = xsc;
struct ifnet *ifp;
u_int32_t istat;
ifp = sc->bfe_ifp;
BFE_LOCK(sc);
istat = CSR_READ_4(sc, BFE_ISTAT);
2004-08-07 20:55:53 +00:00
/*
* Defer unsolicited interrupts - This is necessary because setting the
* chips interrupt mask register to 0 doesn't actually stop the
* interrupts
*/
istat &= BFE_IMASK_DEF;
CSR_WRITE_4(sc, BFE_ISTAT, istat);
CSR_READ_4(sc, BFE_ISTAT);
/* not expecting this interrupt, disregard it */
if (istat == 0 || (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
BFE_UNLOCK(sc);
return;
}
/* A packet was received */
if (istat & BFE_ISTAT_RX)
bfe_rxeof(sc);
/* A packet was sent */
if (istat & BFE_ISTAT_TX)
bfe_txeof(sc);
if (istat & BFE_ISTAT_ERRORS) {
if (istat & BFE_ISTAT_DSCE) {
device_printf(sc->bfe_dev, "Descriptor Error\n");
bfe_stop(sc);
BFE_UNLOCK(sc);
return;
}
if (istat & BFE_ISTAT_DPE) {
device_printf(sc->bfe_dev,
"Descriptor Protocol Error\n");
bfe_stop(sc);
BFE_UNLOCK(sc);
return;
}
ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
bfe_init_locked(sc);
}
2004-08-07 20:55:53 +00:00
/* We have packets pending, fire them out */
if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
bfe_start_locked(ifp);
BFE_UNLOCK(sc);
}
static int
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
bfe_encap(struct bfe_softc *sc, struct mbuf **m_head)
{
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
struct bfe_desc *d;
struct bfe_tx_data *r, *r1;
struct mbuf *m;
bus_dmamap_t map;
bus_dma_segment_t txsegs[BFE_MAXTXSEGS];
uint32_t cur, si;
int error, i, nsegs;
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
BFE_LOCK_ASSERT(sc);
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
M_ASSERTPKTHDR((*m_head));
si = cur = sc->bfe_tx_prod;
r = &sc->bfe_tx_ring[cur];
error = bus_dmamap_load_mbuf_sg(sc->bfe_txmbuf_tag, r->bfe_map, *m_head,
txsegs, &nsegs, 0);
if (error == EFBIG) {
m = m_collapse(*m_head, M_NOWAIT, BFE_MAXTXSEGS);
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
if (m == NULL) {
m_freem(*m_head);
*m_head = NULL;
return (ENOMEM);
}
*m_head = m;
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
error = bus_dmamap_load_mbuf_sg(sc->bfe_txmbuf_tag, r->bfe_map,
*m_head, txsegs, &nsegs, 0);
if (error != 0) {
m_freem(*m_head);
*m_head = NULL;
return (error);
}
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
} else if (error != 0)
return (error);
if (nsegs == 0) {
m_freem(*m_head);
*m_head = NULL;
return (EIO);
}
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
if (sc->bfe_tx_cnt + nsegs > BFE_TX_LIST_CNT - 1) {
bus_dmamap_unload(sc->bfe_txmbuf_tag, r->bfe_map);
2004-08-07 20:55:53 +00:00
return (ENOBUFS);
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
}
for (i = 0; i < nsegs; i++) {
d = &sc->bfe_tx_list[cur];
d->bfe_ctrl = htole32(txsegs[i].ds_len & BFE_DESC_LEN);
d->bfe_ctrl |= htole32(BFE_DESC_IOC);
if (cur == BFE_TX_LIST_CNT - 1)
/*
* Tell the chip to wrap to the start of
* the descriptor list.
