freebsd-dev/sys/dev/vge/if_vge.c

2930 lines
79 KiB
C
Raw Normal View History

/*-
* Copyright (c) 2004
* Bill Paul <wpaul@windriver.com>. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. All advertising materials mentioning features or use of this software
* must display the following acknowledgement:
* This product includes software developed by Bill Paul.
* 4. Neither the name of the author nor the names of any co-contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
* BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
* THE POSSIBILITY OF SUCH DAMAGE.
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
/*
* VIA Networking Technologies VT612x PCI gigabit ethernet NIC driver.
*
* Written by Bill Paul <wpaul@windriver.com>
* Senior Networking Software Engineer
* Wind River Systems
*/
/*
* The VIA Networking VT6122 is a 32bit, 33/66Mhz PCI device that
* combines a tri-speed ethernet MAC and PHY, with the following
* features:
*
* o Jumbo frame support up to 16K
* o Transmit and receive flow control
* o IPv4 checksum offload
* o VLAN tag insertion and stripping
* o TCP large send
* o 64-bit multicast hash table filter
* o 64 entry CAM filter
* o 16K RX FIFO and 48K TX FIFO memory
* o Interrupt moderation
*
* The VT6122 supports up to four transmit DMA queues. The descriptors
* in the transmit ring can address up to 7 data fragments; frames which
* span more than 7 data buffers must be coalesced, but in general the
* BSD TCP/IP stack rarely generates frames more than 2 or 3 fragments
* long. The receive descriptors address only a single buffer.
*
* There are two peculiar design issues with the VT6122. One is that
* receive data buffers must be aligned on a 32-bit boundary. This is
* not a problem where the VT6122 is used as a LOM device in x86-based
* systems, but on architectures that generate unaligned access traps, we
* have to do some copying.
*
* The other issue has to do with the way 64-bit addresses are handled.
* The DMA descriptors only allow you to specify 48 bits of addressing
* information. The remaining 16 bits are specified using one of the
* I/O registers. If you only have a 32-bit system, then this isn't
* an issue, but if you have a 64-bit system and more than 4GB of
* memory, you must have to make sure your network data buffers reside
* in the same 48-bit 'segment.'
*
* Special thanks to Ryan Fu at VIA Networking for providing documentation
* and sample NICs for testing.
*/
#ifdef HAVE_KERNEL_OPTION_HEADERS
#include "opt_device_polling.h"
#endif
#include <sys/param.h>
#include <sys/endian.h>
#include <sys/systm.h>
#include <sys/sockio.h>
#include <sys/mbuf.h>
#include <sys/malloc.h>
#include <sys/module.h>
#include <sys/kernel.h>
#include <sys/socket.h>
#include <sys/sysctl.h>
#include <net/if.h>
#include <net/if_arp.h>
#include <net/ethernet.h>
#include <net/if_dl.h>
#include <net/if_var.h>
#include <net/if_media.h>
#include <net/if_types.h>
#include <net/if_vlan_var.h>
#include <net/bpf.h>
#include <machine/bus.h>
#include <machine/resource.h>
#include <sys/bus.h>
#include <sys/rman.h>
#include <dev/mii/mii.h>
#include <dev/mii/miivar.h>
#include <dev/pci/pcireg.h>
#include <dev/pci/pcivar.h>
MODULE_DEPEND(vge, pci, 1, 1, 1);
MODULE_DEPEND(vge, ether, 1, 1, 1);
MODULE_DEPEND(vge, miibus, 1, 1, 1);
/* "device miibus" required. See GENERIC if you get errors here. */
#include "miibus_if.h"
#include <dev/vge/if_vgereg.h>
#include <dev/vge/if_vgevar.h>
#define VGE_CSUM_FEATURES (CSUM_IP | CSUM_TCP | CSUM_UDP)
/* Tunables */
static int msi_disable = 0;
TUNABLE_INT("hw.vge.msi_disable", &msi_disable);
/*
* The SQE error counter of MIB seems to report bogus value.
* Vendor's workaround does not seem to work on PCIe based
* controllers. Disable it until we find better workaround.
*/
#undef VGE_ENABLE_SQEERR
/*
* Various supported device vendors/types and their names.
*/
static struct vge_type vge_devs[] = {
{ VIA_VENDORID, VIA_DEVICEID_61XX,
"VIA Networking Velocity Gigabit Ethernet" },
{ 0, 0, NULL }
};
2009-12-14 22:00:11 +00:00
static int vge_attach(device_t);
static int vge_detach(device_t);
static int vge_probe(device_t);
static int vge_resume(device_t);
static int vge_shutdown(device_t);
static int vge_suspend(device_t);
static void vge_cam_clear(struct vge_softc *);
static int vge_cam_set(struct vge_softc *, uint8_t *);
static void vge_clrwol(struct vge_softc *);
2009-12-14 22:00:11 +00:00
static void vge_discard_rxbuf(struct vge_softc *, int);
static int vge_dma_alloc(struct vge_softc *);
static void vge_dma_free(struct vge_softc *);
static void vge_dmamap_cb(void *, bus_dma_segment_t *, int, int);
#ifdef VGE_EEPROM
2009-12-14 22:00:11 +00:00
static void vge_eeprom_getword(struct vge_softc *, int, uint16_t *);
#endif
static int vge_encap(struct vge_softc *, struct mbuf **);
#ifndef __NO_STRICT_ALIGNMENT
static __inline void
vge_fixup_rx(struct mbuf *);
#endif
2009-12-14 22:00:11 +00:00
static void vge_freebufs(struct vge_softc *);
static void vge_ifmedia_sts(struct ifnet *, struct ifmediareq *);
static int vge_ifmedia_upd(struct ifnet *);
Rework link establishment and link state detection logic. - Remove MIIBUS statchg callback and program VGE_DIAGCTL before initiating link establishment. Previously driver used to program VGE_DIAGCTL after getting a link in statchg callback. It seems the VGE_DIAGCTL register works like a kind of MII register such that it requires setting a 'to be' mode in advance rather than relying on resolved speed/duplex of established link. This means the statchg callback is not needed in driver. In addition, if there was no link at the time of media change, this was not called at all. - Introduce vge_ifmedia_upd_locked() to change current media to configured one. Actual media change is performed only after PHY reset and VGE_DIAGCTL setup. - In WOL configuration, make sure to clear forced mode such that controller can rely on auto-negotiation. - Unlike most other drivers that use miibus(4), vge(4) used controller's auto-polling feature for link state tracking via interrupt. This came from controller's inefficient mechanism to access MII registers. On link state change interrupt, vge(4) used to get current link state with series of MII register accesses. Because vge(4) already enabled auto polling, read PHY status register to resolved speed/duplex/flow control parameters. vge(4) still does not drive MII_TICK to reduce number of MII register accesses which in turn means the driver does not know the status of auto-negotiation. This was a one of long standing issue of vge(4). Probably driver may be able to implement a timer that keeps track of auto-negotiation state and restart auto-negotiation when driver couldn't establish a link within a specified period. However the controller does not provide a reliable way to detect auto-negotiation failure so I'm not sure whether it's worth to implement it in driver. Alternatively driver can completely disable MII auto-polling and let miibus(4) poll link state by driving MII_TICK. This may reduce unnecessary overhead of stopping/restarting MII auto-polling of controller. Unfortunately it was known that some variants of controller does not work correctly if MII auto-polling is disabled.
2011-11-22 20:45:09 +00:00
static int vge_ifmedia_upd_locked(struct vge_softc *);
2009-12-14 22:00:11 +00:00
static void vge_init(void *);
static void vge_init_locked(struct vge_softc *);
static void vge_intr(void *);
Implement interrupt moderation scheme supported by VT61xx controllers. TX/RX interrupt mitigation is controlled by VGE_TXSUPPTHR and VGE_RXSUPPTHR register. These registers suppress generation of interrupts until the programmed frames counter equals to the registers. VT61xx also supports interrupt hold off timer register. If this interrupt hold off timer is active all interrupts would be disabled until the timer reaches to 0. The timer value is reloaded whenever VGE_ISR register written. The timer resolution is about 20us. Previously vge(4) used single shot timer to reduce Tx completion interrupts. This required VGE_CRS1 register access in Tx start/completion handler to rearm new timeout value and it did not show satisfactory result(more than 50k interrupts under load). Rx interrupts was not moderated at all such that vge(4) used to generate too many interrupts which in turn made polling(4) better approach under high network load. This change activates all interrupt moderation mechanism and initial values were tuned to generate interrupt less than 8k per second. That number of interrupts wouldn't add additional packet latencies compared to polling(4). These interrupt parameters could be changed with sysctl. dev.vge.%d.int_holdoff dev.vge.%d.rx_coal_pkt dev.vge.%d.tx_coal_pkt Interface has be brought down and up again before change take effect. With interrupt moderation there is no more need to loop in interrupt handler. This loop always added one more register access. While I'm here remove dead code which tried to implement subset of interrupt moderation.
2009-12-17 18:00:25 +00:00
static void vge_intr_holdoff(struct vge_softc *);
2009-12-14 22:00:11 +00:00
static int vge_ioctl(struct ifnet *, u_long, caddr_t);
static void vge_link_statchg(void *);
2009-12-14 22:00:11 +00:00
static int vge_miibus_readreg(device_t, int, int);
static int vge_miibus_writereg(device_t, int, int, int);
static void vge_miipoll_start(struct vge_softc *);
static void vge_miipoll_stop(struct vge_softc *);
static int vge_newbuf(struct vge_softc *, int);
static void vge_read_eeprom(struct vge_softc *, caddr_t, int, int, int);
static void vge_reset(struct vge_softc *);
static int vge_rx_list_init(struct vge_softc *);
static int vge_rxeof(struct vge_softc *, int);
static void vge_rxfilter(struct vge_softc *);
Rework link establishment and link state detection logic. - Remove MIIBUS statchg callback and program VGE_DIAGCTL before initiating link establishment. Previously driver used to program VGE_DIAGCTL after getting a link in statchg callback. It seems the VGE_DIAGCTL register works like a kind of MII register such that it requires setting a 'to be' mode in advance rather than relying on resolved speed/duplex of established link. This means the statchg callback is not needed in driver. In addition, if there was no link at the time of media change, this was not called at all. - Introduce vge_ifmedia_upd_locked() to change current media to configured one. Actual media change is performed only after PHY reset and VGE_DIAGCTL setup. - In WOL configuration, make sure to clear forced mode such that controller can rely on auto-negotiation. - Unlike most other drivers that use miibus(4), vge(4) used controller's auto-polling feature for link state tracking via interrupt. This came from controller's inefficient mechanism to access MII registers. On link state change interrupt, vge(4) used to get current link state with series of MII register accesses. Because vge(4) already enabled auto polling, read PHY status register to resolved speed/duplex/flow control parameters. vge(4) still does not drive MII_TICK to reduce number of MII register accesses which in turn means the driver does not know the status of auto-negotiation. This was a one of long standing issue of vge(4). Probably driver may be able to implement a timer that keeps track of auto-negotiation state and restart auto-negotiation when driver couldn't establish a link within a specified period. However the controller does not provide a reliable way to detect auto-negotiation failure so I'm not sure whether it's worth to implement it in driver. Alternatively driver can completely disable MII auto-polling and let miibus(4) poll link state by driving MII_TICK. This may reduce unnecessary overhead of stopping/restarting MII auto-polling of controller. Unfortunately it was known that some variants of controller does not work correctly if MII auto-polling is disabled.
2011-11-22 20:45:09 +00:00
static void vge_setmedia(struct vge_softc *);
static void vge_setvlan(struct vge_softc *);
static void vge_setwol(struct vge_softc *);
2009-12-14 22:00:11 +00:00
static void vge_start(struct ifnet *);
static void vge_start_locked(struct ifnet *);
static void vge_stats_clear(struct vge_softc *);
static void vge_stats_update(struct vge_softc *);
2009-12-14 22:00:11 +00:00
static void vge_stop(struct vge_softc *);
static void vge_sysctl_node(struct vge_softc *);
2009-12-14 22:00:11 +00:00
static int vge_tx_list_init(struct vge_softc *);
static void vge_txeof(struct vge_softc *);
static void vge_watchdog(void *);
static device_method_t vge_methods[] = {
/* Device interface */
DEVMETHOD(device_probe, vge_probe),
DEVMETHOD(device_attach, vge_attach),
DEVMETHOD(device_detach, vge_detach),
DEVMETHOD(device_suspend, vge_suspend),
DEVMETHOD(device_resume, vge_resume),
DEVMETHOD(device_shutdown, vge_shutdown),
/* MII interface */
DEVMETHOD(miibus_readreg, vge_miibus_readreg),
DEVMETHOD(miibus_writereg, vge_miibus_writereg),
DEVMETHOD_END
};
static driver_t vge_driver = {
"vge",
vge_methods,
sizeof(struct vge_softc)
};
static devclass_t vge_devclass;
DRIVER_MODULE(vge, pci, vge_driver, vge_devclass, 0, 0);
DRIVER_MODULE(miibus, vge, miibus_driver, miibus_devclass, 0, 0);
#ifdef VGE_EEPROM
/*
* Read a word of data stored in the EEPROM at address 'addr.'
*/
static void
2009-12-14 19:53:57 +00:00
vge_eeprom_getword(struct vge_softc *sc, int addr, uint16_t *dest)
{
2009-12-14 20:07:25 +00:00
int i;
uint16_t word = 0;
/*
* Enter EEPROM embedded programming mode. In order to
* access the EEPROM at all, we first have to set the
* EELOAD bit in the CHIPCFG2 register.
*/
CSR_SETBIT_1(sc, VGE_CHIPCFG2, VGE_CHIPCFG2_EELOAD);
CSR_SETBIT_1(sc, VGE_EECSR, VGE_EECSR_EMBP/*|VGE_EECSR_ECS*/);
/* Select the address of the word we want to read */
CSR_WRITE_1(sc, VGE_EEADDR, addr);
/* Issue read command */
CSR_SETBIT_1(sc, VGE_EECMD, VGE_EECMD_ERD);
/* Wait for the done bit to be set. */
for (i = 0; i < VGE_TIMEOUT; i++) {
if (CSR_READ_1(sc, VGE_EECMD) & VGE_EECMD_EDONE)
break;
}
if (i == VGE_TIMEOUT) {
device_printf(sc->vge_dev, "EEPROM read timed out\n");
*dest = 0;
return;
}
/* Read the result */
word = CSR_READ_2(sc, VGE_EERDDAT);
/* Turn off EEPROM access mode. */
CSR_CLRBIT_1(sc, VGE_EECSR, VGE_EECSR_EMBP/*|VGE_EECSR_ECS*/);
CSR_CLRBIT_1(sc, VGE_CHIPCFG2, VGE_CHIPCFG2_EELOAD);
*dest = word;
}
#endif
/*
* Read a sequence of words from the EEPROM.
*/
static void
2009-12-14 19:44:54 +00:00
vge_read_eeprom(struct vge_softc *sc, caddr_t dest, int off, int cnt, int swap)
{
2009-12-14 20:07:25 +00:00
int i;
#ifdef VGE_EEPROM
2009-12-14 20:07:25 +00:00
uint16_t word = 0, *ptr;
for (i = 0; i < cnt; i++) {
vge_eeprom_getword(sc, off + i, &word);
2009-12-14 19:53:57 +00:00
ptr = (uint16_t *)(dest + (i * 2));
if (swap)
*ptr = ntohs(word);
else
*ptr = word;
}
#else
for (i = 0; i < ETHER_ADDR_LEN; i++)
dest[i] = CSR_READ_1(sc, VGE_PAR0 + i);
#endif
}
static void
2009-12-14 19:44:54 +00:00
vge_miipoll_stop(struct vge_softc *sc)
{
2009-12-14 20:07:25 +00:00
int i;
CSR_WRITE_1(sc, VGE_MIICMD, 0);
for (i = 0; i < VGE_TIMEOUT; i++) {
DELAY(1);
if (CSR_READ_1(sc, VGE_MIISTS) & VGE_MIISTS_IIDL)
break;
}
if (i == VGE_TIMEOUT)
device_printf(sc->vge_dev, "failed to idle MII autopoll\n");
}
static void
2009-12-14 19:44:54 +00:00
vge_miipoll_start(struct vge_softc *sc)
{
2009-12-14 20:07:25 +00:00
int i;
/* First, make sure we're idle. */
CSR_WRITE_1(sc, VGE_MIICMD, 0);
CSR_WRITE_1(sc, VGE_MIIADDR, VGE_MIIADDR_SWMPL);
for (i = 0; i < VGE_TIMEOUT; i++) {
DELAY(1);
if (CSR_READ_1(sc, VGE_MIISTS) & VGE_MIISTS_IIDL)
break;
}
if (i == VGE_TIMEOUT) {
device_printf(sc->vge_dev, "failed to idle MII autopoll\n");
return;
}
/* Now enable auto poll mode. */
CSR_WRITE_1(sc, VGE_MIICMD, VGE_MIICMD_MAUTO);
/* And make sure it started. */
for (i = 0; i < VGE_TIMEOUT; i++) {
DELAY(1);
if ((CSR_READ_1(sc, VGE_MIISTS) & VGE_MIISTS_IIDL) == 0)
break;
}
if (i == VGE_TIMEOUT)
device_printf(sc->vge_dev, "failed to start MII autopoll\n");
}
static int
2009-12-14 19:44:54 +00:00
vge_miibus_readreg(device_t dev, int phy, int reg)
{
2009-12-14 20:07:25 +00:00
struct vge_softc *sc;
int i;
uint16_t rval = 0;
sc = device_get_softc(dev);
vge_miipoll_stop(sc);
/* Specify the register we want to read. */
CSR_WRITE_1(sc, VGE_MIIADDR, reg);
/* Issue read command. */
CSR_SETBIT_1(sc, VGE_MIICMD, VGE_MIICMD_RCMD);
/* Wait for the read command bit to self-clear. */
for (i = 0; i < VGE_TIMEOUT; i++) {
DELAY(1);
if ((CSR_READ_1(sc, VGE_MIICMD) & VGE_MIICMD_RCMD) == 0)
break;
}
if (i == VGE_TIMEOUT)
device_printf(sc->vge_dev, "MII read timed out\n");
else
rval = CSR_READ_2(sc, VGE_MIIDATA);
vge_miipoll_start(sc);
return (rval);
}
static int
2009-12-14 19:44:54 +00:00
vge_miibus_writereg(device_t dev, int phy, int reg, int data)
{
2009-12-14 20:07:25 +00:00
struct vge_softc *sc;
int i, rval = 0;
sc = device_get_softc(dev);
vge_miipoll_stop(sc);
/* Specify the register we want to write. */
CSR_WRITE_1(sc, VGE_MIIADDR, reg);
/* Specify the data we want to write. */
CSR_WRITE_2(sc, VGE_MIIDATA, data);
/* Issue write command. */
CSR_SETBIT_1(sc, VGE_MIICMD, VGE_MIICMD_WCMD);
/* Wait for the write command bit to self-clear. */
for (i = 0; i < VGE_TIMEOUT; i++) {
DELAY(1);
if ((CSR_READ_1(sc, VGE_MIICMD) & VGE_MIICMD_WCMD) == 0)
break;
}
if (i == VGE_TIMEOUT) {
device_printf(sc->vge_dev, "MII write timed out\n");
rval = EIO;
}
vge_miipoll_start(sc);
return (rval);
}
static void
2009-12-14 19:44:54 +00:00
vge_cam_clear(struct vge_softc *sc)
{
2009-12-14 20:07:25 +00:00
int i;
/*
* Turn off all the mask bits. This tells the chip
* that none of the entries in the CAM filter are valid.