*/
d->bfe_ctrl |= htole32(BFE_DESC_EOT);
/* The chip needs all addresses to be added to BFE_PCI_DMA. */
d->bfe_addr = htole32(BFE_ADDR_LO(txsegs[i].ds_addr) +
BFE_PCI_DMA);
BFE_INC(cur, BFE_TX_LIST_CNT);
}
/* Update producer index. */
sc->bfe_tx_prod = cur;
/* Set EOF on the last descriptor. */
cur = (cur + BFE_TX_LIST_CNT - 1) % BFE_TX_LIST_CNT;
d = &sc->bfe_tx_list[cur];
d->bfe_ctrl |= htole32(BFE_DESC_EOF);
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
/* Lastly set SOF on the first descriptor to avoid races. */
d = &sc->bfe_tx_list[si];
d->bfe_ctrl |= htole32(BFE_DESC_SOF);
r1 = &sc->bfe_tx_ring[cur];
map = r->bfe_map;
r->bfe_map = r1->bfe_map;
r1->bfe_map = map;
r1->bfe_mbuf = *m_head;
sc->bfe_tx_cnt += nsegs;
bus_dmamap_sync(sc->bfe_txmbuf_tag, map, BUS_DMASYNC_PREWRITE);
return (0);
}
/*
* Set up to transmit a packet.
*/
static void
bfe_start(struct ifnet *ifp)
{
BFE_LOCK((struct bfe_softc *)ifp->if_softc);
bfe_start_locked(ifp);
BFE_UNLOCK((struct bfe_softc *)ifp->if_softc);
}
/*
* Set up to transmit a packet. The softc is already locked.
*/
static void
bfe_start_locked(struct ifnet *ifp)
{
struct bfe_softc *sc;
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
struct mbuf *m_head;
int queued;
sc = ifp->if_softc;
BFE_LOCK_ASSERT(sc);
2004-08-07 20:55:53 +00:00
/*
* Not much point trying to send if the link is down
* or we have nothing to send.
*/
if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
IFF_DRV_RUNNING || (sc->bfe_flags & BFE_FLAG_LINK) == 0)
return;
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
for (queued = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) &&
sc->bfe_tx_cnt < BFE_TX_LIST_CNT - 1;) {
IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
if (m_head == NULL)
break;
2004-08-07 20:55:53 +00:00
/*
* Pack the data into the tx ring. If we dont have
* enough room, let the chip drain the ring.
*/
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
if (bfe_encap(sc, &m_head)) {
if (m_head == NULL)
break;
IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
ifp->if_drv_flags |= IFF_DRV_OACTIVE;
break;
}
queued++;
/*
* If there's a BPF listener, bounce a copy of this frame
* to him.
*/
BPF_MTAP(ifp, m_head);
}
if (queued) {
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
bus_dmamap_sync(sc->bfe_tx_tag, sc->bfe_tx_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
/* Transmit - twice due to apparent hardware bug */
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
CSR_WRITE_4(sc, BFE_DMATX_PTR,
sc->bfe_tx_prod * sizeof(struct bfe_desc));
/*
* XXX It seems the following write is not necessary
* to kick Tx command. What might be required would be
* a way flushing PCI posted write. Reading the register
* back ensures the flush operation. In addition,
* hardware will execute PCI posted write in the long
* run and watchdog timer for the kick command was set
* to 5 seconds. Therefore I think the second write
* access is not necessary or could be replaced with
* read operation.
*/
CSR_WRITE_4(sc, BFE_DMATX_PTR,
sc->bfe_tx_prod * sizeof(struct bfe_desc));
/*
* Set a timeout in case the chip goes out to lunch.