* desired entries will be enabled as we fill the filter in.
*/
CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_CAMMASK);
CSR_WRITE_1(sc, VGE_CAMADDR, VGE_CAMADDR_ENABLE);
for (i = 0; i < 8; i++)
CSR_WRITE_1(sc, VGE_CAM0 + i, 0);
/* Clear the VLAN filter too. */
CSR_WRITE_1(sc, VGE_CAMADDR, VGE_CAMADDR_ENABLE|VGE_CAMADDR_AVSEL|0);
for (i = 0; i < 8; i++)
CSR_WRITE_1(sc, VGE_CAM0 + i, 0);
CSR_WRITE_1(sc, VGE_CAMADDR, 0);
CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_MAR);
sc->vge_camidx = 0;
}
static int
2009-12-14 19:44:54 +00:00
vge_cam_set(struct vge_softc *sc, uint8_t *addr)
{
2009-12-14 20:07:25 +00:00
int i, error = 0;
if (sc->vge_camidx == VGE_CAM_MAXADDRS)
2009-12-14 20:07:25 +00:00
return (ENOSPC);
/* Select the CAM data page. */
CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_CAMDATA);
/* Set the filter entry we want to update and enable writing. */
CSR_WRITE_1(sc, VGE_CAMADDR, VGE_CAMADDR_ENABLE|sc->vge_camidx);
/* Write the address to the CAM registers */
for (i = 0; i < ETHER_ADDR_LEN; i++)
CSR_WRITE_1(sc, VGE_CAM0 + i, addr[i]);
/* Issue a write command. */
CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_WRITE);
/* Wake for it to clear. */
for (i = 0; i < VGE_TIMEOUT; i++) {
DELAY(1);
if ((CSR_READ_1(sc, VGE_CAMCTL) & VGE_CAMCTL_WRITE) == 0)
break;
}
if (i == VGE_TIMEOUT) {
device_printf(sc->vge_dev, "setting CAM filter failed\n");
error = EIO;
goto fail;
}
/* Select the CAM mask page. */
CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_CAMMASK);
/* Set the mask bit that enables this filter. */
CSR_SETBIT_1(sc, VGE_CAM0 + (sc->vge_camidx/8),
1<<(sc->vge_camidx & 7));
sc->vge_camidx++;
fail:
/* Turn off access to CAM. */
CSR_WRITE_1(sc, VGE_CAMADDR, 0);
CSR_CLRBIT_1(sc, VGE_CAMCTL, VGE_CAMCTL_PAGESEL);
CSR_SETBIT_1(sc, VGE_CAMCTL, VGE_PAGESEL_MAR);
return (error);
}
static void
vge_setvlan(struct vge_softc *sc)
{
struct ifnet *ifp;
uint8_t cfg;
VGE_LOCK_ASSERT(sc);
ifp = sc->vge_ifp;
cfg = CSR_READ_1(sc, VGE_RXCFG);
if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
cfg |= VGE_VTAG_OPT2;
else
cfg &= ~VGE_VTAG_OPT2;
CSR_WRITE_1(sc, VGE_RXCFG, cfg);
}
/*
* Program the multicast filter. We use the 64-entry CAM filter
* for perfect filtering. If there's more than 64 multicast addresses,
2009-12-14 18:19:56 +00:00
* we use the hash filter instead.
*/
static void
vge_rxfilter(struct vge_softc *sc)
{
2009-12-14 20:07:25 +00:00
struct ifnet *ifp;
struct ifmultiaddr *ifma;
uint32_t h, hashes[2];
uint8_t rxcfg;
int error = 0;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
VGE_LOCK_ASSERT(sc);
/* First, zot all the multicast entries. */
hashes[0] = 0;
hashes[1] = 0;
rxcfg = CSR_READ_1(sc, VGE_RXCTL);
rxcfg &= ~(VGE_RXCTL_RX_MCAST | VGE_RXCTL_RX_BCAST |
VGE_RXCTL_RX_PROMISC);
/*
* Always allow VLAN oversized frames and frames for
* this host.
*/
rxcfg |= VGE_RXCTL_RX_GIANT | VGE_RXCTL_RX_UCAST;
ifp = sc->vge_ifp;
if ((ifp->if_flags & IFF_BROADCAST) != 0)
rxcfg |= VGE_RXCTL_RX_BCAST;
if ((ifp->if_flags & (IFF_PROMISC | IFF_ALLMULTI)) != 0) {
if ((ifp->if_flags & IFF_PROMISC) != 0)
rxcfg |= VGE_RXCTL_RX_PROMISC;
if ((ifp->if_flags & IFF_ALLMULTI) != 0) {
hashes[0] = 0xFFFFFFFF;
hashes[1] = 0xFFFFFFFF;
}
goto done;
}
vge_cam_clear(sc);
/* Now program new ones */
if_maddr_rlock(ifp);
TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
if (ifma->ifma_addr->sa_family != AF_LINK)
continue;
error = vge_cam_set(sc,
LLADDR((struct sockaddr_dl *)ifma->ifma_addr));
if (error)
break;
}
/* If there were too many addresses, use the hash filter. */
if (error) {
vge_cam_clear(sc);
TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
if (ifma->ifma_addr->sa_family != AF_LINK)
continue;
h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
ifma->ifma_addr), ETHER_ADDR_LEN) >> 26;
if (h < 32)
hashes[0] |= (1 << h);
else
hashes[1] |= (1 << (h - 32));
}
}
if_maddr_runlock(ifp);
done:
if (hashes[0] != 0 || hashes[1] != 0)
rxcfg |= VGE_RXCTL_RX_MCAST;
CSR_WRITE_4(sc, VGE_MAR0, hashes[0]);
CSR_WRITE_4(sc, VGE_MAR1, hashes[1]);
CSR_WRITE_1(sc, VGE_RXCTL, rxcfg);
}
static void
2009-12-14 19:44:54 +00:00
vge_reset(struct vge_softc *sc)
{
2009-12-14 20:07:25 +00:00
int i;
CSR_WRITE_1(sc, VGE_CRS1, VGE_CR1_SOFTRESET);
for (i = 0; i < VGE_TIMEOUT; i++) {
DELAY(5);
if ((CSR_READ_1(sc, VGE_CRS1) & VGE_CR1_SOFTRESET) == 0)
break;
}
if (i == VGE_TIMEOUT) {
device_printf(sc->vge_dev, "soft reset timed out\n");
CSR_WRITE_1(sc, VGE_CRS3, VGE_CR3_STOP_FORCE);
DELAY(2000);
}
DELAY(5000);
}
/*
* Probe for a VIA gigabit chip. Check the PCI vendor and device
* IDs against our list and return a device name if we find a match.
*/
static int
2009-12-14 19:44:54 +00:00
vge_probe(device_t dev)
{
2009-12-14 20:07:25 +00:00
struct vge_type *t;
t = vge_devs;
while (t->vge_name != NULL) {
if ((pci_get_vendor(dev) == t->vge_vid) &&
(pci_get_device(dev) == t->vge_did)) {
device_set_desc(dev, t->vge_name);
return (BUS_PROBE_DEFAULT);
}
t++;
}
return (ENXIO);
}
/*
* Map a single buffer address.
*/
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
struct vge_dmamap_arg {
bus_addr_t vge_busaddr;
};
static void
2009-12-14 19:44:54 +00:00
vge_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error)
{
2009-12-14 20:07:25 +00:00
struct vge_dmamap_arg *ctx;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
if (error != 0)
return;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
KASSERT(nsegs == 1, ("%s: %d segments returned!", __func__, nsegs));
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
ctx = (struct vge_dmamap_arg *)arg;
ctx->vge_busaddr = segs[0].ds_addr;
}
static int
2009-12-14 19:44:54 +00:00
vge_dma_alloc(struct vge_softc *sc)
{
2009-12-14 20:07:25 +00:00
struct vge_dmamap_arg ctx;
struct vge_txdesc *txd;
struct vge_rxdesc *rxd;
bus_addr_t lowaddr, tx_ring_end, rx_ring_end;
int error, i;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
/*
* It seems old PCI controllers do not support DAC. DAC
* configuration can be enabled by accessing VGE_CHIPCFG3
* register but honor EEPROM configuration instead of
* blindly overriding DAC configuration. PCIe based
* controllers are supposed to support 64bit DMA so enable
* 64bit DMA on these controllers.
*/
if ((sc->vge_flags & VGE_FLAG_PCIE) != 0)
lowaddr = BUS_SPACE_MAXADDR;
else
lowaddr = BUS_SPACE_MAXADDR_32BIT;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
again:
/* Create parent ring tag. */
error = bus_dma_tag_create(bus_get_dma_tag(sc->vge_dev),/* parent */
1, 0, /* algnmnt, boundary */
lowaddr, /* 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->vge_cdata.vge_ring_tag);
if (error != 0) {
device_printf(sc->vge_dev,
"could not create parent DMA tag.\n");
goto fail;
}
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
/* Create tag for Tx ring. */
error = bus_dma_tag_create(sc->vge_cdata.vge_ring_tag,/* parent */
VGE_TX_RING_ALIGN, 0, /* algnmnt, boundary */
BUS_SPACE_MAXADDR, /* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
VGE_TX_LIST_SZ, /* maxsize */
1, /* nsegments */
VGE_TX_LIST_SZ, /* maxsegsize */
0, /* flags */
NULL, NULL, /* lockfunc, lockarg */
&sc->vge_cdata.vge_tx_ring_tag);
if (error != 0) {
device_printf(sc->vge_dev,
"could not allocate Tx ring DMA tag.\n");
goto fail;
}
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
/* Create tag for Rx ring. */
error = bus_dma_tag_create(sc->vge_cdata.vge_ring_tag,/* parent */
VGE_RX_RING_ALIGN, 0, /* algnmnt, boundary */
BUS_SPACE_MAXADDR, /* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
VGE_RX_LIST_SZ, /* maxsize */
1, /* nsegments */
VGE_RX_LIST_SZ, /* maxsegsize */
0, /* flags */
NULL, NULL, /* lockfunc, lockarg */
&sc->vge_cdata.vge_rx_ring_tag);
if (error != 0) {
device_printf(sc->vge_dev,
"could not allocate Rx ring DMA tag.\n");
goto fail;
}
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
/* Allocate DMA'able memory and load the DMA map for Tx ring. */
error = bus_dmamem_alloc(sc->vge_cdata.vge_tx_ring_tag,
(void **)&sc->vge_rdata.vge_tx_ring,
BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT,
&sc->vge_cdata.vge_tx_ring_map);
if (error != 0) {
device_printf(sc->vge_dev,
"could not allocate DMA'able memory for Tx ring.\n");
goto fail;
}
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
ctx.vge_busaddr = 0;
error = bus_dmamap_load(sc->vge_cdata.vge_tx_ring_tag,
sc->vge_cdata.vge_tx_ring_map, sc->vge_rdata.vge_tx_ring,
VGE_TX_LIST_SZ, vge_dmamap_cb, &ctx, BUS_DMA_NOWAIT);
if (error != 0 || ctx.vge_busaddr == 0) {
device_printf(sc->vge_dev,
"could not load DMA'able memory for Tx ring.\n");
goto fail;
}
sc->vge_rdata.vge_tx_ring_paddr = ctx.vge_busaddr;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
/* Allocate DMA'able memory and load the DMA map for Rx ring. */
error = bus_dmamem_alloc(sc->vge_cdata.vge_rx_ring_tag,
(void **)&sc->vge_rdata.vge_rx_ring,
BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT,
&sc->vge_cdata.vge_rx_ring_map);
if (error != 0) {
device_printf(sc->vge_dev,
"could not allocate DMA'able memory for Rx ring.\n");
goto fail;
}
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
ctx.vge_busaddr = 0;
error = bus_dmamap_load(sc->vge_cdata.vge_rx_ring_tag,
sc->vge_cdata.vge_rx_ring_map, sc->vge_rdata.vge_rx_ring,
VGE_RX_LIST_SZ, vge_dmamap_cb, &ctx, BUS_DMA_NOWAIT);
if (error != 0 || ctx.vge_busaddr == 0) {
device_printf(sc->vge_dev,
"could not load DMA'able memory for Rx ring.\n");
goto fail;
}
sc->vge_rdata.vge_rx_ring_paddr = ctx.vge_busaddr;
/* Tx/Rx descriptor queue should reside within 4GB boundary. */
tx_ring_end = sc->vge_rdata.vge_tx_ring_paddr + VGE_TX_LIST_SZ;
rx_ring_end = sc->vge_rdata.vge_rx_ring_paddr + VGE_RX_LIST_SZ;
if ((VGE_ADDR_HI(tx_ring_end) !=
VGE_ADDR_HI(sc->vge_rdata.vge_tx_ring_paddr)) ||
(VGE_ADDR_HI(rx_ring_end) !=
VGE_ADDR_HI(sc->vge_rdata.vge_rx_ring_paddr)) ||
VGE_ADDR_HI(tx_ring_end) != VGE_ADDR_HI(rx_ring_end)) {
device_printf(sc->vge_dev, "4GB boundary crossed, "
"switching to 32bit DMA address mode.\n");
vge_dma_free(sc);
/* Limit DMA address space to 32bit and try again. */
lowaddr = BUS_SPACE_MAXADDR_32BIT;
goto again;
}
if ((sc->vge_flags & VGE_FLAG_PCIE) != 0)
lowaddr = VGE_BUF_DMA_MAXADDR;
else
lowaddr = BUS_SPACE_MAXADDR_32BIT;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
/* Create parent buffer tag. */
error = bus_dma_tag_create(bus_get_dma_tag(sc->vge_dev),/* parent */
1, 0, /* algnmnt, boundary */
lowaddr, /* lowaddr */
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
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->vge_cdata.vge_buffer_tag);
if (error != 0) {
device_printf(sc->vge_dev,
"could not create parent buffer DMA tag.\n");
goto fail;
}
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
/* Create tag for Tx buffers. */
error = bus_dma_tag_create(sc->vge_cdata.vge_buffer_tag,/* parent */
1, 0, /* algnmnt, boundary */
BUS_SPACE_MAXADDR, /* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
MCLBYTES * VGE_MAXTXSEGS, /* maxsize */
VGE_MAXTXSEGS, /* nsegments */
MCLBYTES, /* maxsegsize */
0, /* flags */
NULL, NULL, /* lockfunc, lockarg */
&sc->vge_cdata.vge_tx_tag);
if (error != 0) {
device_printf(sc->vge_dev, "could not create Tx DMA tag.\n");
goto fail;
}
/* Create tag for Rx buffers. */
error = bus_dma_tag_create(sc->vge_cdata.vge_buffer_tag,/* parent */
VGE_RX_BUF_ALIGN, 0, /* algnmnt, 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->vge_cdata.vge_rx_tag);
if (error != 0) {
device_printf(sc->vge_dev, "could not create Rx DMA tag.\n");
goto fail;
}
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
/* Create DMA maps for Tx buffers. */
for (i = 0; i < VGE_TX_DESC_CNT; i++) {
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
txd = &sc->vge_cdata.vge_txdesc[i];
txd->tx_m = NULL;
txd->tx_dmamap = NULL;
error = bus_dmamap_create(sc->vge_cdata.vge_tx_tag, 0,
&txd->tx_dmamap);
if (error != 0) {
device_printf(sc->vge_dev,
"could not create Tx dmamap.\n");
goto fail;
}
}
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
/* Create DMA maps for Rx buffers. */
if ((error = bus_dmamap_create(sc->vge_cdata.vge_rx_tag, 0,
&sc->vge_cdata.vge_rx_sparemap)) != 0) {
device_printf(sc->vge_dev,
"could not create spare Rx dmamap.\n");
goto fail;
}
for (i = 0; i < VGE_RX_DESC_CNT; i++) {
rxd = &sc->vge_cdata.vge_rxdesc[i];
rxd->rx_m = NULL;
rxd->rx_dmamap = NULL;
error = bus_dmamap_create(sc->vge_cdata.vge_rx_tag, 0,
&rxd->rx_dmamap);
if (error != 0) {
device_printf(sc->vge_dev,
"could not create Rx dmamap.\n");
goto fail;
}
}
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
fail:
return (error);
}
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
static void
2009-12-14 19:44:54 +00:00
vge_dma_free(struct vge_softc *sc)
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
{
2009-12-14 20:07:25 +00:00
struct vge_txdesc *txd;
struct vge_rxdesc *rxd;
int i;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
/* Tx ring. */
if (sc->vge_cdata.vge_tx_ring_tag != NULL) {
if (sc->vge_rdata.vge_tx_ring_paddr)
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
bus_dmamap_unload(sc->vge_cdata.vge_tx_ring_tag,
sc->vge_cdata.vge_tx_ring_map);
if (sc->vge_rdata.vge_tx_ring)
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
bus_dmamem_free(sc->vge_cdata.vge_tx_ring_tag,
sc->vge_rdata.vge_tx_ring,
sc->vge_cdata.vge_tx_ring_map);
sc->vge_rdata.vge_tx_ring = NULL;
sc->vge_rdata.vge_tx_ring_paddr = 0;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
bus_dma_tag_destroy(sc->vge_cdata.vge_tx_ring_tag);
sc->vge_cdata.vge_tx_ring_tag = NULL;
}
/* Rx ring. */
if (sc->vge_cdata.vge_rx_ring_tag != NULL) {
if (sc->vge_rdata.vge_rx_ring_paddr)
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
bus_dmamap_unload(sc->vge_cdata.vge_rx_ring_tag,
sc->vge_cdata.vge_rx_ring_map);
if (sc->vge_rdata.vge_rx_ring)
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
bus_dmamem_free(sc->vge_cdata.vge_rx_ring_tag,
sc->vge_rdata.vge_rx_ring,
sc->vge_cdata.vge_rx_ring_map);
sc->vge_rdata.vge_rx_ring = NULL;
sc->vge_rdata.vge_rx_ring_paddr = 0;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
bus_dma_tag_destroy(sc->vge_cdata.vge_rx_ring_tag);
sc->vge_cdata.vge_rx_ring_tag = NULL;
}
/* Tx buffers. */
if (sc->vge_cdata.vge_tx_tag != NULL) {
for (i = 0; i < VGE_TX_DESC_CNT; i++) {
txd = &sc->vge_cdata.vge_txdesc[i];
if (txd->tx_dmamap != NULL) {
bus_dmamap_destroy(sc->vge_cdata.vge_tx_tag,
txd->tx_dmamap);
txd->tx_dmamap = NULL;
}
}
bus_dma_tag_destroy(sc->vge_cdata.vge_tx_tag);
sc->vge_cdata.vge_tx_tag = NULL;
}
/* Rx buffers. */
if (sc->vge_cdata.vge_rx_tag != NULL) {
for (i = 0; i < VGE_RX_DESC_CNT; i++) {
rxd = &sc->vge_cdata.vge_rxdesc[i];
if (rxd->rx_dmamap != NULL) {
bus_dmamap_destroy(sc->vge_cdata.vge_rx_tag,
rxd->rx_dmamap);
rxd->rx_dmamap = NULL;
}
}
if (sc->vge_cdata.vge_rx_sparemap != NULL) {
bus_dmamap_destroy(sc->vge_cdata.vge_rx_tag,
sc->vge_cdata.vge_rx_sparemap);
sc->vge_cdata.vge_rx_sparemap = NULL;
}
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
bus_dma_tag_destroy(sc->vge_cdata.vge_rx_tag);
sc->vge_cdata.vge_rx_tag = NULL;
}
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
if (sc->vge_cdata.vge_buffer_tag != NULL) {
bus_dma_tag_destroy(sc->vge_cdata.vge_buffer_tag);
sc->vge_cdata.vge_buffer_tag = NULL;
}
if (sc->vge_cdata.vge_ring_tag != NULL) {
bus_dma_tag_destroy(sc->vge_cdata.vge_ring_tag);
sc->vge_cdata.vge_ring_tag = NULL;
}
}
/*
* Attach the interface. Allocate softc structures, do ifmedia
* setup and ethernet/BPF attach.
*/
static int
2009-12-14 19:44:54 +00:00
vge_attach(device_t dev)
{
2009-12-14 20:07:25 +00:00
u_char eaddr[ETHER_ADDR_LEN];
struct vge_softc *sc;
struct ifnet *ifp;
int error = 0, cap, i, msic, rid;
sc = device_get_softc(dev);
sc->vge_dev = dev;
mtx_init(&sc->vge_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
MTX_DEF);
callout_init_mtx(&sc->vge_watchdog, &sc->vge_mtx, 0);
/*
* Map control/status registers.
*/
pci_enable_busmaster(dev);
rid = PCIR_BAR(1);
sc->vge_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid,
RF_ACTIVE);
if (sc->vge_res == NULL) {
2009-12-14 18:00:08 +00:00
device_printf(dev, "couldn't map ports/memory\n");
error = ENXIO;
goto fail;
}
if (pci_find_cap(dev, PCIY_EXPRESS, &cap) == 0) {
sc->vge_flags |= VGE_FLAG_PCIE;
sc->vge_expcap = cap;
} else
sc->vge_flags |= VGE_FLAG_JUMBO;
if (pci_find_cap(dev, PCIY_PMG, &cap) == 0) {
sc->vge_flags |= VGE_FLAG_PMCAP;
sc->vge_pmcap = cap;
}
rid = 0;
msic = pci_msi_count(dev);
if (msi_disable == 0 && msic > 0) {
msic = 1;
if (pci_alloc_msi(dev, &msic) == 0) {
if (msic == 1) {
sc->vge_flags |= VGE_FLAG_MSI;
device_printf(dev, "Using %d MSI message\n",
msic);
rid = 1;
} else
pci_release_msi(dev);
}
}
/* Allocate interrupt */
sc->vge_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
((sc->vge_flags & VGE_FLAG_MSI) ? 0 : RF_SHAREABLE) | RF_ACTIVE);
if (sc->vge_irq == NULL) {
2009-12-14 18:00:08 +00:00
device_printf(dev, "couldn't map interrupt\n");
error = ENXIO;
goto fail;
}
/* Reset the adapter. */
vge_reset(sc);
/* Reload EEPROM. */
CSR_WRITE_1(sc, VGE_EECSR, VGE_EECSR_RELOAD);
for (i = 0; i < VGE_TIMEOUT; i++) {
DELAY(5);
if ((CSR_READ_1(sc, VGE_EECSR) & VGE_EECSR_RELOAD) == 0)
break;
}
if (i == VGE_TIMEOUT)
device_printf(dev, "EEPROM reload timed out\n");
/*
* Clear PACPI as EEPROM reload will set the bit. Otherwise
* MAC will receive magic packet which in turn confuses
* controller.
*/
CSR_CLRBIT_1(sc, VGE_CHIPCFG0, VGE_CHIPCFG0_PACPI);
/*
* Get station address from the EEPROM.
*/
vge_read_eeprom(sc, (caddr_t)eaddr, VGE_EE_EADDR, 3, 0);
/*
* Save configured PHY address.
* It seems the PHY address of PCIe controllers just
* reflects media jump strapping status so we assume the
* internal PHY address of PCIe controller is at 1.
*/
if ((sc->vge_flags & VGE_FLAG_PCIE) != 0)
sc->vge_phyaddr = 1;
else
sc->vge_phyaddr = CSR_READ_1(sc, VGE_MIICFG) &
VGE_MIICFG_PHYADDR;
/* Clear WOL and take hardware from powerdown. */
vge_clrwol(sc);
vge_sysctl_node(sc);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
error = vge_dma_alloc(sc);
if (error)
goto fail;
ifp = sc->vge_ifp = if_alloc(IFT_ETHER);
if (ifp == NULL) {
device_printf(dev, "can not if_alloc()\n");
error = ENOSPC;
goto fail;
}
vge_miipoll_start(sc);
/* Do MII setup */
error = mii_attach(dev, &sc->vge_miibus, ifp, vge_ifmedia_upd,
vge_ifmedia_sts, BMSR_DEFCAPMASK, sc->vge_phyaddr, MII_OFFSET_ANY,
MIIF_DOPAUSE);
if (error != 0) {
device_printf(dev, "attaching PHYs failed\n");
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 = vge_ioctl;
ifp->if_capabilities = IFCAP_VLAN_MTU;
ifp->if_start = vge_start;
ifp->if_hwassist = VGE_CSUM_FEATURES;
ifp->if_capabilities |= IFCAP_HWCSUM | IFCAP_VLAN_HWCSUM |
IFCAP_VLAN_HWTAGGING;
if ((sc->vge_flags & VGE_FLAG_PMCAP) != 0)
ifp->if_capabilities |= IFCAP_WOL;
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
ifp->if_capenable = ifp->if_capabilities;
#ifdef DEVICE_POLLING
ifp->if_capabilities |= IFCAP_POLLING;
#endif
ifp->if_init = vge_init;
IFQ_SET_MAXLEN(&ifp->if_snd, VGE_TX_DESC_CNT - 1);
ifp->if_snd.ifq_drv_maxlen = VGE_TX_DESC_CNT - 1;
IFQ_SET_READY(&ifp->if_snd);
/*
* Call MI attach routine.
*/
ether_ifattach(ifp, eaddr);
/* Tell the upper layer(s) we support long frames. */
ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
/* Hook interrupt last to avoid having to lock softc */
error = bus_setup_intr(dev, sc->vge_irq, INTR_TYPE_NET|INTR_MPSAFE,
NULL, vge_intr, sc, &sc->vge_intrhand);
if (error) {
2009-12-14 18:00:08 +00:00
device_printf(dev, "couldn't set up irq\n");
ether_ifdetach(ifp);
goto fail;
}
fail:
if (error)
vge_detach(dev);
return (error);
}
/*
* Shutdown hardware and free up resources. This can be called any
* time after the mutex has been initialized. It is called in both
* the error case in attach and the normal detach case so it needs
* to be careful about only freeing resources that have actually been
* allocated.
*/
static int
2009-12-14 19:44:54 +00:00
vge_detach(device_t dev)
{
2009-12-14 20:07:25 +00:00
struct vge_softc *sc;
struct ifnet *ifp;
sc = device_get_softc(dev);
KASSERT(mtx_initialized(&sc->vge_mtx), ("vge mutex not initialized"));
ifp = sc->vge_ifp;
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
#ifdef DEVICE_POLLING
if (ifp->if_capenable & IFCAP_POLLING)
ether_poll_deregister(ifp);
#endif
/* These should only be active if attach succeeded */
if (device_is_attached(dev)) {
ether_ifdetach(ifp);
VGE_LOCK(sc);
vge_stop(sc);
VGE_UNLOCK(sc);
callout_drain(&sc->vge_watchdog);
}
if (sc->vge_miibus)
device_delete_child(dev, sc->vge_miibus);
bus_generic_detach(dev);
if (sc->vge_intrhand)
bus_teardown_intr(dev, sc->vge_irq, sc->vge_intrhand);
if (sc->vge_irq)
bus_release_resource(dev, SYS_RES_IRQ,
sc->vge_flags & VGE_FLAG_MSI ? 1 : 0, sc->vge_irq);
if (sc->vge_flags & VGE_FLAG_MSI)
pci_release_msi(dev);
if (sc->vge_res)
bus_release_resource(dev, SYS_RES_MEMORY,
PCIR_BAR(1), sc->vge_res);
if (ifp)
if_free(ifp);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
vge_dma_free(sc);
mtx_destroy(&sc->vge_mtx);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
return (0);
}
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
static void
2009-12-14 19:44:54 +00:00
vge_discard_rxbuf(struct vge_softc *sc, int prod)
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
{
2009-12-14 20:07:25 +00:00
struct vge_rxdesc *rxd;
int i;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
rxd = &sc->vge_cdata.vge_rxdesc[prod];
rxd->rx_desc->vge_sts = 0;
rxd->rx_desc->vge_ctl = 0;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
/*
* Note: the manual fails to document the fact that for
* proper opration, the driver needs to replentish the RX
* DMA ring 4 descriptors at a time (rather than one at a
* time, like most chips). We can allocate the new buffers
* but we should not set the OWN bits until we're ready
* to hand back 4 of them in one shot.
*/
if ((prod % VGE_RXCHUNK) == (VGE_RXCHUNK - 1)) {
for (i = VGE_RXCHUNK; i > 0; i--) {
rxd->rx_desc->vge_sts = htole32(VGE_RDSTS_OWN);
rxd = rxd->rxd_prev;
}
sc->vge_cdata.vge_rx_commit += VGE_RXCHUNK;
}
}
static int
2009-12-14 19:44:54 +00:00
vge_newbuf(struct vge_softc *sc, int prod)
{
2009-12-14 20:07:25 +00:00
struct vge_rxdesc *rxd;
struct mbuf *m;
bus_dma_segment_t segs[1];
bus_dmamap_t map;
int i, nsegs;
m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
if (m == NULL)
return (ENOBUFS);
/*
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
* This is part of an evil trick to deal with strict-alignment
* architectures. The VIA chip requires RX buffers to be aligned
* on 32-bit boundaries, but that will hose strict-alignment
* architectures. To get around this, we leave some empty space
* at the start of each buffer and for non-strict-alignment hosts,
* we copy the buffer back two bytes to achieve word alignment.
* This is slightly more efficient than allocating a new buffer,
* copying the contents, and discarding the old buffer.
*/
m->m_len = m->m_pkthdr.len = MCLBYTES;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
m_adj(m, VGE_RX_BUF_ALIGN);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
if (bus_dmamap_load_mbuf_sg(sc->vge_cdata.vge_rx_tag,
sc->vge_cdata.vge_rx_sparemap, m, segs, &nsegs, 0) != 0) {
m_freem(m);
return (ENOBUFS);
}
KASSERT(nsegs == 1, ("%s: %d segments returned!", __func__, nsegs));
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
rxd = &sc->vge_cdata.vge_rxdesc[prod];
if (rxd->rx_m != NULL) {
bus_dmamap_sync(sc->vge_cdata.vge_rx_tag, rxd->rx_dmamap,
BUS_DMASYNC_POSTREAD);
bus_dmamap_unload(sc->vge_cdata.vge_rx_tag, rxd->rx_dmamap);
}
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
map = rxd->rx_dmamap;
rxd->rx_dmamap = sc->vge_cdata.vge_rx_sparemap;
sc->vge_cdata.vge_rx_sparemap = map;
bus_dmamap_sync(sc->vge_cdata.vge_rx_tag, rxd->rx_dmamap,
BUS_DMASYNC_PREREAD);
rxd->rx_m = m;
rxd->rx_desc->vge_sts = 0;
rxd->rx_desc->vge_ctl = 0;
rxd->rx_desc->vge_addrlo = htole32(VGE_ADDR_LO(segs[0].ds_addr));
rxd->rx_desc->vge_addrhi = htole32(VGE_ADDR_HI(segs[0].ds_addr) |
(VGE_BUFLEN(segs[0].ds_len) << 16) | VGE_RXDESC_I);
/*
* Note: the manual fails to document the fact that for
2009-12-14 18:19:56 +00:00
* proper operation, the driver needs to replenish the RX
* DMA ring 4 descriptors at a time (rather than one at a
* time, like most chips). We can allocate the new buffers
* but we should not set the OWN bits until we're ready
* to hand back 4 of them in one shot.
*/
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
if ((prod % VGE_RXCHUNK) == (VGE_RXCHUNK - 1)) {
for (i = VGE_RXCHUNK; i > 0; i--) {
rxd->rx_desc->vge_sts = htole32(VGE_RDSTS_OWN);
rxd = rxd->rxd_prev;
}
sc->vge_cdata.vge_rx_commit += VGE_RXCHUNK;
}
return (0);
}
static int
2009-12-14 19:44:54 +00:00
vge_tx_list_init(struct vge_softc *sc)
{
2009-12-14 20:07:25 +00:00
struct vge_ring_data *rd;
struct vge_txdesc *txd;
int i;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
VGE_LOCK_ASSERT(sc);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
sc->vge_cdata.vge_tx_prodidx = 0;
sc->vge_cdata.vge_tx_considx = 0;
sc->vge_cdata.vge_tx_cnt = 0;
rd = &sc->vge_rdata;
bzero(rd->vge_tx_ring, VGE_TX_LIST_SZ);
for (i = 0; i < VGE_TX_DESC_CNT; i++) {
txd = &sc->vge_cdata.vge_txdesc[i];
txd->tx_m = NULL;
txd->tx_desc = &rd->vge_tx_ring[i];
}
bus_dmamap_sync(sc->vge_cdata.vge_tx_ring_tag,
sc->vge_cdata.vge_tx_ring_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
return (0);
}
static int
2009-12-14 19:44:54 +00:00
vge_rx_list_init(struct vge_softc *sc)
{
2009-12-14 20:07:25 +00:00
struct vge_ring_data *rd;
struct vge_rxdesc *rxd;
int i;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
VGE_LOCK_ASSERT(sc);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
sc->vge_cdata.vge_rx_prodidx = 0;
sc->vge_cdata.vge_head = NULL;
sc->vge_cdata.vge_tail = NULL;
sc->vge_cdata.vge_rx_commit = 0;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
rd = &sc->vge_rdata;
bzero(rd->vge_rx_ring, VGE_RX_LIST_SZ);
for (i = 0; i < VGE_RX_DESC_CNT; i++) {
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
rxd = &sc->vge_cdata.vge_rxdesc[i];
rxd->rx_m = NULL;
rxd->rx_desc = &rd->vge_rx_ring[i];
if (i == 0)
rxd->rxd_prev =
&sc->vge_cdata.vge_rxdesc[VGE_RX_DESC_CNT - 1];
else
rxd->rxd_prev = &sc->vge_cdata.vge_rxdesc[i - 1];
if (vge_newbuf(sc, i) != 0)
return (ENOBUFS);
}
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
bus_dmamap_sync(sc->vge_cdata.vge_rx_ring_tag,
sc->vge_cdata.vge_rx_ring_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
sc->vge_cdata.vge_rx_commit = 0;
return (0);
}
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
static void
2009-12-14 19:44:54 +00:00
vge_freebufs(struct vge_softc *sc)
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
{
2009-12-14 20:07:25 +00:00
struct vge_txdesc *txd;
struct vge_rxdesc *rxd;
struct ifnet *ifp;
int i;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
VGE_LOCK_ASSERT(sc);
ifp = sc->vge_ifp;
/*
* Free RX and TX mbufs still in the queues.
*/
for (i = 0; i < VGE_RX_DESC_CNT; i++) {
rxd = &sc->vge_cdata.vge_rxdesc[i];
if (rxd->rx_m != NULL) {
bus_dmamap_sync(sc->vge_cdata.vge_rx_tag,
rxd->rx_dmamap, BUS_DMASYNC_POSTREAD);
bus_dmamap_unload(sc->vge_cdata.vge_rx_tag,
rxd->rx_dmamap);
m_freem(rxd->rx_m);
rxd->rx_m = NULL;
}
}
for (i = 0; i < VGE_TX_DESC_CNT; i++) {
txd = &sc->vge_cdata.vge_txdesc[i];
if (txd->tx_m != NULL) {
bus_dmamap_sync(sc->vge_cdata.vge_tx_tag,
txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
bus_dmamap_unload(sc->vge_cdata.vge_tx_tag,
txd->tx_dmamap);
m_freem(txd->tx_m);
txd->tx_m = NULL;
ifp->if_oerrors++;
}
}
}
#ifndef __NO_STRICT_ALIGNMENT
static __inline void
2009-12-14 19:44:54 +00:00
vge_fixup_rx(struct mbuf *m)
{
2009-12-14 20:07:25 +00:00
int i;
uint16_t *src, *dst;
src = mtod(m, uint16_t *);
dst = src - 1;
for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++)
*dst++ = *src++;
m->m_data -= ETHER_ALIGN;
}
#endif
/*
* RX handler. We support the reception of jumbo frames that have
* been fragmented across multiple 2K mbuf cluster buffers.