*/
sc->bfe_watchdog_timer = 5;
}
}
static void
bfe_init(void *xsc)
{
BFE_LOCK((struct bfe_softc *)xsc);
bfe_init_locked(xsc);
BFE_UNLOCK((struct bfe_softc *)xsc);
}
static void
bfe_init_locked(void *xsc)
{
struct bfe_softc *sc = (struct bfe_softc*)xsc;
struct ifnet *ifp = sc->bfe_ifp;
struct mii_data *mii;
BFE_LOCK_ASSERT(sc);
mii = device_get_softc(sc->bfe_miibus);
if (ifp->if_drv_flags & IFF_DRV_RUNNING)
return;
bfe_stop(sc);
bfe_chip_reset(sc);
if (bfe_list_rx_init(sc) == ENOBUFS) {
device_printf(sc->bfe_dev,
"%s: Not enough memory for list buffers\n", __func__);
bfe_stop(sc);
return;
}
o Sort includes and add <endian.h> to support endianness. o Removed unneeded header files. o bus_dma(9) fix: - created parent tag with 1GB dma address limit with no alignment restrictions. - set 4096 alignment limit for Tx/Rx descriptor rings. - separate Rx buffer tag from Tx buffer tag such that Tx tag allows up-to 16 segments while Rx buffer tag only allows single segment. - it seems the controller has no alignment restrictions on Tx/Rx buffers. Remove ETHER_ALIGN alignment restriction in Tx/Rx buffers. - created a spare Rx dma map which would be used to cope with failure of loading a dma map. - make sure to load full Tx/Rx descriptor size for Tx/Rx descriptor dma maps, previously bfe(4) used to load single descriptor size for each descriptor rings. I have no idea how it could be run without problems. - don't blindly cast bus_addr_t type to 32bits in bfe_dma_map(). - created bfe_dma_free() to free allocated dma memory/tags. - make sure to invoke bus_dmamap_sync(9) before/after processing descriptor rings/buffers. Because the hardware has severe dma address space limitation, bounce-buffers would be always used on systems with more than 1GB memory during descriptors/buffers access. - added Tx descriptor ring initialization function, bfe_list_tx_init(). - moved producer/consumer index initialization to bfe_list_tx_init() and bfe_list_rx_init() from bfe_chip_reset(). - added bfe_discard_buf() which will update loaded descriptors without unloading/reloading the dma map to speed up error recovery. - implemented Tx side bus_dmamap_load_mbuf_sg(9). The number of segments allowed was chosen to be 16 which should be enough for non-TSO capable hardwares. Setting SOF bit of Tx descriptor is done in the last to avoid potential race. - don't give up sending frames in bfe_start() until the hardware lacks free descriptors. - added XXX comment to second kick command and possible workaround. - implemented Rx side bus_dmamap_load_mbuf_sg(9). - removed bfe_dma_map_desc() as it's not needed anymore after the conversion to bus_dmamap_load_mbuf_sg(9). - added endianness support. With this change bfe(4) should work on any architectures that can create bounce buffers within 1GB address range. - add missing bus_dmamap_sync() in bfe_tx_eof()/bfe_rx_eof(). o Use PCI_BAR instead of hardcoded value to set BARs. Simplified register access with bus_write_4(9)/bus_read_4(9) and removed bfe_btag, bfe_bhandle, bfe_vhandle in softc as it's not used anymore. o Reorder device detach logic such that bfe_detach() is also used for handling driver attach failure case. o Remove unnecessary KASSERT in bfe_detach(). o Remove bfe_rx_cnt, bfe_up, bfe_vpd_prodname, bfe_vpd_readonly in softc. It's not used at all. o Remove BFE_RX_RING_SIZE/BFE_RX_RING_SIZE/BFE_LINK_DOWN. Tested by: kib, Gleb Kurtsou gleb.kurtsou at gmail dot com Ulrich Spoerlein uspoerlein at gmail dot com
2008-08-21 04:21:53 +00:00
bfe_list_tx_init(sc);
bfe_set_rx_mode(sc);
/* Enable the chip and core */
BFE_OR(sc, BFE_ENET_CTRL, BFE_ENET_ENABLE);
/* Enable interrupts */
CSR_WRITE_4(sc, BFE_IMASK, BFE_IMASK_DEF);
/* Clear link state and change media. */
sc->bfe_flags &= ~BFE_FLAG_LINK;
mii_mediachg(mii);
ifp->if_drv_flags |= IFF_DRV_RUNNING;
ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
callout_reset(&sc->bfe_stat_co, hz, bfe_tick, sc);
}
/*
* Set media options.