*/
static int
2009-12-14 19:44:54 +00:00
vge_rxeof(struct vge_softc *sc, int count)
{
2009-12-14 20:07:25 +00:00
struct mbuf *m;
struct ifnet *ifp;
int prod, prog, total_len;
struct vge_rxdesc *rxd;
struct vge_rx_desc *cur_rx;
uint32_t rxstat, rxctl;
VGE_LOCK_ASSERT(sc);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
ifp = sc->vge_ifp;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
bus_dmamap_sync(sc->vge_cdata.vge_rx_ring_tag,
sc->vge_cdata.vge_rx_ring_map,
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
prod = sc->vge_cdata.vge_rx_prodidx;
for (prog = 0; count > 0 &&
(ifp->if_drv_flags & IFF_DRV_RUNNING) != 0;
VGE_RX_DESC_INC(prod)) {
cur_rx = &sc->vge_rdata.vge_rx_ring[prod];
rxstat = le32toh(cur_rx->vge_sts);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
if ((rxstat & VGE_RDSTS_OWN) != 0)
break;
count--;
prog++;
rxctl = le32toh(cur_rx->vge_ctl);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
total_len = VGE_RXBYTES(rxstat);
rxd = &sc->vge_cdata.vge_rxdesc[prod];
m = rxd->rx_m;
/*
* If the 'start of frame' bit is set, this indicates
* either the first fragment in a multi-fragment receive,
* or an intermediate fragment. Either way, we want to
* accumulate the buffers.
*/
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
if ((rxstat & VGE_RXPKT_SOF) != 0) {
if (vge_newbuf(sc, prod) != 0) {
ifp->if_iqdrops++;
VGE_CHAIN_RESET(sc);
vge_discard_rxbuf(sc, prod);
continue;
}
m->m_len = MCLBYTES - VGE_RX_BUF_ALIGN;
if (sc->vge_cdata.vge_head == NULL) {
sc->vge_cdata.vge_head = m;
sc->vge_cdata.vge_tail = m;
} else {
m->m_flags &= ~M_PKTHDR;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
sc->vge_cdata.vge_tail->m_next = m;
sc->vge_cdata.vge_tail = m;
}
continue;
}
/*
* Bad/error frames will have the RXOK bit cleared.
* However, there's one error case we want to allow:
* if a VLAN tagged frame arrives and the chip can't
* match it against the CAM filter, it considers this
* a 'VLAN CAM filter miss' and clears the 'RXOK' bit.
* We don't want to drop the frame though: our VLAN
* filtering is done in software.
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
* We also want to receive bad-checksummed frames and
* and frames with bad-length.
*/
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
if ((rxstat & VGE_RDSTS_RXOK) == 0 &&
(rxstat & (VGE_RDSTS_VIDM | VGE_RDSTS_RLERR |
VGE_RDSTS_CSUMERR)) == 0) {
ifp->if_ierrors++;
/*
* If this is part of a multi-fragment packet,
* discard all the pieces.
*/
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
VGE_CHAIN_RESET(sc);
vge_discard_rxbuf(sc, prod);
continue;
}
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
if (vge_newbuf(sc, prod) != 0) {
ifp->if_iqdrops++;
VGE_CHAIN_RESET(sc);
vge_discard_rxbuf(sc, prod);
continue;
}
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
/* Chain received mbufs. */
if (sc->vge_cdata.vge_head != NULL) {
m->m_len = total_len % (MCLBYTES - VGE_RX_BUF_ALIGN);
/*
* Special case: if there's 4 bytes or less
* in this buffer, the mbuf can be discarded:
* the last 4 bytes is the CRC, which we don't
* care about anyway.
*/
if (m->m_len <= ETHER_CRC_LEN) {
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
sc->vge_cdata.vge_tail->m_len -=
(ETHER_CRC_LEN - m->m_len);
m_freem(m);
} else {
m->m_len -= ETHER_CRC_LEN;
m->m_flags &= ~M_PKTHDR;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
sc->vge_cdata.vge_tail->m_next = m;
}
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
m = sc->vge_cdata.vge_head;
m->m_flags |= M_PKTHDR;
m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
} else {
m->m_flags |= M_PKTHDR;
m->m_pkthdr.len = m->m_len =
(total_len - ETHER_CRC_LEN);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
}
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
#ifndef __NO_STRICT_ALIGNMENT
vge_fixup_rx(m);
#endif
m->m_pkthdr.rcvif = ifp;
/* Do RX checksumming if enabled */
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
if ((ifp->if_capenable & IFCAP_RXCSUM) != 0 &&
(rxctl & VGE_RDCTL_FRAG) == 0) {
/* Check IP header checksum */
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
if ((rxctl & VGE_RDCTL_IPPKT) != 0)
m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
if ((rxctl & VGE_RDCTL_IPCSUMOK) != 0)
m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
/* Check TCP/UDP checksum */
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
if (rxctl & (VGE_RDCTL_TCPPKT | VGE_RDCTL_UDPPKT) &&
rxctl & VGE_RDCTL_PROTOCSUMOK) {
m->m_pkthdr.csum_flags |=
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
m->m_pkthdr.csum_data = 0xffff;
}
}
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
if ((rxstat & VGE_RDSTS_VTAG) != 0) {
/*
* The 32-bit rxctl register is stored in little-endian.
* However, the 16-bit vlan tag is stored in big-endian,
* so we have to byte swap it.
*/
m->m_pkthdr.ether_vtag =
bswap16(rxctl & VGE_RDCTL_VLANID);
m->m_flags |= M_VLANTAG;
}
VGE_UNLOCK(sc);
(*ifp->if_input)(ifp, m);
VGE_LOCK(sc);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
sc->vge_cdata.vge_head = NULL;
sc->vge_cdata.vge_tail = NULL;
}
if (prog > 0) {
sc->vge_cdata.vge_rx_prodidx = prod;
bus_dmamap_sync(sc->vge_cdata.vge_rx_ring_tag,
sc->vge_cdata.vge_rx_ring_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
/* Update residue counter. */
if (sc->vge_cdata.vge_rx_commit != 0) {
CSR_WRITE_2(sc, VGE_RXDESC_RESIDUECNT,
sc->vge_cdata.vge_rx_commit);
sc->vge_cdata.vge_rx_commit = 0;
}
}
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
return (prog);
}
static void
2009-12-14 19:44:54 +00:00
vge_txeof(struct vge_softc *sc)
{
2009-12-14 20:07:25 +00:00
struct ifnet *ifp;
struct vge_tx_desc *cur_tx;
struct vge_txdesc *txd;
uint32_t txstat;
int cons, prod;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
VGE_LOCK_ASSERT(sc);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
ifp = sc->vge_ifp;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
if (sc->vge_cdata.vge_tx_cnt == 0)
return;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
bus_dmamap_sync(sc->vge_cdata.vge_tx_ring_tag,
sc->vge_cdata.vge_tx_ring_map,
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
/*
* Go through our tx list and free mbufs for those
* frames that have been transmitted.
*/
cons = sc->vge_cdata.vge_tx_considx;
prod = sc->vge_cdata.vge_tx_prodidx;
for (; cons != prod; VGE_TX_DESC_INC(cons)) {
cur_tx = &sc->vge_rdata.vge_tx_ring[cons];
txstat = le32toh(cur_tx->vge_sts);
if ((txstat & VGE_TDSTS_OWN) != 0)
break;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
sc->vge_cdata.vge_tx_cnt--;
ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
txd = &sc->vge_cdata.vge_txdesc[cons];
bus_dmamap_sync(sc->vge_cdata.vge_tx_tag, txd->tx_dmamap,
BUS_DMASYNC_POSTWRITE);
bus_dmamap_unload(sc->vge_cdata.vge_tx_tag, txd->tx_dmamap);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
KASSERT(txd->tx_m != NULL, ("%s: freeing NULL mbuf!\n",
__func__));
m_freem(txd->tx_m);
txd->tx_m = NULL;
txd->tx_desc->vge_frag[0].vge_addrhi = 0;
}
bus_dmamap_sync(sc->vge_cdata.vge_tx_ring_tag,
sc->vge_cdata.vge_tx_ring_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
sc->vge_cdata.vge_tx_considx = cons;
if (sc->vge_cdata.vge_tx_cnt == 0)
sc->vge_timer = 0;
}
static void
vge_link_statchg(void *xsc)
{
2009-12-14 20:07:25 +00:00
struct vge_softc *sc;
struct ifnet *ifp;
Rework link establishment and link state detection logic. - Remove MIIBUS statchg callback and program VGE_DIAGCTL before initiating link establishment. Previously driver used to program VGE_DIAGCTL after getting a link in statchg callback. It seems the VGE_DIAGCTL register works like a kind of MII register such that it requires setting a 'to be' mode in advance rather than relying on resolved speed/duplex of established link. This means the statchg callback is not needed in driver. In addition, if there was no link at the time of media change, this was not called at all. - Introduce vge_ifmedia_upd_locked() to change current media to configured one. Actual media change is performed only after PHY reset and VGE_DIAGCTL setup. - In WOL configuration, make sure to clear forced mode such that controller can rely on auto-negotiation. - Unlike most other drivers that use miibus(4), vge(4) used controller's auto-polling feature for link state tracking via interrupt. This came from controller's inefficient mechanism to access MII registers. On link state change interrupt, vge(4) used to get current link state with series of MII register accesses. Because vge(4) already enabled auto polling, read PHY status register to resolved speed/duplex/flow control parameters. vge(4) still does not drive MII_TICK to reduce number of MII register accesses which in turn means the driver does not know the status of auto-negotiation. This was a one of long standing issue of vge(4). Probably driver may be able to implement a timer that keeps track of auto-negotiation state and restart auto-negotiation when driver couldn't establish a link within a specified period. However the controller does not provide a reliable way to detect auto-negotiation failure so I'm not sure whether it's worth to implement it in driver. Alternatively driver can completely disable MII auto-polling and let miibus(4) poll link state by driving MII_TICK. This may reduce unnecessary overhead of stopping/restarting MII auto-polling of controller. Unfortunately it was known that some variants of controller does not work correctly if MII auto-polling is disabled.
2011-11-22 20:45:09 +00:00
uint8_t physts;
sc = xsc;
ifp = sc->vge_ifp;
VGE_LOCK_ASSERT(sc);
Rework link establishment and link state detection logic. - Remove MIIBUS statchg callback and program VGE_DIAGCTL before initiating link establishment. Previously driver used to program VGE_DIAGCTL after getting a link in statchg callback. It seems the VGE_DIAGCTL register works like a kind of MII register such that it requires setting a 'to be' mode in advance rather than relying on resolved speed/duplex of established link. This means the statchg callback is not needed in driver. In addition, if there was no link at the time of media change, this was not called at all. - Introduce vge_ifmedia_upd_locked() to change current media to configured one. Actual media change is performed only after PHY reset and VGE_DIAGCTL setup. - In WOL configuration, make sure to clear forced mode such that controller can rely on auto-negotiation. - Unlike most other drivers that use miibus(4), vge(4) used controller's auto-polling feature for link state tracking via interrupt. This came from controller's inefficient mechanism to access MII registers. On link state change interrupt, vge(4) used to get current link state with series of MII register accesses. Because vge(4) already enabled auto polling, read PHY status register to resolved speed/duplex/flow control parameters. vge(4) still does not drive MII_TICK to reduce number of MII register accesses which in turn means the driver does not know the status of auto-negotiation. This was a one of long standing issue of vge(4). Probably driver may be able to implement a timer that keeps track of auto-negotiation state and restart auto-negotiation when driver couldn't establish a link within a specified period. However the controller does not provide a reliable way to detect auto-negotiation failure so I'm not sure whether it's worth to implement it in driver. Alternatively driver can completely disable MII auto-polling and let miibus(4) poll link state by driving MII_TICK. This may reduce unnecessary overhead of stopping/restarting MII auto-polling of controller. Unfortunately it was known that some variants of controller does not work correctly if MII auto-polling is disabled.
2011-11-22 20:45:09 +00:00
physts = CSR_READ_1(sc, VGE_PHYSTS0);
if ((physts & VGE_PHYSTS_RESETSTS) == 0) {
if ((physts & VGE_PHYSTS_LINK) == 0) {
sc->vge_flags &= ~VGE_FLAG_LINK;
if_link_state_change(sc->vge_ifp,
LINK_STATE_DOWN);
Rework link establishment and link state detection logic. - Remove MIIBUS statchg callback and program VGE_DIAGCTL before initiating link establishment. Previously driver used to program VGE_DIAGCTL after getting a link in statchg callback. It seems the VGE_DIAGCTL register works like a kind of MII register such that it requires setting a 'to be' mode in advance rather than relying on resolved speed/duplex of established link. This means the statchg callback is not needed in driver. In addition, if there was no link at the time of media change, this was not called at all. - Introduce vge_ifmedia_upd_locked() to change current media to configured one. Actual media change is performed only after PHY reset and VGE_DIAGCTL setup. - In WOL configuration, make sure to clear forced mode such that controller can rely on auto-negotiation. - Unlike most other drivers that use miibus(4), vge(4) used controller's auto-polling feature for link state tracking via interrupt. This came from controller's inefficient mechanism to access MII registers. On link state change interrupt, vge(4) used to get current link state with series of MII register accesses. Because vge(4) already enabled auto polling, read PHY status register to resolved speed/duplex/flow control parameters. vge(4) still does not drive MII_TICK to reduce number of MII register accesses which in turn means the driver does not know the status of auto-negotiation. This was a one of long standing issue of vge(4). Probably driver may be able to implement a timer that keeps track of auto-negotiation state and restart auto-negotiation when driver couldn't establish a link within a specified period. However the controller does not provide a reliable way to detect auto-negotiation failure so I'm not sure whether it's worth to implement it in driver. Alternatively driver can completely disable MII auto-polling and let miibus(4) poll link state by driving MII_TICK. This may reduce unnecessary overhead of stopping/restarting MII auto-polling of controller. Unfortunately it was known that some variants of controller does not work correctly if MII auto-polling is disabled.
2011-11-22 20:45:09 +00:00
} else {
sc->vge_flags |= VGE_FLAG_LINK;
if_link_state_change(sc->vge_ifp,
LINK_STATE_UP);
Rework link establishment and link state detection logic. - Remove MIIBUS statchg callback and program VGE_DIAGCTL before initiating link establishment. Previously driver used to program VGE_DIAGCTL after getting a link in statchg callback. It seems the VGE_DIAGCTL register works like a kind of MII register such that it requires setting a 'to be' mode in advance rather than relying on resolved speed/duplex of established link. This means the statchg callback is not needed in driver. In addition, if there was no link at the time of media change, this was not called at all. - Introduce vge_ifmedia_upd_locked() to change current media to configured one. Actual media change is performed only after PHY reset and VGE_DIAGCTL setup. - In WOL configuration, make sure to clear forced mode such that controller can rely on auto-negotiation. - Unlike most other drivers that use miibus(4), vge(4) used controller's auto-polling feature for link state tracking via interrupt. This came from controller's inefficient mechanism to access MII registers. On link state change interrupt, vge(4) used to get current link state with series of MII register accesses. Because vge(4) already enabled auto polling, read PHY status register to resolved speed/duplex/flow control parameters. vge(4) still does not drive MII_TICK to reduce number of MII register accesses which in turn means the driver does not know the status of auto-negotiation. This was a one of long standing issue of vge(4). Probably driver may be able to implement a timer that keeps track of auto-negotiation state and restart auto-negotiation when driver couldn't establish a link within a specified period. However the controller does not provide a reliable way to detect auto-negotiation failure so I'm not sure whether it's worth to implement it in driver. Alternatively driver can completely disable MII auto-polling and let miibus(4) poll link state by driving MII_TICK. This may reduce unnecessary overhead of stopping/restarting MII auto-polling of controller. Unfortunately it was known that some variants of controller does not work correctly if MII auto-polling is disabled.
2011-11-22 20:45:09 +00:00
CSR_WRITE_1(sc, VGE_CRC2, VGE_CR2_FDX_TXFLOWCTL_ENABLE |
VGE_CR2_FDX_RXFLOWCTL_ENABLE);
if ((physts & VGE_PHYSTS_FDX) != 0) {
if ((physts & VGE_PHYSTS_TXFLOWCAP) != 0)
CSR_WRITE_1(sc, VGE_CRS2,
VGE_CR2_FDX_TXFLOWCTL_ENABLE);
if ((physts & VGE_PHYSTS_RXFLOWCAP) != 0)
CSR_WRITE_1(sc, VGE_CRS2,
VGE_CR2_FDX_RXFLOWCTL_ENABLE);
}
if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
vge_start_locked(ifp);
}
}
Rework link establishment and link state detection logic. - Remove MIIBUS statchg callback and program VGE_DIAGCTL before initiating link establishment. Previously driver used to program VGE_DIAGCTL after getting a link in statchg callback. It seems the VGE_DIAGCTL register works like a kind of MII register such that it requires setting a 'to be' mode in advance rather than relying on resolved speed/duplex of established link. This means the statchg callback is not needed in driver. In addition, if there was no link at the time of media change, this was not called at all. - Introduce vge_ifmedia_upd_locked() to change current media to configured one. Actual media change is performed only after PHY reset and VGE_DIAGCTL setup. - In WOL configuration, make sure to clear forced mode such that controller can rely on auto-negotiation. - Unlike most other drivers that use miibus(4), vge(4) used controller's auto-polling feature for link state tracking via interrupt. This came from controller's inefficient mechanism to access MII registers. On link state change interrupt, vge(4) used to get current link state with series of MII register accesses. Because vge(4) already enabled auto polling, read PHY status register to resolved speed/duplex/flow control parameters. vge(4) still does not drive MII_TICK to reduce number of MII register accesses which in turn means the driver does not know the status of auto-negotiation. This was a one of long standing issue of vge(4). Probably driver may be able to implement a timer that keeps track of auto-negotiation state and restart auto-negotiation when driver couldn't establish a link within a specified period. However the controller does not provide a reliable way to detect auto-negotiation failure so I'm not sure whether it's worth to implement it in driver. Alternatively driver can completely disable MII auto-polling and let miibus(4) poll link state by driving MII_TICK. This may reduce unnecessary overhead of stopping/restarting MII auto-polling of controller. Unfortunately it was known that some variants of controller does not work correctly if MII auto-polling is disabled.
2011-11-22 20:45:09 +00:00
/*
* Restart MII auto-polling because link state change interrupt
* will disable it.
*/
vge_miipoll_start(sc);
}
#ifdef DEVICE_POLLING
static int
vge_poll (struct ifnet *ifp, enum poll_cmd cmd, int count)
{
struct vge_softc *sc = ifp->if_softc;
int rx_npkts = 0;
VGE_LOCK(sc);
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
goto done;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
rx_npkts = vge_rxeof(sc, count);
vge_txeof(sc);
if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
vge_start_locked(ifp);
if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
2009-12-14 19:53:57 +00:00
uint32_t status;
status = CSR_READ_4(sc, VGE_ISR);
if (status == 0xFFFFFFFF)
goto done;
if (status)
CSR_WRITE_4(sc, VGE_ISR, status);
/*
* XXX check behaviour on receiver stalls.
*/
if (status & VGE_ISR_TXDMA_STALL ||
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
status & VGE_ISR_RXDMA_STALL) {
ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
vge_init_locked(sc);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
}
if (status & (VGE_ISR_RXOFLOW|VGE_ISR_RXNODESC)) {
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
vge_rxeof(sc, count);
CSR_WRITE_1(sc, VGE_RXQCSRS, VGE_RXQCSR_RUN);
CSR_WRITE_1(sc, VGE_RXQCSRS, VGE_RXQCSR_WAK);
}
}
done:
VGE_UNLOCK(sc);
return (rx_npkts);
}
#endif /* DEVICE_POLLING */
static void
2009-12-14 19:44:54 +00:00
vge_intr(void *arg)
{
2009-12-14 20:07:25 +00:00
struct vge_softc *sc;
struct ifnet *ifp;
uint32_t status;
sc = arg;
VGE_LOCK(sc);
ifp = sc->vge_ifp;
if ((sc->vge_flags & VGE_FLAG_SUSPENDED) != 0 ||
(ifp->if_flags & IFF_UP) == 0) {
VGE_UNLOCK(sc);
return;
}
#ifdef DEVICE_POLLING
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
if (ifp->if_capenable & IFCAP_POLLING) {
status = CSR_READ_4(sc, VGE_ISR);
CSR_WRITE_4(sc, VGE_ISR, status);
if (status != 0xFFFFFFFF && (status & VGE_ISR_LINKSTS) != 0)
vge_link_statchg(sc);
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
VGE_UNLOCK(sc);
return;
}
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
#endif
/* Disable interrupts */
CSR_WRITE_1(sc, VGE_CRC3, VGE_CR3_INT_GMSK);
Implement interrupt moderation scheme supported by VT61xx controllers. TX/RX interrupt mitigation is controlled by VGE_TXSUPPTHR and VGE_RXSUPPTHR register. These registers suppress generation of interrupts until the programmed frames counter equals to the registers. VT61xx also supports interrupt hold off timer register. If this interrupt hold off timer is active all interrupts would be disabled until the timer reaches to 0. The timer value is reloaded whenever VGE_ISR register written. The timer resolution is about 20us. Previously vge(4) used single shot timer to reduce Tx completion interrupts. This required VGE_CRS1 register access in Tx start/completion handler to rearm new timeout value and it did not show satisfactory result(more than 50k interrupts under load). Rx interrupts was not moderated at all such that vge(4) used to generate too many interrupts which in turn made polling(4) better approach under high network load. This change activates all interrupt moderation mechanism and initial values were tuned to generate interrupt less than 8k per second. That number of interrupts wouldn't add additional packet latencies compared to polling(4). These interrupt parameters could be changed with sysctl. dev.vge.%d.int_holdoff dev.vge.%d.rx_coal_pkt dev.vge.%d.tx_coal_pkt Interface has be brought down and up again before change take effect. With interrupt moderation there is no more need to loop in interrupt handler. This loop always added one more register access. While I'm here remove dead code which tried to implement subset of interrupt moderation.
2009-12-17 18:00:25 +00:00
status = CSR_READ_4(sc, VGE_ISR);
CSR_WRITE_4(sc, VGE_ISR, status | VGE_ISR_HOLDOFF_RELOAD);
/* If the card has gone away the read returns 0xffff. */
if (status == 0xFFFFFFFF || (status & VGE_INTRS) == 0)
goto done;
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
if (status & (VGE_ISR_RXOK|VGE_ISR_RXOK_HIPRIO))
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
vge_rxeof(sc, VGE_RX_DESC_CNT);
if (status & (VGE_ISR_RXOFLOW|VGE_ISR_RXNODESC)) {
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
vge_rxeof(sc, VGE_RX_DESC_CNT);
CSR_WRITE_1(sc, VGE_RXQCSRS, VGE_RXQCSR_RUN);
CSR_WRITE_1(sc, VGE_RXQCSRS, VGE_RXQCSR_WAK);
}
Implement interrupt moderation scheme supported by VT61xx controllers. TX/RX interrupt mitigation is controlled by VGE_TXSUPPTHR and VGE_RXSUPPTHR register. These registers suppress generation of interrupts until the programmed frames counter equals to the registers. VT61xx also supports interrupt hold off timer register. If this interrupt hold off timer is active all interrupts would be disabled until the timer reaches to 0. The timer value is reloaded whenever VGE_ISR register written. The timer resolution is about 20us. Previously vge(4) used single shot timer to reduce Tx completion interrupts. This required VGE_CRS1 register access in Tx start/completion handler to rearm new timeout value and it did not show satisfactory result(more than 50k interrupts under load). Rx interrupts was not moderated at all such that vge(4) used to generate too many interrupts which in turn made polling(4) better approach under high network load. This change activates all interrupt moderation mechanism and initial values were tuned to generate interrupt less than 8k per second. That number of interrupts wouldn't add additional packet latencies compared to polling(4). These interrupt parameters could be changed with sysctl. dev.vge.%d.int_holdoff dev.vge.%d.rx_coal_pkt dev.vge.%d.tx_coal_pkt Interface has be brought down and up again before change take effect. With interrupt moderation there is no more need to loop in interrupt handler. This loop always added one more register access. While I'm here remove dead code which tried to implement subset of interrupt moderation.
2009-12-17 18:00:25 +00:00
if (status & (VGE_ISR_TXOK0|VGE_ISR_TXOK_HIPRIO))
vge_txeof(sc);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
if (status & (VGE_ISR_TXDMA_STALL|VGE_ISR_RXDMA_STALL)) {
ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
vge_init_locked(sc);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
}
if (status & VGE_ISR_LINKSTS)
vge_link_statchg(sc);
}
Implement interrupt moderation scheme supported by VT61xx controllers. TX/RX interrupt mitigation is controlled by VGE_TXSUPPTHR and VGE_RXSUPPTHR register. These registers suppress generation of interrupts until the programmed frames counter equals to the registers. VT61xx also supports interrupt hold off timer register. If this interrupt hold off timer is active all interrupts would be disabled until the timer reaches to 0. The timer value is reloaded whenever VGE_ISR register written. The timer resolution is about 20us. Previously vge(4) used single shot timer to reduce Tx completion interrupts. This required VGE_CRS1 register access in Tx start/completion handler to rearm new timeout value and it did not show satisfactory result(more than 50k interrupts under load). Rx interrupts was not moderated at all such that vge(4) used to generate too many interrupts which in turn made polling(4) better approach under high network load. This change activates all interrupt moderation mechanism and initial values were tuned to generate interrupt less than 8k per second. That number of interrupts wouldn't add additional packet latencies compared to polling(4). These interrupt parameters could be changed with sysctl. dev.vge.%d.int_holdoff dev.vge.%d.rx_coal_pkt dev.vge.%d.tx_coal_pkt Interface has be brought down and up again before change take effect. With interrupt moderation there is no more need to loop in interrupt handler. This loop always added one more register access. While I'm here remove dead code which tried to implement subset of interrupt moderation.
2009-12-17 18:00:25 +00:00
done:
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
/* Re-enable interrupts */
CSR_WRITE_1(sc, VGE_CRS3, VGE_CR3_INT_GMSK);
Implement interrupt moderation scheme supported by VT61xx controllers. TX/RX interrupt mitigation is controlled by VGE_TXSUPPTHR and VGE_RXSUPPTHR register. These registers suppress generation of interrupts until the programmed frames counter equals to the registers. VT61xx also supports interrupt hold off timer register. If this interrupt hold off timer is active all interrupts would be disabled until the timer reaches to 0. The timer value is reloaded whenever VGE_ISR register written. The timer resolution is about 20us. Previously vge(4) used single shot timer to reduce Tx completion interrupts. This required VGE_CRS1 register access in Tx start/completion handler to rearm new timeout value and it did not show satisfactory result(more than 50k interrupts under load). Rx interrupts was not moderated at all such that vge(4) used to generate too many interrupts which in turn made polling(4) better approach under high network load. This change activates all interrupt moderation mechanism and initial values were tuned to generate interrupt less than 8k per second. That number of interrupts wouldn't add additional packet latencies compared to polling(4). These interrupt parameters could be changed with sysctl. dev.vge.%d.int_holdoff dev.vge.%d.rx_coal_pkt dev.vge.%d.tx_coal_pkt Interface has be brought down and up again before change take effect. With interrupt moderation there is no more need to loop in interrupt handler. This loop always added one more register access. While I'm here remove dead code which tried to implement subset of interrupt moderation.
2009-12-17 18:00:25 +00:00
if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
vge_start_locked(ifp);
}
VGE_UNLOCK(sc);
}
static int
2009-12-14 19:44:54 +00:00
vge_encap(struct vge_softc *sc, struct mbuf **m_head)
{
2009-12-14 20:07:25 +00:00
struct vge_txdesc *txd;
struct vge_tx_frag *frag;
struct mbuf *m;
bus_dma_segment_t txsegs[VGE_MAXTXSEGS];
int error, i, nsegs, padlen;
uint32_t cflags;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
VGE_LOCK_ASSERT(sc);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
M_ASSERTPKTHDR((*m_head));
/* Argh. This chip does not autopad short frames. */
if ((*m_head)->m_pkthdr.len < VGE_MIN_FRAMELEN) {
m = *m_head;
padlen = VGE_MIN_FRAMELEN - m->m_pkthdr.len;
if (M_WRITABLE(m) == 0) {
/* Get a writable copy. */
m = m_dup(*m_head, M_NOWAIT);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
m_freem(*m_head);
if (m == NULL) {
*m_head = NULL;
return (ENOBUFS);
}
*m_head = m;
}
if (M_TRAILINGSPACE(m) < padlen) {
m = m_defrag(m, M_NOWAIT);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
if (m == NULL) {
m_freem(*m_head);
*m_head = NULL;
return (ENOBUFS);
}
}
/*
* Manually pad short frames, and zero the pad space
* to avoid leaking data.
*/
bzero(mtod(m, char *) + m->m_pkthdr.len, padlen);
m->m_pkthdr.len += padlen;
m->m_len = m->m_pkthdr.len;
*m_head = m;
}
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
txd = &sc->vge_cdata.vge_txdesc[sc->vge_cdata.vge_tx_prodidx];
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
error = bus_dmamap_load_mbuf_sg(sc->vge_cdata.vge_tx_tag,
txd->tx_dmamap, *m_head, txsegs, &nsegs, 0);
if (error == EFBIG) {
m = m_collapse(*m_head, M_NOWAIT, VGE_MAXTXSEGS);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
if (m == NULL) {
m_freem(*m_head);
*m_head = NULL;
return (ENOMEM);
}
*m_head = m;
error = bus_dmamap_load_mbuf_sg(sc->vge_cdata.vge_tx_tag,
txd->tx_dmamap, *m_head, txsegs, &nsegs, 0);
if (error != 0) {
m_freem(*m_head);
*m_head = NULL;
return (error);
}
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
} else if (error != 0)
return (error);
bus_dmamap_sync(sc->vge_cdata.vge_tx_tag, txd->tx_dmamap,
BUS_DMASYNC_PREWRITE);
m = *m_head;
cflags = 0;
/* Configure checksum offload. */
if ((m->m_pkthdr.csum_flags & CSUM_IP) != 0)
cflags |= VGE_TDCTL_IPCSUM;
if ((m->m_pkthdr.csum_flags & CSUM_TCP) != 0)
cflags |= VGE_TDCTL_TCPCSUM;
if ((m->m_pkthdr.csum_flags & CSUM_UDP) != 0)
cflags |= VGE_TDCTL_UDPCSUM;
/* Configure VLAN. */
if ((m->m_flags & M_VLANTAG) != 0)
cflags |= m->m_pkthdr.ether_vtag | VGE_TDCTL_VTAG;
txd->tx_desc->vge_sts = htole32(m->m_pkthdr.len << 16);
/*
* XXX
* Velocity family seems to support TSO but no information
* for MSS configuration is available. Also the number of
* fragments supported by a descriptor is too small to hold
* entire 64KB TCP/IP segment. Maybe VGE_TD_LS_MOF,
* VGE_TD_LS_SOF and VGE_TD_LS_EOF could be used to build
* longer chain of buffers but no additional information is
* available.
*
* When telling the chip how many segments there are, we
* must use nsegs + 1 instead of just nsegs. Darned if I
* know why. This also means we can't use the last fragment
* field of Tx descriptor.
*/
txd->tx_desc->vge_ctl = htole32(cflags | ((nsegs + 1) << 28) |
VGE_TD_LS_NORM);
for (i = 0; i < nsegs; i++) {
frag = &txd->tx_desc->vge_frag[i];
frag->vge_addrlo = htole32(VGE_ADDR_LO(txsegs[i].ds_addr));
frag->vge_addrhi = htole32(VGE_ADDR_HI(txsegs[i].ds_addr) |
(VGE_BUFLEN(txsegs[i].ds_len) << 16));
}
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
sc->vge_cdata.vge_tx_cnt++;
VGE_TX_DESC_INC(sc->vge_cdata.vge_tx_prodidx);
/*
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
* Finally request interrupt and give the first descriptor
* ownership to hardware.
*/
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
txd->tx_desc->vge_ctl |= htole32(VGE_TDCTL_TIC);
txd->tx_desc->vge_sts |= htole32(VGE_TDSTS_OWN);
txd->tx_m = m;
return (0);
}
/*
* Main transmit routine.
*/
static void
2009-12-14 19:44:54 +00:00
vge_start(struct ifnet *ifp)
{
2009-12-14 20:07:25 +00:00
struct vge_softc *sc;
sc = ifp->if_softc;
VGE_LOCK(sc);
vge_start_locked(ifp);
VGE_UNLOCK(sc);
}
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
static void
2009-12-14 19:44:54 +00:00
vge_start_locked(struct ifnet *ifp)
{
2009-12-14 20:07:25 +00:00
struct vge_softc *sc;
struct vge_txdesc *txd;
struct mbuf *m_head;
int enq, idx;
sc = ifp->if_softc;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
VGE_LOCK_ASSERT(sc);
if ((sc->vge_flags & VGE_FLAG_LINK) == 0 ||
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
(ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
IFF_DRV_RUNNING)
return;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
idx = sc->vge_cdata.vge_tx_prodidx;
VGE_TX_DESC_DEC(idx);
for (enq = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) &&
sc->vge_cdata.vge_tx_cnt < VGE_TX_DESC_CNT - 1; ) {
IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
if (m_head == NULL)
break;
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
/*
* Pack the data into the transmit ring. If we
* don't have room, set the OACTIVE flag and wait
* for the NIC to drain the ring.
*/
if (vge_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;
}
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
txd = &sc->vge_cdata.vge_txdesc[idx];
txd->tx_desc->vge_frag[0].vge_addrhi |= htole32(VGE_TXDESC_Q);
VGE_TX_DESC_INC(idx);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
enq++;
/*
* If there's a BPF listener, bounce a copy of this frame
* to him.
*/
ETHER_BPF_MTAP(ifp, m_head);
}
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
if (enq > 0) {
bus_dmamap_sync(sc->vge_cdata.vge_tx_ring_tag,
sc->vge_cdata.vge_tx_ring_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
/* Issue a transmit command. */
CSR_WRITE_2(sc, VGE_TXQCSRS, VGE_TXQCSR_WAK0);
/*
* Set a timeout in case the chip goes out to lunch.
*/
sc->vge_timer = 5;
}
}
static void
2009-12-14 19:44:54 +00:00
vge_init(void *xsc)
{
2009-12-14 20:07:25 +00:00
struct vge_softc *sc = xsc;
VGE_LOCK(sc);
vge_init_locked(sc);
VGE_UNLOCK(sc);
}
static void
vge_init_locked(struct vge_softc *sc)
{
2009-12-14 20:07:25 +00:00
struct ifnet *ifp = sc->vge_ifp;
int error, i;
VGE_LOCK_ASSERT(sc);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
return;
/*
* Cancel pending I/O and free all RX/TX buffers.
*/
vge_stop(sc);
vge_reset(sc);
vge_miipoll_start(sc);
/*
* Initialize the RX and TX descriptors and mbufs.
*/
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
error = vge_rx_list_init(sc);
if (error != 0) {
device_printf(sc->vge_dev, "no memory for Rx buffers.\n");
return;
}
vge_tx_list_init(sc);
/* Clear MAC statistics. */
vge_stats_clear(sc);
/* Set our station address */
for (i = 0; i < ETHER_ADDR_LEN; i++)
CSR_WRITE_1(sc, VGE_PAR0 + i, IF_LLADDR(sc->vge_ifp)[i]);
/*
* Set receive FIFO threshold. Also allow transmission and
* reception of VLAN tagged frames.