*/
static int
bfe_ifmedia_upd(struct ifnet *ifp)
{
struct bfe_softc *sc;
struct mii_data *mii;
- Remove attempts to implement setting of BMCR_LOOP/MIIF_NOLOOP (reporting IFM_LOOP based on BMCR_LOOP is left in place though as it might provide useful for debugging). For most mii(4) drivers it was unclear whether the PHYs driven by them actually support loopback or not. Moreover, typically loopback mode also needs to be activated on the MAC, which none of the Ethernet drivers using mii(4) implements. Given that loopback media has no real use (and obviously hardly had a chance to actually work) besides for driver development (which just loopback mode should be sufficient for though, i.e one doesn't necessary need support for loopback media) support for it is just dropped as both NetBSD and OpenBSD already did quite some time ago. - Let mii_phy_add_media() also announce the support of IFM_NONE. - Restructure the PHY entry points to use a structure of entry points instead of discrete function pointers, and extend this to include a "reset" entry point. Make sure any PHY-specific reset routine is always used, and provide one for lxtphy(4) which disables MII interrupts (as is done for a few other PHYs we have drivers for). This includes changing NIC drivers which previously just called the generic mii_phy_reset() to now actually call the PHY-specific reset routine, which might be crucial in some cases. While at it, the redundant checks in these NIC drivers for mii->mii_instance not being zero before calling the reset routines were removed because as soon as one PHY driver attaches mii->mii_instance is incremented and we hardly can end up in their media change callbacks etc if no PHY driver has attached as mii_attach() would have failed in that case and not attach a miibus(4) instance. Consequently, NIC drivers now no longer should call mii_phy_reset() directly, so it was removed from EXPORT_SYMS. - Add a mii_phy_dev_attach() as a companion helper to mii_phy_dev_probe(). The purpose of that function is to perform the common steps to attach a PHY driver instance and to hook it up to the miibus(4) instance and to optionally also handle the probing, addition and initialization of the supported media. So all a PHY driver without any special requirements has to do in its bus attach method is to call mii_phy_dev_attach() along with PHY-specific MIIF_* flags, a pointer to its PHY functions and the add_media set to one. All PHY drivers were updated to take advantage of mii_phy_dev_attach() as appropriate. Along with these changes the capability mask was added to the mii_softc structure so PHY drivers taking advantage of mii_phy_dev_attach() but still handling media on their own do not need to fiddle with the MII attach arguments anyway. - Keep track of the PHY offset in the mii_softc structure. This is done for compatibility with NetBSD/OpenBSD. - Keep track of the PHY's OUI, model and revision in the mii_softc structure. Several PHY drivers require this information also after attaching and previously had to wrap their own softc around mii_softc. NetBSD/OpenBSD also keep track of the model and revision on their mii_softc structure. All PHY drivers were updated to take advantage as appropriate. - Convert the mebers of the MII data structure to unsigned where appropriate. This is partly inspired by NetBSD/OpenBSD. - According to IEEE 802.3-2002 the bits actually have to be reversed when mapping an OUI to the MII ID registers. All PHY drivers and miidevs where changed as necessary. Actually this now again allows to largely share miidevs with NetBSD, which fixed this problem already 9 years ago. Consequently miidevs was synced as far as possible. - Add MIIF_NOMANPAUSE and mii_phy_flowstatus() calls to drivers that weren't explicitly converted to support flow control before. It's unclear whether flow control actually works with these but typically it should and their net behavior should be more correct with these changes in place than without if the MAC driver sets MIIF_DOPAUSE. Obtained from: NetBSD (partially) Reviewed by: yongari (earlier version), silence on arch@ and net@
2011-05-03 19:51:29 +00:00
struct mii_softc *miisc;
int error;
sc = ifp->if_softc;
BFE_LOCK(sc);
mii = device_get_softc(sc->bfe_miibus);