*/
CSR_CLRBIT_1(sc, VGE_RXCFG, VGE_RXCFG_FIFO_THR|VGE_RXCFG_VTAGOPT);
CSR_SETBIT_1(sc, VGE_RXCFG, VGE_RXFIFOTHR_128BYTES);
/* Set DMA burst length */
CSR_CLRBIT_1(sc, VGE_DMACFG0, VGE_DMACFG0_BURSTLEN);
CSR_SETBIT_1(sc, VGE_DMACFG0, VGE_DMABURST_128);
CSR_SETBIT_1(sc, VGE_TXCFG, VGE_TXCFG_ARB_PRIO|VGE_TXCFG_NONBLK);
/* Set collision backoff algorithm */
CSR_CLRBIT_1(sc, VGE_CHIPCFG1, VGE_CHIPCFG1_CRANDOM|
VGE_CHIPCFG1_CAP|VGE_CHIPCFG1_MBA|VGE_CHIPCFG1_BAKOPT);
CSR_SETBIT_1(sc, VGE_CHIPCFG1, VGE_CHIPCFG1_OFSET);
/* Disable LPSEL field in priority resolution */
CSR_SETBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_LPSEL_DIS);
/*
* Load the addresses of the DMA queues into the chip.
* Note that we only use one transmit queue.
*/
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
CSR_WRITE_4(sc, VGE_TXDESC_HIADDR,
VGE_ADDR_HI(sc->vge_rdata.vge_tx_ring_paddr));
CSR_WRITE_4(sc, VGE_TXDESC_ADDR_LO0,
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
VGE_ADDR_LO(sc->vge_rdata.vge_tx_ring_paddr));
CSR_WRITE_2(sc, VGE_TXDESCNUM, VGE_TX_DESC_CNT - 1);
CSR_WRITE_4(sc, VGE_RXDESC_ADDR_LO,
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
VGE_ADDR_LO(sc->vge_rdata.vge_rx_ring_paddr));
CSR_WRITE_2(sc, VGE_RXDESCNUM, VGE_RX_DESC_CNT - 1);
CSR_WRITE_2(sc, VGE_RXDESC_RESIDUECNT, VGE_RX_DESC_CNT);
Implement interrupt moderation scheme supported by VT61xx controllers. TX/RX interrupt mitigation is controlled by VGE_TXSUPPTHR and VGE_RXSUPPTHR register. These registers suppress generation of interrupts until the programmed frames counter equals to the registers. VT61xx also supports interrupt hold off timer register. If this interrupt hold off timer is active all interrupts would be disabled until the timer reaches to 0. The timer value is reloaded whenever VGE_ISR register written. The timer resolution is about 20us. Previously vge(4) used single shot timer to reduce Tx completion interrupts. This required VGE_CRS1 register access in Tx start/completion handler to rearm new timeout value and it did not show satisfactory result(more than 50k interrupts under load). Rx interrupts was not moderated at all such that vge(4) used to generate too many interrupts which in turn made polling(4) better approach under high network load. This change activates all interrupt moderation mechanism and initial values were tuned to generate interrupt less than 8k per second. That number of interrupts wouldn't add additional packet latencies compared to polling(4). These interrupt parameters could be changed with sysctl. dev.vge.%d.int_holdoff dev.vge.%d.rx_coal_pkt dev.vge.%d.tx_coal_pkt Interface has be brought down and up again before change take effect. With interrupt moderation there is no more need to loop in interrupt handler. This loop always added one more register access. While I'm here remove dead code which tried to implement subset of interrupt moderation.
2009-12-17 18:00:25 +00:00
/* Configure interrupt moderation. */
vge_intr_holdoff(sc);
/* Enable and wake up the RX descriptor queue */
CSR_WRITE_1(sc, VGE_RXQCSRS, VGE_RXQCSR_RUN);
CSR_WRITE_1(sc, VGE_RXQCSRS, VGE_RXQCSR_WAK);
/* Enable the TX descriptor queue */
CSR_WRITE_2(sc, VGE_TXQCSRS, VGE_TXQCSR_RUN0);
/* Init the cam filter. */
vge_cam_clear(sc);
/* Set up receiver filter. */
vge_rxfilter(sc);
vge_setvlan(sc);
/* Initialize pause timer. */
CSR_WRITE_2(sc, VGE_TX_PAUSE_TIMER, 0xFFFF);
/*
* Initialize flow control parameters.
* TX XON high threshold : 48
* TX pause low threshold : 24
* Disable hald-duplex flow control
*/
CSR_WRITE_1(sc, VGE_CRC2, 0xFF);
CSR_WRITE_1(sc, VGE_CRS2, VGE_CR2_XON_ENABLE | 0x0B);
/* Enable jumbo frame reception (if desired) */
/* Start the MAC. */
CSR_WRITE_1(sc, VGE_CRC0, VGE_CR0_STOP);
CSR_WRITE_1(sc, VGE_CRS1, VGE_CR1_NOPOLL);
CSR_WRITE_1(sc, VGE_CRS0,
VGE_CR0_TX_ENABLE|VGE_CR0_RX_ENABLE|VGE_CR0_START);
#ifdef DEVICE_POLLING
/*
* Disable interrupts except link state change if we are polling.
*/
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
if (ifp->if_capenable & IFCAP_POLLING) {
CSR_WRITE_4(sc, VGE_IMR, VGE_INTRS_POLLING);
} else /* otherwise ... */
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
#endif
{
/*
* Enable interrupts.
*/
CSR_WRITE_4(sc, VGE_IMR, VGE_INTRS);
}
CSR_WRITE_4(sc, VGE_ISR, 0xFFFFFFFF);
CSR_WRITE_1(sc, VGE_CRS3, VGE_CR3_INT_GMSK);
sc->vge_flags &= ~VGE_FLAG_LINK;
Rework link establishment and link state detection logic. - Remove MIIBUS statchg callback and program VGE_DIAGCTL before initiating link establishment. Previously driver used to program VGE_DIAGCTL after getting a link in statchg callback. It seems the VGE_DIAGCTL register works like a kind of MII register such that it requires setting a 'to be' mode in advance rather than relying on resolved speed/duplex of established link. This means the statchg callback is not needed in driver. In addition, if there was no link at the time of media change, this was not called at all. - Introduce vge_ifmedia_upd_locked() to change current media to configured one. Actual media change is performed only after PHY reset and VGE_DIAGCTL setup. - In WOL configuration, make sure to clear forced mode such that controller can rely on auto-negotiation. - Unlike most other drivers that use miibus(4), vge(4) used controller's auto-polling feature for link state tracking via interrupt. This came from controller's inefficient mechanism to access MII registers. On link state change interrupt, vge(4) used to get current link state with series of MII register accesses. Because vge(4) already enabled auto polling, read PHY status register to resolved speed/duplex/flow control parameters. vge(4) still does not drive MII_TICK to reduce number of MII register accesses which in turn means the driver does not know the status of auto-negotiation. This was a one of long standing issue of vge(4). Probably driver may be able to implement a timer that keeps track of auto-negotiation state and restart auto-negotiation when driver couldn't establish a link within a specified period. However the controller does not provide a reliable way to detect auto-negotiation failure so I'm not sure whether it's worth to implement it in driver. Alternatively driver can completely disable MII auto-polling and let miibus(4) poll link state by driving MII_TICK. This may reduce unnecessary overhead of stopping/restarting MII auto-polling of controller. Unfortunately it was known that some variants of controller does not work correctly if MII auto-polling is disabled.
2011-11-22 20:45:09 +00:00
vge_ifmedia_upd_locked(sc);
ifp->if_drv_flags |= IFF_DRV_RUNNING;
ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
callout_reset(&sc->vge_watchdog, hz, vge_watchdog, sc);
}
/*
* Set media options.
*/
static int
2009-12-14 19:44:54 +00:00
vge_ifmedia_upd(struct ifnet *ifp)
{
2009-12-14 20:07:25 +00:00
struct vge_softc *sc;
int error;
sc = ifp->if_softc;
VGE_LOCK(sc);
Rework link establishment and link state detection logic. - Remove MIIBUS statchg callback and program VGE_DIAGCTL before initiating link establishment. Previously driver used to program VGE_DIAGCTL after getting a link in statchg callback. It seems the VGE_DIAGCTL register works like a kind of MII register such that it requires setting a 'to be' mode in advance rather than relying on resolved speed/duplex of established link. This means the statchg callback is not needed in driver. In addition, if there was no link at the time of media change, this was not called at all. - Introduce vge_ifmedia_upd_locked() to change current media to configured one. Actual media change is performed only after PHY reset and VGE_DIAGCTL setup. - In WOL configuration, make sure to clear forced mode such that controller can rely on auto-negotiation. - Unlike most other drivers that use miibus(4), vge(4) used controller's auto-polling feature for link state tracking via interrupt. This came from controller's inefficient mechanism to access MII registers. On link state change interrupt, vge(4) used to get current link state with series of MII register accesses. Because vge(4) already enabled auto polling, read PHY status register to resolved speed/duplex/flow control parameters. vge(4) still does not drive MII_TICK to reduce number of MII register accesses which in turn means the driver does not know the status of auto-negotiation. This was a one of long standing issue of vge(4). Probably driver may be able to implement a timer that keeps track of auto-negotiation state and restart auto-negotiation when driver couldn't establish a link within a specified period. However the controller does not provide a reliable way to detect auto-negotiation failure so I'm not sure whether it's worth to implement it in driver. Alternatively driver can completely disable MII auto-polling and let miibus(4) poll link state by driving MII_TICK. This may reduce unnecessary overhead of stopping/restarting MII auto-polling of controller. Unfortunately it was known that some variants of controller does not work correctly if MII auto-polling is disabled.
2011-11-22 20:45:09 +00:00
error = vge_ifmedia_upd_locked(sc);
VGE_UNLOCK(sc);
return (error);
}
static int
vge_ifmedia_upd_locked(struct vge_softc *sc)
{
struct mii_data *mii;
struct mii_softc *miisc;
int error;
mii = device_get_softc(sc->vge_miibus);
Rework link establishment and link state detection logic. - Remove MIIBUS statchg callback and program VGE_DIAGCTL before initiating link establishment. Previously driver used to program VGE_DIAGCTL after getting a link in statchg callback. It seems the VGE_DIAGCTL register works like a kind of MII register such that it requires setting a 'to be' mode in advance rather than relying on resolved speed/duplex of established link. This means the statchg callback is not needed in driver. In addition, if there was no link at the time of media change, this was not called at all. - Introduce vge_ifmedia_upd_locked() to change current media to configured one. Actual media change is performed only after PHY reset and VGE_DIAGCTL setup. - In WOL configuration, make sure to clear forced mode such that controller can rely on auto-negotiation. - Unlike most other drivers that use miibus(4), vge(4) used controller's auto-polling feature for link state tracking via interrupt. This came from controller's inefficient mechanism to access MII registers. On link state change interrupt, vge(4) used to get current link state with series of MII register accesses. Because vge(4) already enabled auto polling, read PHY status register to resolved speed/duplex/flow control parameters. vge(4) still does not drive MII_TICK to reduce number of MII register accesses which in turn means the driver does not know the status of auto-negotiation. This was a one of long standing issue of vge(4). Probably driver may be able to implement a timer that keeps track of auto-negotiation state and restart auto-negotiation when driver couldn't establish a link within a specified period. However the controller does not provide a reliable way to detect auto-negotiation failure so I'm not sure whether it's worth to implement it in driver. Alternatively driver can completely disable MII auto-polling and let miibus(4) poll link state by driving MII_TICK. This may reduce unnecessary overhead of stopping/restarting MII auto-polling of controller. Unfortunately it was known that some variants of controller does not work correctly if MII auto-polling is disabled.
2011-11-22 20:45:09 +00:00
LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
PHY_RESET(miisc);
vge_setmedia(sc);
error = mii_mediachg(mii);
return (error);
}
/*
* Report current media status.
*/
static void
2009-12-14 19:44:54 +00:00
vge_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
{
2009-12-14 20:07:25 +00:00
struct vge_softc *sc;
struct mii_data *mii;
sc = ifp->if_softc;
mii = device_get_softc(sc->vge_miibus);
VGE_LOCK(sc);
if ((ifp->if_flags & IFF_UP) == 0) {
VGE_UNLOCK(sc);
return;
}
mii_pollstat(mii);
ifmr->ifm_active = mii->mii_media_active;
ifmr->ifm_status = mii->mii_media_status;
VGE_UNLOCK(sc);
}
static void
Rework link establishment and link state detection logic. - Remove MIIBUS statchg callback and program VGE_DIAGCTL before initiating link establishment. Previously driver used to program VGE_DIAGCTL after getting a link in statchg callback. It seems the VGE_DIAGCTL register works like a kind of MII register such that it requires setting a 'to be' mode in advance rather than relying on resolved speed/duplex of established link. This means the statchg callback is not needed in driver. In addition, if there was no link at the time of media change, this was not called at all. - Introduce vge_ifmedia_upd_locked() to change current media to configured one. Actual media change is performed only after PHY reset and VGE_DIAGCTL setup. - In WOL configuration, make sure to clear forced mode such that controller can rely on auto-negotiation. - Unlike most other drivers that use miibus(4), vge(4) used controller's auto-polling feature for link state tracking via interrupt. This came from controller's inefficient mechanism to access MII registers. On link state change interrupt, vge(4) used to get current link state with series of MII register accesses. Because vge(4) already enabled auto polling, read PHY status register to resolved speed/duplex/flow control parameters. vge(4) still does not drive MII_TICK to reduce number of MII register accesses which in turn means the driver does not know the status of auto-negotiation. This was a one of long standing issue of vge(4). Probably driver may be able to implement a timer that keeps track of auto-negotiation state and restart auto-negotiation when driver couldn't establish a link within a specified period. However the controller does not provide a reliable way to detect auto-negotiation failure so I'm not sure whether it's worth to implement it in driver. Alternatively driver can completely disable MII auto-polling and let miibus(4) poll link state by driving MII_TICK. This may reduce unnecessary overhead of stopping/restarting MII auto-polling of controller. Unfortunately it was known that some variants of controller does not work correctly if MII auto-polling is disabled.
2011-11-22 20:45:09 +00:00
vge_setmedia(struct vge_softc *sc)
{
2009-12-14 20:07:25 +00:00
struct mii_data *mii;
struct ifmedia_entry *ife;
mii = device_get_softc(sc->vge_miibus);
ife = mii->mii_media.ifm_cur;
/*
* If the user manually selects a media mode, we need to turn
* on the forced MAC mode bit in the DIAGCTL register. If the
* user happens to choose a full duplex mode, we also need to
* set the 'force full duplex' bit. This applies only to
* 10Mbps and 100Mbps speeds. In autoselect mode, forced MAC
* mode is disabled, and in 1000baseT mode, full duplex is
* always implied, so we turn on the forced mode bit but leave
* the FDX bit cleared.
*/
switch (IFM_SUBTYPE(ife->ifm_media)) {
case IFM_AUTO:
CSR_CLRBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_MACFORCE);
CSR_CLRBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_FDXFORCE);
break;
case IFM_1000_T:
CSR_SETBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_MACFORCE);
CSR_CLRBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_FDXFORCE);
break;
case IFM_100_TX:
case IFM_10_T:
CSR_SETBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_MACFORCE);
if ((ife->ifm_media & IFM_GMASK) == IFM_FDX) {
CSR_SETBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_FDXFORCE);
} else {
CSR_CLRBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_FDXFORCE);
}
break;
default:
Rework link establishment and link state detection logic. - Remove MIIBUS statchg callback and program VGE_DIAGCTL before initiating link establishment. Previously driver used to program VGE_DIAGCTL after getting a link in statchg callback. It seems the VGE_DIAGCTL register works like a kind of MII register such that it requires setting a 'to be' mode in advance rather than relying on resolved speed/duplex of established link. This means the statchg callback is not needed in driver. In addition, if there was no link at the time of media change, this was not called at all. - Introduce vge_ifmedia_upd_locked() to change current media to configured one. Actual media change is performed only after PHY reset and VGE_DIAGCTL setup. - In WOL configuration, make sure to clear forced mode such that controller can rely on auto-negotiation. - Unlike most other drivers that use miibus(4), vge(4) used controller's auto-polling feature for link state tracking via interrupt. This came from controller's inefficient mechanism to access MII registers. On link state change interrupt, vge(4) used to get current link state with series of MII register accesses. Because vge(4) already enabled auto polling, read PHY status register to resolved speed/duplex/flow control parameters. vge(4) still does not drive MII_TICK to reduce number of MII register accesses which in turn means the driver does not know the status of auto-negotiation. This was a one of long standing issue of vge(4). Probably driver may be able to implement a timer that keeps track of auto-negotiation state and restart auto-negotiation when driver couldn't establish a link within a specified period. However the controller does not provide a reliable way to detect auto-negotiation failure so I'm not sure whether it's worth to implement it in driver. Alternatively driver can completely disable MII auto-polling and let miibus(4) poll link state by driving MII_TICK. This may reduce unnecessary overhead of stopping/restarting MII auto-polling of controller. Unfortunately it was known that some variants of controller does not work correctly if MII auto-polling is disabled.
2011-11-22 20:45:09 +00:00
device_printf(sc->vge_dev, "unknown media type: %x\n",
IFM_SUBTYPE(ife->ifm_media));
break;
}
}
static int
2009-12-14 19:44:54 +00:00
vge_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
{
2009-12-14 20:07:25 +00:00
struct vge_softc *sc = ifp->if_softc;
struct ifreq *ifr = (struct ifreq *) data;
struct mii_data *mii;
int error = 0, mask;
switch (command) {
case SIOCSIFMTU:
VGE_LOCK(sc);
if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > VGE_JUMBO_MTU)
error = EINVAL;
else if (ifp->if_mtu != ifr->ifr_mtu) {
if (ifr->ifr_mtu > ETHERMTU &&
(sc->vge_flags & VGE_FLAG_JUMBO) == 0)
error = EINVAL;
else
ifp->if_mtu = ifr->ifr_mtu;
}
VGE_UNLOCK(sc);
break;
case SIOCSIFFLAGS:
VGE_LOCK(sc);
if ((ifp->if_flags & IFF_UP) != 0) {
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0 &&
((ifp->if_flags ^ sc->vge_if_flags) &
(IFF_PROMISC | IFF_ALLMULTI)) != 0)
vge_rxfilter(sc);
else
vge_init_locked(sc);
} else if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
vge_stop(sc);
sc->vge_if_flags = ifp->if_flags;
VGE_UNLOCK(sc);
break;
case SIOCADDMULTI:
case SIOCDELMULTI:
VGE_LOCK(sc);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
if (ifp->if_drv_flags & IFF_DRV_RUNNING)
vge_rxfilter(sc);
VGE_UNLOCK(sc);
break;
case SIOCGIFMEDIA:
case SIOCSIFMEDIA:
mii = device_get_softc(sc->vge_miibus);
error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command);
break;
case SIOCSIFCAP:
mask = ifr->ifr_reqcap ^ ifp->if_capenable;
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
#ifdef DEVICE_POLLING
if (mask & IFCAP_POLLING) {
if (ifr->ifr_reqcap & IFCAP_POLLING) {
error = ether_poll_register(vge_poll, ifp);
if (error)
2009-12-14 20:07:25 +00:00
return (error);
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
VGE_LOCK(sc);
/* Disable interrupts */
CSR_WRITE_4(sc, VGE_IMR, VGE_INTRS_POLLING);
CSR_WRITE_4(sc, VGE_ISR, 0xFFFFFFFF);
CSR_WRITE_1(sc, VGE_CRS3, VGE_CR3_INT_GMSK);
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
ifp->if_capenable |= IFCAP_POLLING;
VGE_UNLOCK(sc);
} else {
error = ether_poll_deregister(ifp);
/* Enable interrupts. */
VGE_LOCK(sc);
CSR_WRITE_4(sc, VGE_IMR, VGE_INTRS);
CSR_WRITE_4(sc, VGE_ISR, 0xFFFFFFFF);
CSR_WRITE_1(sc, VGE_CRS3, VGE_CR3_INT_GMSK);
ifp->if_capenable &= ~IFCAP_POLLING;
VGE_UNLOCK(sc);
}
}
#endif /* DEVICE_POLLING */
VGE_LOCK(sc);
if ((mask & IFCAP_TXCSUM) != 0 &&
(ifp->if_capabilities & IFCAP_TXCSUM) != 0) {
ifp->if_capenable ^= IFCAP_TXCSUM;
if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
ifp->if_hwassist |= VGE_CSUM_FEATURES;
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
else
ifp->if_hwassist &= ~VGE_CSUM_FEATURES;
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
}
if ((mask & IFCAP_RXCSUM) != 0 &&
(ifp->if_capabilities & IFCAP_RXCSUM) != 0)
ifp->if_capenable ^= IFCAP_RXCSUM;
if ((mask & IFCAP_WOL_UCAST) != 0 &&
(ifp->if_capabilities & IFCAP_WOL_UCAST) != 0)
ifp->if_capenable ^= IFCAP_WOL_UCAST;
if ((mask & IFCAP_WOL_MCAST) != 0 &&
(ifp->if_capabilities & IFCAP_WOL_MCAST) != 0)
ifp->if_capenable ^= IFCAP_WOL_MCAST;
if ((mask & IFCAP_WOL_MAGIC) != 0 &&
(ifp->if_capabilities & IFCAP_WOL_MAGIC) != 0)
ifp->if_capenable ^= IFCAP_WOL_MAGIC;
if ((mask & IFCAP_VLAN_HWCSUM) != 0 &&
(ifp->if_capabilities & IFCAP_VLAN_HWCSUM) != 0)
ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
(IFCAP_VLAN_HWTAGGING & ifp->if_capabilities) != 0) {
ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
vge_setvlan(sc);
}
VGE_UNLOCK(sc);
VLAN_CAPABILITIES(ifp);
break;
default:
error = ether_ioctl(ifp, command, data);
break;
}
return (error);
}
static void
vge_watchdog(void *arg)
{
struct vge_softc *sc;
struct ifnet *ifp;
sc = arg;
VGE_LOCK_ASSERT(sc);
vge_stats_update(sc);
callout_reset(&sc->vge_watchdog, hz, vge_watchdog, sc);
if (sc->vge_timer == 0 || --sc->vge_timer > 0)
return;
ifp = sc->vge_ifp;
if_printf(ifp, "watchdog timeout\n");
ifp->if_oerrors++;
vge_txeof(sc);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
vge_rxeof(sc, VGE_RX_DESC_CNT);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
vge_init_locked(sc);
}
/*
* Stop the adapter and free any mbufs allocated to the
* RX and TX lists.
*/
static void
2009-12-14 19:44:54 +00:00
vge_stop(struct vge_softc *sc)
{
2009-12-14 20:07:25 +00:00
struct ifnet *ifp;
VGE_LOCK_ASSERT(sc);
ifp = sc->vge_ifp;
sc->vge_timer = 0;
callout_stop(&sc->vge_watchdog);
ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
CSR_WRITE_1(sc, VGE_CRC3, VGE_CR3_INT_GMSK);
CSR_WRITE_1(sc, VGE_CRS0, VGE_CR0_STOP);
CSR_WRITE_4(sc, VGE_ISR, 0xFFFFFFFF);
CSR_WRITE_2(sc, VGE_TXQCSRC, 0xFFFF);
CSR_WRITE_1(sc, VGE_RXQCSRC, 0xFF);
CSR_WRITE_4(sc, VGE_RXDESC_ADDR_LO, 0);
vge_stats_update(sc);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
VGE_CHAIN_RESET(sc);
vge_txeof(sc);
vge_freebufs(sc);
}
/*
* Device suspend routine. Stop the interface and save some PCI
* settings in case the BIOS doesn't restore them properly on
* resume.
*/
static int
2009-12-14 19:44:54 +00:00
vge_suspend(device_t dev)
{
2009-12-14 20:07:25 +00:00
struct vge_softc *sc;
sc = device_get_softc(dev);
VGE_LOCK(sc);
vge_stop(sc);
vge_setwol(sc);
sc->vge_flags |= VGE_FLAG_SUSPENDED;
VGE_UNLOCK(sc);
return (0);
}
/*
* Device resume routine. Restore some PCI settings in case the BIOS
* doesn't, re-enable busmastering, and restart the interface if
* appropriate.
*/
static int
2009-12-14 19:44:54 +00:00
vge_resume(device_t dev)
{
2009-12-14 20:07:25 +00:00
struct vge_softc *sc;
struct ifnet *ifp;
uint16_t pmstat;
sc = device_get_softc(dev);
VGE_LOCK(sc);
if ((sc->vge_flags & VGE_FLAG_PMCAP) != 0) {
/* Disable PME and clear PME status. */
pmstat = pci_read_config(sc->vge_dev,
sc->vge_pmcap + PCIR_POWER_STATUS, 2);
if ((pmstat & PCIM_PSTAT_PMEENABLE) != 0) {
pmstat &= ~PCIM_PSTAT_PMEENABLE;
pci_write_config(sc->vge_dev,
sc->vge_pmcap + PCIR_POWER_STATUS, pmstat, 2);
}
}
vge_clrwol(sc);
/* Restart MII auto-polling. */
vge_miipoll_start(sc);
ifp = sc->vge_ifp;
/* Reinitialize interface if necessary. */
if ((ifp->if_flags & IFF_UP) != 0) {
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
vge_init_locked(sc);
Overhaul bus_dma(9) usage and fix various things. o Separate TX/RX buffer DMA tag from TX/RX descriptor ring DMA tag. o Separate RX buffer DMA tag from common buffer DMA tag. RX DMA tag has different restriction compared to TX DMA tag. o Add 40bit DMA address support. o Adjust TX/RX descriptor ring alignment to 64 bytes from 256 bytes as documented in datasheet. o Added check to ensure TX/RX ring reside within a 4GB boundary. Since TX/RX ring shares the same high address register they should have the same high address. o TX/RX side bus_dmamap_load_mbuf_sg(9) support. o Add lock assertion to vge_setmulti(). o Add RX spare DMA map to recover from DMA map load failure. o Add optimized RX buffer handler, vge_discard_rxbuf which is activated when vge(4) sees bad frames. o Don't blindly update VGE_RXDESC_RESIDUECNT register. Datasheet says the register should be updated only when number of available RX descriptors are multiple of 4. o Use __NO_STRICT_ALIGNMENT instead of defining VGE_FIXUP_RX which is only set for i386 architecture. Previously vge(4) also performed expensive copy operation to align IP header on amd64. This change should give RX performance boost on amd64 architecture. o Don't reinitialize controller if driver is already running. This should reduce number of link state flipping. o Since vge(4) drops a driver lock before passing received frame to upper layer, make sure vge(4) is still running after re-acquiring driver lock. o Add second argument count to vge_rxeof(). The argument will limit number of packets could be processed in RX handler. o Rearrange vge_rxeof() not to allocate RX buffer if received frame was bad packet. o Removed if_printf that prints DMA map failure. This type of message shouldn't be used in fast path of driver. o Reduce number of allowed TX buffer fragments to 6 from 7. A TX descriptor allows 7 fragments of a frame. However the CMZ field of descriptor has just 3bits and the controller wants to see fragment + 1 in the field. So if we have 7 fragments the field value would be 0 which seems to cause unexpected results under certain conditions. This change should fix occasional TX hang observed on vge(4). o Simplify vge_stat_locked() and add number of available TX descriptor check. o vge(4) controllers lack padding short frames. Make sure to fill zero for the padded bytes. This closes unintended information disclosure. o Don't set VGE_TDCTL_JUMBO flag. Datasheet is not clear whether this bit should be set by driver or write-back status bit after transmission. At least vendor's driver does not set this bit so remove it. Without this bit vge(4) still can send jumbo frames. o Don't start driver when vge(4) know there are not enough RX buffers. o Remove volatile keyword in RX descriptor structure. This should be handled by bus_dma(9). o Collapse two 16bits member of TX/RX descriptor into single 32bits member. o Reduce number of RX descriptors to 252 from 256. The VGE_RXDESCNUM is 16bits register but only lower 8bits are valid. So the maximum number of RX descriptors would be 255. However the number of should be multiple of 4 as controller wants to update 4 RX descriptors at a time. This limits the maximum number of RX descriptor to be 252. Tested by: Dewayne Geraghty (dewayne.geraghty <> heuristicsystems dot com dot au) Carey Jones (m.carey.jones <> gmail dot com) Yoshiaki Kasahara (kasahara <> nc dor kyushu-u dot ac dotjp)
2009-12-14 18:44:23 +00:00
}
sc->vge_flags &= ~VGE_FLAG_SUSPENDED;
VGE_UNLOCK(sc);
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
2009-12-14 19:44:54 +00:00
vge_shutdown(device_t dev)
{
return (vge_suspend(dev));
}
#define VGE_SYSCTL_STAT_ADD32(c, h, n, p, d) \
SYSCTL_ADD_UINT(c, h, OID_AUTO, n, CTLFLAG_RD, p, 0, d)
static void
vge_sysctl_node(struct vge_softc *sc)
{
struct sysctl_ctx_list *ctx;
struct sysctl_oid_list *child, *parent;
struct sysctl_oid *tree;
struct vge_hw_stats *stats;
stats = &sc->vge_stats;
ctx = device_get_sysctl_ctx(sc->vge_dev);
child = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->vge_dev));
Implement interrupt moderation scheme supported by VT61xx controllers. TX/RX interrupt mitigation is controlled by VGE_TXSUPPTHR and VGE_RXSUPPTHR register. These registers suppress generation of interrupts until the programmed frames counter equals to the registers. VT61xx also supports interrupt hold off timer register. If this interrupt hold off timer is active all interrupts would be disabled until the timer reaches to 0. The timer value is reloaded whenever VGE_ISR register written. The timer resolution is about 20us. Previously vge(4) used single shot timer to reduce Tx completion interrupts. This required VGE_CRS1 register access in Tx start/completion handler to rearm new timeout value and it did not show satisfactory result(more than 50k interrupts under load). Rx interrupts was not moderated at all such that vge(4) used to generate too many interrupts which in turn made polling(4) better approach under high network load. This change activates all interrupt moderation mechanism and initial values were tuned to generate interrupt less than 8k per second. That number of interrupts wouldn't add additional packet latencies compared to polling(4). These interrupt parameters could be changed with sysctl. dev.vge.%d.int_holdoff dev.vge.%d.rx_coal_pkt dev.vge.%d.tx_coal_pkt Interface has be brought down and up again before change take effect. With interrupt moderation there is no more need to loop in interrupt handler. This loop always added one more register access. While I'm here remove dead code which tried to implement subset of interrupt moderation.
2009-12-17 18:00:25 +00:00
SYSCTL_ADD_INT(ctx, child, OID_AUTO, "int_holdoff",
CTLFLAG_RW, &sc->vge_int_holdoff, 0, "interrupt holdoff");
SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rx_coal_pkt",
CTLFLAG_RW, &sc->vge_rx_coal_pkt, 0, "rx coalescing packet");
SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_coal_pkt",
CTLFLAG_RW, &sc->vge_tx_coal_pkt, 0, "tx coalescing packet");
/* Pull in device tunables. */
sc->vge_int_holdoff = VGE_INT_HOLDOFF_DEFAULT;
resource_int_value(device_get_name(sc->vge_dev),
device_get_unit(sc->vge_dev), "int_holdoff", &sc->vge_int_holdoff);
sc->vge_rx_coal_pkt = VGE_RX_COAL_PKT_DEFAULT;
resource_int_value(device_get_name(sc->vge_dev),
device_get_unit(sc->vge_dev), "rx_coal_pkt", &sc->vge_rx_coal_pkt);
sc->vge_tx_coal_pkt = VGE_TX_COAL_PKT_DEFAULT;
resource_int_value(device_get_name(sc->vge_dev),
device_get_unit(sc->vge_dev), "tx_coal_pkt", &sc->vge_tx_coal_pkt);
tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "stats", CTLFLAG_RD,
NULL, "VGE statistics");
parent = SYSCTL_CHILDREN(tree);
/* Rx statistics. */
tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "rx", CTLFLAG_RD,
NULL, "RX MAC statistics");
child = SYSCTL_CHILDREN(tree);
VGE_SYSCTL_STAT_ADD32(ctx, child, "frames",
&stats->rx_frames, "frames");
VGE_SYSCTL_STAT_ADD32(ctx, child, "good_frames",
&stats->rx_good_frames, "Good frames");
VGE_SYSCTL_STAT_ADD32(ctx, child, "fifo_oflows",
&stats->rx_fifo_oflows, "FIFO overflows");
VGE_SYSCTL_STAT_ADD32(ctx, child, "runts",
&stats->rx_runts, "Too short frames");
VGE_SYSCTL_STAT_ADD32(ctx, child, "runts_errs",
&stats->rx_runts_errs, "Too short frames with errors");
VGE_SYSCTL_STAT_ADD32(ctx, child, "frames_64",
&stats->rx_pkts_64, "64 bytes frames");
VGE_SYSCTL_STAT_ADD32(ctx, child, "frames_65_127",
&stats->rx_pkts_65_127, "65 to 127 bytes frames");
VGE_SYSCTL_STAT_ADD32(ctx, child, "frames_128_255",
&stats->rx_pkts_128_255, "128 to 255 bytes frames");
VGE_SYSCTL_STAT_ADD32(ctx, child, "frames_256_511",
&stats->rx_pkts_256_511, "256 to 511 bytes frames");
VGE_SYSCTL_STAT_ADD32(ctx, child, "frames_512_1023",
&stats->rx_pkts_512_1023, "512 to 1023 bytes frames");
VGE_SYSCTL_STAT_ADD32(ctx, child, "frames_1024_1518",
&stats->rx_pkts_1024_1518, "1024 to 1518 bytes frames");
VGE_SYSCTL_STAT_ADD32(ctx, child, "frames_1519_max",
&stats->rx_pkts_1519_max, "1519 to max frames");
VGE_SYSCTL_STAT_ADD32(ctx, child, "frames_1519_max_errs",
&stats->rx_pkts_1519_max_errs, "1519 to max frames with error");
VGE_SYSCTL_STAT_ADD32(ctx, child, "frames_jumbo",
&stats->rx_jumbos, "Jumbo frames");
VGE_SYSCTL_STAT_ADD32(ctx, child, "crcerrs",
&stats->rx_crcerrs, "CRC errors");
VGE_SYSCTL_STAT_ADD32(ctx, child, "pause_frames",
&stats->rx_pause_frames, "CRC errors");
VGE_SYSCTL_STAT_ADD32(ctx, child, "align_errs",
&stats->rx_alignerrs, "Alignment errors");
VGE_SYSCTL_STAT_ADD32(ctx, child, "nobufs",
&stats->rx_nobufs, "Frames with no buffer event");
VGE_SYSCTL_STAT_ADD32(ctx, child, "sym_errs",