- Remove attempts to implement setting of BMCR_LOOP/MIIF_NOLOOP (reporting IFM_LOOP based on BMCR_LOOP is left in place though as it might provide useful for debugging). For most mii(4) drivers it was unclear whether the PHYs driven by them actually support loopback or not. Moreover, typically loopback mode also needs to be activated on the MAC, which none of the Ethernet drivers using mii(4) implements. Given that loopback media has no real use (and obviously hardly had a chance to actually work) besides for driver development (which just loopback mode should be sufficient for though, i.e one doesn't necessary need support for loopback media) support for it is just dropped as both NetBSD and OpenBSD already did quite some time ago. - Let mii_phy_add_media() also announce the support of IFM_NONE. - Restructure the PHY entry points to use a structure of entry points instead of discrete function pointers, and extend this to include a "reset" entry point. Make sure any PHY-specific reset routine is always used, and provide one for lxtphy(4) which disables MII interrupts (as is done for a few other PHYs we have drivers for). This includes changing NIC drivers which previously just called the generic mii_phy_reset() to now actually call the PHY-specific reset routine, which might be crucial in some cases. While at it, the redundant checks in these NIC drivers for mii->mii_instance not being zero before calling the reset routines were removed because as soon as one PHY driver attaches mii->mii_instance is incremented and we hardly can end up in their media change callbacks etc if no PHY driver has attached as mii_attach() would have failed in that case and not attach a miibus(4) instance. Consequently, NIC drivers now no longer should call mii_phy_reset() directly, so it was removed from EXPORT_SYMS. - Add a mii_phy_dev_attach() as a companion helper to mii_phy_dev_probe(). The purpose of that function is to perform the common steps to attach a PHY driver instance and to hook it up to the miibus(4) instance and to optionally also handle the probing, addition and initialization of the supported media. So all a PHY driver without any special requirements has to do in its bus attach method is to call mii_phy_dev_attach() along with PHY-specific MIIF_* flags, a pointer to its PHY functions and the add_media set to one. All PHY drivers were updated to take advantage of mii_phy_dev_attach() as appropriate. Along with these changes the capability mask was added to the mii_softc structure so PHY drivers taking advantage of mii_phy_dev_attach() but still handling media on their own do not need to fiddle with the MII attach arguments anyway. - Keep track of the PHY offset in the mii_softc structure. This is done for compatibility with NetBSD/OpenBSD. - Keep track of the PHY's OUI, model and revision in the mii_softc structure. Several PHY drivers require this information also after attaching and previously had to wrap their own softc around mii_softc. NetBSD/OpenBSD also keep track of the model and revision on their mii_softc structure. All PHY drivers were updated to take advantage as appropriate. - Convert the mebers of the MII data structure to unsigned where appropriate. This is partly inspired by NetBSD/OpenBSD. - According to IEEE 802.3-2002 the bits actually have to be reversed when mapping an OUI to the MII ID registers. All PHY drivers and miidevs where changed as necessary. Actually this now again allows to largely share miidevs with NetBSD, which fixed this problem already 9 years ago. Consequently miidevs was synced as far as possible. - Add MIIF_NOMANPAUSE and mii_phy_flowstatus() calls to drivers that weren't explicitly converted to support flow control before. It's unclear whether flow control actually works with these but typically it should and their net behavior should be more correct with these changes in place than without if the MAC driver sets MIIF_DOPAUSE. Obtained from: NetBSD (partially) Reviewed by: yongari (earlier version), silence on arch@ and net@
2011-05-03 19:51:29 +00:00
LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
PHY_RESET(miisc);
error = mii_mediachg(mii);
BFE_UNLOCK(sc);
return (error);
}
/*
* Report current media status.