&stats->rx_symerrs, "Frames with symbol errors");
VGE_SYSCTL_STAT_ADD32(ctx, child, "len_errs",
&stats->rx_lenerrs, "Frames with length mismatched");
/* Tx statistics. */
tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "tx", CTLFLAG_RD,
NULL, "TX MAC statistics");
child = SYSCTL_CHILDREN(tree);
VGE_SYSCTL_STAT_ADD32(ctx, child, "good_frames",
&stats->tx_good_frames, "Good frames");
VGE_SYSCTL_STAT_ADD32(ctx, child, "frames_64",
&stats->tx_pkts_64, "64 bytes frames");
VGE_SYSCTL_STAT_ADD32(ctx, child, "frames_65_127",
&stats->tx_pkts_65_127, "65 to 127 bytes frames");
VGE_SYSCTL_STAT_ADD32(ctx, child, "frames_128_255",
&stats->tx_pkts_128_255, "128 to 255 bytes frames");
VGE_SYSCTL_STAT_ADD32(ctx, child, "frames_256_511",
&stats->tx_pkts_256_511, "256 to 511 bytes frames");
VGE_SYSCTL_STAT_ADD32(ctx, child, "frames_512_1023",
&stats->tx_pkts_512_1023, "512 to 1023 bytes frames");
VGE_SYSCTL_STAT_ADD32(ctx, child, "frames_1024_1518",
&stats->tx_pkts_1024_1518, "1024 to 1518 bytes frames");
VGE_SYSCTL_STAT_ADD32(ctx, child, "frames_jumbo",
&stats->tx_jumbos, "Jumbo frames");
VGE_SYSCTL_STAT_ADD32(ctx, child, "colls",
&stats->tx_colls, "Collisions");
VGE_SYSCTL_STAT_ADD32(ctx, child, "late_colls",
&stats->tx_latecolls, "Late collisions");
VGE_SYSCTL_STAT_ADD32(ctx, child, "pause_frames",
&stats->tx_pause, "Pause frames");
#ifdef VGE_ENABLE_SQEERR
VGE_SYSCTL_STAT_ADD32(ctx, child, "sqeerrs",
&stats->tx_sqeerrs, "SQE errors");
#endif
/* Clear MAC statistics. */
vge_stats_clear(sc);
}
#undef VGE_SYSCTL_STAT_ADD32
static void
vge_stats_clear(struct vge_softc *sc)
{
int i;
CSR_WRITE_1(sc, VGE_MIBCSR,
CSR_READ_1(sc, VGE_MIBCSR) | VGE_MIBCSR_FREEZE);
CSR_WRITE_1(sc, VGE_MIBCSR,
CSR_READ_1(sc, VGE_MIBCSR) | VGE_MIBCSR_CLR);
for (i = VGE_TIMEOUT; i > 0; i--) {
DELAY(1);
if ((CSR_READ_1(sc, VGE_MIBCSR) & VGE_MIBCSR_CLR) == 0)
break;
}
if (i == 0)
device_printf(sc->vge_dev, "MIB clear timed out!\n");
CSR_WRITE_1(sc, VGE_MIBCSR, CSR_READ_1(sc, VGE_MIBCSR) &
~VGE_MIBCSR_FREEZE);
}
static void
vge_stats_update(struct vge_softc *sc)
{
struct vge_hw_stats *stats;
struct ifnet *ifp;
uint32_t mib[VGE_MIB_CNT], val;
int i;
VGE_LOCK_ASSERT(sc);
stats = &sc->vge_stats;
ifp = sc->vge_ifp;
CSR_WRITE_1(sc, VGE_MIBCSR,
CSR_READ_1(sc, VGE_MIBCSR) | VGE_MIBCSR_FLUSH);
for (i = VGE_TIMEOUT; i > 0; i--) {
DELAY(1);
if ((CSR_READ_1(sc, VGE_MIBCSR) & VGE_MIBCSR_FLUSH) == 0)
break;
}
if (i == 0) {
device_printf(sc->vge_dev, "MIB counter dump timed out!\n");
vge_stats_clear(sc);
return;
}
bzero(mib, sizeof(mib));
reset_idx:
/* Set MIB read index to 0. */
CSR_WRITE_1(sc, VGE_MIBCSR,
CSR_READ_1(sc, VGE_MIBCSR) | VGE_MIBCSR_RINI);
for (i = 0; i < VGE_MIB_CNT; i++) {
val = CSR_READ_4(sc, VGE_MIBDATA);
if (i != VGE_MIB_DATA_IDX(val)) {
/* Reading interrupted. */
goto reset_idx;
}
mib[i] = val & VGE_MIB_DATA_MASK;
}
/* Rx stats. */
stats->rx_frames += mib[VGE_MIB_RX_FRAMES];
stats->rx_good_frames += mib[VGE_MIB_RX_GOOD_FRAMES];
stats->rx_fifo_oflows += mib[VGE_MIB_RX_FIFO_OVERRUNS];
stats->rx_runts += mib[VGE_MIB_RX_RUNTS];
stats->rx_runts_errs += mib[VGE_MIB_RX_RUNTS_ERRS];
stats->rx_pkts_64 += mib[VGE_MIB_RX_PKTS_64];
stats->rx_pkts_65_127 += mib[VGE_MIB_RX_PKTS_65_127];
stats->rx_pkts_128_255 += mib[VGE_MIB_RX_PKTS_128_255];
stats->rx_pkts_256_511 += mib[VGE_MIB_RX_PKTS_256_511];
stats->rx_pkts_512_1023 += mib[VGE_MIB_RX_PKTS_512_1023];
stats->rx_pkts_1024_1518 += mib[VGE_MIB_RX_PKTS_1024_1518];
stats->rx_pkts_1519_max += mib[VGE_MIB_RX_PKTS_1519_MAX];
stats->rx_pkts_1519_max_errs += mib[VGE_MIB_RX_PKTS_1519_MAX_ERRS];
stats->rx_jumbos += mib[VGE_MIB_RX_JUMBOS];
stats->rx_crcerrs += mib[VGE_MIB_RX_CRCERRS];
stats->rx_pause_frames += mib[VGE_MIB_RX_PAUSE];
stats->rx_alignerrs += mib[VGE_MIB_RX_ALIGNERRS];
stats->rx_nobufs += mib[VGE_MIB_RX_NOBUFS];
stats->rx_symerrs += mib[VGE_MIB_RX_SYMERRS];
stats->rx_lenerrs += mib[VGE_MIB_RX_LENERRS];
/* Tx stats. */
stats->tx_good_frames += mib[VGE_MIB_TX_GOOD_FRAMES];
stats->tx_pkts_64 += mib[VGE_MIB_TX_PKTS_64];
stats->tx_pkts_65_127 += mib[VGE_MIB_TX_PKTS_65_127];
stats->tx_pkts_128_255 += mib[VGE_MIB_TX_PKTS_128_255];
stats->tx_pkts_256_511 += mib[VGE_MIB_TX_PKTS_256_511];
stats->tx_pkts_512_1023 += mib[VGE_MIB_TX_PKTS_512_1023];
stats->tx_pkts_1024_1518 += mib[VGE_MIB_TX_PKTS_1024_1518];
stats->tx_jumbos += mib[VGE_MIB_TX_JUMBOS];
stats->tx_colls += mib[VGE_MIB_TX_COLLS];
stats->tx_pause += mib[VGE_MIB_TX_PAUSE];
#ifdef VGE_ENABLE_SQEERR
stats->tx_sqeerrs += mib[VGE_MIB_TX_SQEERRS];
#endif
stats->tx_latecolls += mib[VGE_MIB_TX_LATECOLLS];
/* Update counters in ifnet. */
ifp->if_opackets += mib[VGE_MIB_TX_GOOD_FRAMES];
ifp->if_collisions += mib[VGE_MIB_TX_COLLS] +
mib[VGE_MIB_TX_LATECOLLS];
ifp->if_oerrors += mib[VGE_MIB_TX_COLLS] +
mib[VGE_MIB_TX_LATECOLLS];
ifp->if_ipackets += mib[VGE_MIB_RX_GOOD_FRAMES];
ifp->if_ierrors += mib[VGE_MIB_RX_FIFO_OVERRUNS] +
mib[VGE_MIB_RX_RUNTS] +
mib[VGE_MIB_RX_RUNTS_ERRS] +
mib[VGE_MIB_RX_CRCERRS] +
mib[VGE_MIB_RX_ALIGNERRS] +
mib[VGE_MIB_RX_NOBUFS] +
mib[VGE_MIB_RX_SYMERRS] +
mib[VGE_MIB_RX_LENERRS];
}
Implement interrupt moderation scheme supported by VT61xx controllers. TX/RX interrupt mitigation is controlled by VGE_TXSUPPTHR and VGE_RXSUPPTHR register. These registers suppress generation of interrupts until the programmed frames counter equals to the registers. VT61xx also supports interrupt hold off timer register. If this interrupt hold off timer is active all interrupts would be disabled until the timer reaches to 0. The timer value is reloaded whenever VGE_ISR register written. The timer resolution is about 20us. Previously vge(4) used single shot timer to reduce Tx completion interrupts. This required VGE_CRS1 register access in Tx start/completion handler to rearm new timeout value and it did not show satisfactory result(more than 50k interrupts under load). Rx interrupts was not moderated at all such that vge(4) used to generate too many interrupts which in turn made polling(4) better approach under high network load. This change activates all interrupt moderation mechanism and initial values were tuned to generate interrupt less than 8k per second. That number of interrupts wouldn't add additional packet latencies compared to polling(4). These interrupt parameters could be changed with sysctl. dev.vge.%d.int_holdoff dev.vge.%d.rx_coal_pkt dev.vge.%d.tx_coal_pkt Interface has be brought down and up again before change take effect. With interrupt moderation there is no more need to loop in interrupt handler. This loop always added one more register access. While I'm here remove dead code which tried to implement subset of interrupt moderation.
2009-12-17 18:00:25 +00:00
static void
vge_intr_holdoff(struct vge_softc *sc)
{
uint8_t intctl;
VGE_LOCK_ASSERT(sc);
/*
* Set Tx interrupt supression threshold.
* It's possible to use single-shot timer in VGE_CRS1 register
* in Tx path such that driver can remove most of Tx completion
* interrupts. However this requires additional access to
* VGE_CRS1 register to reload the timer in addintion to
* activating Tx kick command. Another downside is we don't know
* what single-shot timer value should be used in advance so
* reclaiming transmitted mbufs could be delayed a lot which in
* turn slows down Tx operation.
*/
CSR_WRITE_1(sc, VGE_CAMCTL, VGE_PAGESEL_TXSUPPTHR);
CSR_WRITE_1(sc, VGE_TXSUPPTHR, sc->vge_tx_coal_pkt);
/* Set Rx interrupt suppresion threshold. */
CSR_WRITE_1(sc, VGE_CAMCTL, VGE_PAGESEL_RXSUPPTHR);
CSR_WRITE_1(sc, VGE_RXSUPPTHR, sc->vge_rx_coal_pkt);
intctl = CSR_READ_1(sc, VGE_INTCTL1);
intctl &= ~VGE_INTCTL_SC_RELOAD;
intctl |= VGE_INTCTL_HC_RELOAD;
if (sc->vge_tx_coal_pkt <= 0)
intctl |= VGE_INTCTL_TXINTSUP_DISABLE;
else
intctl &= ~VGE_INTCTL_TXINTSUP_DISABLE;
if (sc->vge_rx_coal_pkt <= 0)
intctl |= VGE_INTCTL_RXINTSUP_DISABLE;
else
intctl &= ~VGE_INTCTL_RXINTSUP_DISABLE;
CSR_WRITE_1(sc, VGE_INTCTL1, intctl);
CSR_WRITE_1(sc, VGE_CRC3, VGE_CR3_INT_HOLDOFF);
if (sc->vge_int_holdoff > 0) {
/* Set interrupt holdoff timer. */
CSR_WRITE_1(sc, VGE_CAMCTL, VGE_PAGESEL_INTHLDOFF);
CSR_WRITE_1(sc, VGE_INTHOLDOFF,
VGE_INT_HOLDOFF_USEC(sc->vge_int_holdoff));
/* Enable holdoff timer. */
CSR_WRITE_1(sc, VGE_CRS3, VGE_CR3_INT_HOLDOFF);
}
}
static void
vge_setlinkspeed(struct vge_softc *sc)
{
struct mii_data *mii;
int aneg, i;
VGE_LOCK_ASSERT(sc);
mii = device_get_softc(sc->vge_miibus);
mii_pollstat(mii);
aneg = 0;
if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
(IFM_ACTIVE | IFM_AVALID)) {
switch IFM_SUBTYPE(mii->mii_media_active) {
case IFM_10_T:
case IFM_100_TX:
return;
case IFM_1000_T:
aneg++;
default:
break;
}
}
Rework link establishment and link state detection logic. - Remove MIIBUS statchg callback and program VGE_DIAGCTL before initiating link establishment. Previously driver used to program VGE_DIAGCTL after getting a link in statchg callback. It seems the VGE_DIAGCTL register works like a kind of MII register such that it requires setting a 'to be' mode in advance rather than relying on resolved speed/duplex of established link. This means the statchg callback is not needed in driver. In addition, if there was no link at the time of media change, this was not called at all. - Introduce vge_ifmedia_upd_locked() to change current media to configured one. Actual media change is performed only after PHY reset and VGE_DIAGCTL setup. - In WOL configuration, make sure to clear forced mode such that controller can rely on auto-negotiation. - Unlike most other drivers that use miibus(4), vge(4) used controller's auto-polling feature for link state tracking via interrupt. This came from controller's inefficient mechanism to access MII registers. On link state change interrupt, vge(4) used to get current link state with series of MII register accesses. Because vge(4) already enabled auto polling, read PHY status register to resolved speed/duplex/flow control parameters. vge(4) still does not drive MII_TICK to reduce number of MII register accesses which in turn means the driver does not know the status of auto-negotiation. This was a one of long standing issue of vge(4). Probably driver may be able to implement a timer that keeps track of auto-negotiation state and restart auto-negotiation when driver couldn't establish a link within a specified period. However the controller does not provide a reliable way to detect auto-negotiation failure so I'm not sure whether it's worth to implement it in driver. Alternatively driver can completely disable MII auto-polling and let miibus(4) poll link state by driving MII_TICK. This may reduce unnecessary overhead of stopping/restarting MII auto-polling of controller. Unfortunately it was known that some variants of controller does not work correctly if MII auto-polling is disabled.
2011-11-22 20:45:09 +00:00
/* Clear forced MAC speed/duplex configuration. */
CSR_CLRBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_MACFORCE);
CSR_CLRBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_FDXFORCE);
vge_miibus_writereg(sc->vge_dev, sc->vge_phyaddr, MII_100T2CR, 0);
vge_miibus_writereg(sc->vge_dev, sc->vge_phyaddr, MII_ANAR,
ANAR_TX_FD | ANAR_TX | ANAR_10_FD | ANAR_10 | ANAR_CSMA);
vge_miibus_writereg(sc->vge_dev, sc->vge_phyaddr, MII_BMCR,
BMCR_AUTOEN | BMCR_STARTNEG);
DELAY(1000);
if (aneg != 0) {
/* Poll link state until vge(4) get a 10/100 link. */
for (i = 0; i < MII_ANEGTICKS_GIGE; i++) {
mii_pollstat(mii);
if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID))
== (IFM_ACTIVE | IFM_AVALID)) {
switch (IFM_SUBTYPE(mii->mii_media_active)) {
case IFM_10_T:
case IFM_100_TX:
return;
default:
break;
}
}
VGE_UNLOCK(sc);
pause("vgelnk", hz);
VGE_LOCK(sc);
}
if (i == MII_ANEGTICKS_GIGE)
device_printf(sc->vge_dev, "establishing link failed, "
"WOL may not work!");
}
/*
* No link, force MAC to have 100Mbps, full-duplex link.
* This is the last resort and may/may not work.
*/
mii->mii_media_status = IFM_AVALID | IFM_ACTIVE;
mii->mii_media_active = IFM_ETHER | IFM_100_TX | IFM_FDX;
}
static void
vge_setwol(struct vge_softc *sc)
{
struct ifnet *ifp;
uint16_t pmstat;
uint8_t val;
VGE_LOCK_ASSERT(sc);
if ((sc->vge_flags & VGE_FLAG_PMCAP) == 0) {
/* No PME capability, PHY power down. */
vge_miibus_writereg(sc->vge_dev, sc->vge_phyaddr, MII_BMCR,
BMCR_PDOWN);
vge_miipoll_stop(sc);
return;
}
ifp = sc->vge_ifp;
/* Clear WOL on pattern match. */
CSR_WRITE_1(sc, VGE_WOLCR0C, VGE_WOLCR0_PATTERN_ALL);
/* Disable WOL on magic/unicast packet. */
CSR_WRITE_1(sc, VGE_WOLCR1C, 0x0F);
CSR_WRITE_1(sc, VGE_WOLCFGC, VGE_WOLCFG_SAB | VGE_WOLCFG_SAM |
VGE_WOLCFG_PMEOVR);
if ((ifp->if_capenable & IFCAP_WOL) != 0) {
vge_setlinkspeed(sc);
val = 0;
if ((ifp->if_capenable & IFCAP_WOL_UCAST) != 0)
val |= VGE_WOLCR1_UCAST;
if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
val |= VGE_WOLCR1_MAGIC;
CSR_WRITE_1(sc, VGE_WOLCR1S, val);
val = 0;
if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
val |= VGE_WOLCFG_SAM | VGE_WOLCFG_SAB;
CSR_WRITE_1(sc, VGE_WOLCFGS, val | VGE_WOLCFG_PMEOVR);
/* Disable MII auto-polling. */
vge_miipoll_stop(sc);
}
CSR_SETBIT_1(sc, VGE_DIAGCTL,
VGE_DIAGCTL_MACFORCE | VGE_DIAGCTL_FDXFORCE);
CSR_CLRBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_GMII);
/* Clear WOL status on pattern match. */
CSR_WRITE_1(sc, VGE_WOLSR0C, 0xFF);
CSR_WRITE_1(sc, VGE_WOLSR1C, 0xFF);
val = CSR_READ_1(sc, VGE_PWRSTAT);
val |= VGE_STICKHW_SWPTAG;
CSR_WRITE_1(sc, VGE_PWRSTAT, val);
/* Put hardware into sleep. */
val = CSR_READ_1(sc, VGE_PWRSTAT);
val |= VGE_STICKHW_DS0 | VGE_STICKHW_DS1;
CSR_WRITE_1(sc, VGE_PWRSTAT, val);
/* Request PME if WOL is requested. */
pmstat = pci_read_config(sc->vge_dev, sc->vge_pmcap +
PCIR_POWER_STATUS, 2);
pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
if ((ifp->if_capenable & IFCAP_WOL) != 0)
pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
pci_write_config(sc->vge_dev, sc->vge_pmcap + PCIR_POWER_STATUS,
pmstat, 2);
}
static void
vge_clrwol(struct vge_softc *sc)
{
uint8_t val;
val = CSR_READ_1(sc, VGE_PWRSTAT);
val &= ~VGE_STICKHW_SWPTAG;
CSR_WRITE_1(sc, VGE_PWRSTAT, val);
/* Disable WOL and clear power state indicator. */
val = CSR_READ_1(sc, VGE_PWRSTAT);
val &= ~(VGE_STICKHW_DS0 | VGE_STICKHW_DS1);
CSR_WRITE_1(sc, VGE_PWRSTAT, val);
CSR_CLRBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_GMII);
CSR_CLRBIT_1(sc, VGE_DIAGCTL, VGE_DIAGCTL_MACFORCE);
/* Clear WOL on pattern match. */
CSR_WRITE_1(sc, VGE_WOLCR0C, VGE_WOLCR0_PATTERN_ALL);
/* Disable WOL on magic/unicast packet. */
CSR_WRITE_1(sc, VGE_WOLCR1C, 0x0F);
CSR_WRITE_1(sc, VGE_WOLCFGC, VGE_WOLCFG_SAB | VGE_WOLCFG_SAM |
VGE_WOLCFG_PMEOVR);
/* Clear WOL status on pattern match. */
CSR_WRITE_1(sc, VGE_WOLSR0C, 0xFF);
CSR_WRITE_1(sc, VGE_WOLSR1C, 0xFF);
}