*/
static void
bfe_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
{
struct bfe_softc *sc = ifp->if_softc;
struct mii_data *mii;
BFE_LOCK(sc);
mii = device_get_softc(sc->bfe_miibus);
mii_pollstat(mii);
ifmr->ifm_active = mii->mii_media_active;
ifmr->ifm_status = mii->mii_media_status;
BFE_UNLOCK(sc);
}
static int
bfe_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
{
struct bfe_softc *sc = ifp->if_softc;
struct ifreq *ifr = (struct ifreq *) data;
struct mii_data *mii;
int error = 0;
switch (command) {
case SIOCSIFFLAGS:
BFE_LOCK(sc);
if (ifp->if_flags & IFF_UP) {
if (ifp->if_drv_flags & IFF_DRV_RUNNING)
bfe_set_rx_mode(sc);
else if ((sc->bfe_flags & BFE_FLAG_DETACH) == 0)
bfe_init_locked(sc);
} else if (ifp->if_drv_flags & IFF_DRV_RUNNING)
bfe_stop(sc);
BFE_UNLOCK(sc);
break;
case SIOCADDMULTI:
case SIOCDELMULTI:
BFE_LOCK(sc);
if (ifp->if_drv_flags & IFF_DRV_RUNNING)
bfe_set_rx_mode(sc);
BFE_UNLOCK(sc);
break;
case SIOCGIFMEDIA:
case SIOCSIFMEDIA:
mii = device_get_softc(sc->bfe_miibus);
error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
break;
default:
error = ether_ioctl(ifp, command, data);
break;
}
2004-08-07 20:55:53 +00:00
return (error);
}
static void
bfe_watchdog(struct bfe_softc *sc)
{
struct ifnet *ifp;
BFE_LOCK_ASSERT(sc);
if (sc->bfe_watchdog_timer == 0 || --sc->bfe_watchdog_timer)
return;
ifp = sc->bfe_ifp;
device_printf(sc->bfe_dev, "watchdog timeout -- resetting\n");
ifp->if_oerrors++;
ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
bfe_init_locked(sc);
if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
bfe_start_locked(ifp);
}
static void
bfe_tick(void *xsc)
{
struct bfe_softc *sc = xsc;
struct mii_data *mii;
BFE_LOCK_ASSERT(sc);
mii = device_get_softc(sc->bfe_miibus);
mii_tick(mii);
bfe_stats_update(sc);
bfe_watchdog(sc);
callout_reset(&sc->bfe_stat_co, hz, bfe_tick, sc);
}
/*
* Stop the adapter and free any mbufs allocated to the
* RX and TX lists.
*/
static void
bfe_stop(struct bfe_softc *sc)
{
struct ifnet *ifp;
BFE_LOCK_ASSERT(sc);
ifp = sc->bfe_ifp;
ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
sc->bfe_flags &= ~BFE_FLAG_LINK;
callout_stop(&sc->bfe_stat_co);
sc->bfe_watchdog_timer = 0;
bfe_chip_halt(sc);
bfe_tx_ring_free(sc);
bfe_rx_ring_free(sc);
}
static int
sysctl_bfe_stats(SYSCTL_HANDLER_ARGS)
{
struct bfe_softc *sc;
struct bfe_hw_stats *stats;
int error, result;
result = -1;
error = sysctl_handle_int(oidp, &result, 0, req);
if (error != 0 || req->newptr == NULL)
return (error);
if (result != 1)
return (error);
sc = (struct bfe_softc *)arg1;
stats = &sc->bfe_stats;
printf("%s statistics:\n", device_get_nameunit(sc->bfe_dev));
printf("Transmit good octets : %ju\n",
(uintmax_t)stats->tx_good_octets);
printf("Transmit good frames : %ju\n",
(uintmax_t)stats->tx_good_frames);
printf("Transmit octets : %ju\n",
(uintmax_t)stats->tx_octets);
printf("Transmit frames : %ju\n",
(uintmax_t)stats->tx_frames);
printf("Transmit broadcast frames : %ju\n",
(uintmax_t)stats->tx_bcast_frames);
printf("Transmit multicast frames : %ju\n",
(uintmax_t)stats->tx_mcast_frames);
printf("Transmit frames 64 bytes : %ju\n",
(uint64_t)stats->tx_pkts_64);
printf("Transmit frames 65 to 127 bytes : %ju\n",
(uint64_t)stats->tx_pkts_65_127);
printf("Transmit frames 128 to 255 bytes : %ju\n",
(uint64_t)stats->tx_pkts_128_255);
printf("Transmit frames 256 to 511 bytes : %ju\n",
(uint64_t)stats->tx_pkts_256_511);
printf("Transmit frames 512 to 1023 bytes : %ju\n",
(uint64_t)stats->tx_pkts_512_1023);
printf("Transmit frames 1024 to max bytes : %ju\n",
(uint64_t)stats->tx_pkts_1024_max);
printf("Transmit jabber errors : %u\n", stats->tx_jabbers);
printf("Transmit oversized frames : %ju\n",
(uint64_t)stats->tx_oversize_frames);
printf("Transmit fragmented frames : %ju\n",
(uint64_t)stats->tx_frag_frames);
printf("Transmit underruns : %u\n", stats->tx_colls);
printf("Transmit total collisions : %u\n", stats->tx_single_colls);
printf("Transmit single collisions : %u\n", stats->tx_single_colls);
printf("Transmit multiple collisions : %u\n", stats->tx_multi_colls);
printf("Transmit excess collisions : %u\n", stats->tx_excess_colls);
printf("Transmit late collisions : %u\n", stats->tx_late_colls);
printf("Transmit deferrals : %u\n", stats->tx_deferrals);
printf("Transmit carrier losts : %u\n", stats->tx_carrier_losts);
printf("Transmit pause frames : %u\n", stats->tx_pause_frames);
printf("Receive good octets : %ju\n",
(uintmax_t)stats->rx_good_octets);
printf("Receive good frames : %ju\n",
(uintmax_t)stats->rx_good_frames);
printf("Receive octets : %ju\n",
(uintmax_t)stats->rx_octets);
printf("Receive frames : %ju\n",
(uintmax_t)stats->rx_frames);
printf("Receive broadcast frames : %ju\n",
(uintmax_t)stats->rx_bcast_frames);
printf("Receive multicast frames : %ju\n",
(uintmax_t)stats->rx_mcast_frames);
printf("Receive frames 64 bytes : %ju\n",
(uint64_t)stats->rx_pkts_64);
printf("Receive frames 65 to 127 bytes : %ju\n",
(uint64_t)stats->rx_pkts_65_127);
printf("Receive frames 128 to 255 bytes : %ju\n",
(uint64_t)stats->rx_pkts_128_255);
printf("Receive frames 256 to 511 bytes : %ju\n",
(uint64_t)stats->rx_pkts_256_511);
printf("Receive frames 512 to 1023 bytes : %ju\n",
(uint64_t)stats->rx_pkts_512_1023);
printf("Receive frames 1024 to max bytes : %ju\n",
(uint64_t)stats->rx_pkts_1024_max);
printf("Receive jabber errors : %u\n", stats->rx_jabbers);
printf("Receive oversized frames : %ju\n",
(uint64_t)stats->rx_oversize_frames);
printf("Receive fragmented frames : %ju\n",
(uint64_t)stats->rx_frag_frames);
printf("Receive missed frames : %u\n", stats->rx_missed_frames);
printf("Receive CRC align errors : %u\n", stats->rx_crc_align_errs);
printf("Receive undersized frames : %u\n", stats->rx_runts);
printf("Receive CRC errors : %u\n", stats->rx_crc_errs);
printf("Receive align errors : %u\n", stats->rx_align_errs);
printf("Receive symbol errors : %u\n", stats->rx_symbol_errs);
printf("Receive pause frames : %u\n", stats->rx_pause_frames);
printf("Receive control frames : %u\n", stats->rx_control_frames);
return (error);
}