freebsd-dev/sys/pci/if_wx.c

1986 lines
45 KiB
C
Raw Normal View History

Add first pass of the Intel Gigabit Ethernet (wiseman) driver. This driver seems relatively functional, but could use some souping up, particularly in the performance area. This has both NetBSD and FreeBSD attachment code and a fair amount of effort has been put into making it easy to port to different *BSD platforms. The basic design is a one tfd per mbuf transmit (with no transmit related interrupts- tfds are gc'd as needed). The receive ring uses a 2K buffer per rfd with a +2 byte adjust for the ethernet header (so the payload is aligned). There's support that *almost* works for doing large packets- the rfd chaining code works, but there's some problem with getting good checksums at the IP reassembly level (ditto for doing short tfd's too). The chip has support for TCP checksums insertion for transmit and TCP checksum calculation on receive (for both you have to do some appropriate backoff && twiddling), but this isn't in place. This is nearly entirely reverse engineered from the released Intel driver, so there's a lot of "We have to do this but do not know why" stuff. There is somebody who has the chip specs who works in FreeBSD but they're being a bit standoffish about even sharing hints which is somewhat annoying. It's also apparent that all I had to work with were the first rev boards. This driver has been lightly tested on intel && alpha, but only point-to-point. There may be some issues with switches- use of boot time environment variables that override EEPROM settings (e.g., 'set wx_ilos=1' which inverts the sense of optical signal loss) may help with this. I had this out for review for three weeks, and nobody said anything negative or positive, ergo, this checkin has no 'reviewed by' field which I would have preferred.
2000-01-04 11:12:42 +00:00
/*
* Copyright (c) 1999, Traakan Software
* 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 unmodified, this list of conditions, and the following
* disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* $FreeBSD$
*/
/*
* Intel Gigabit Ethernet (82452) Driver.
* Inspired by fxp driver by David Greenman for FreeBSD, and by
* Bill Paul's work in other FreeBSD network drivers.
*/
/*
* Options
*/
/*
* Use only every other 16 byte receive descriptor, leaving the ones
* in between empty. This card is most efficient at reading/writing
* 32 byte cache lines, so avoid all the (not working for early rev
* cards) MWI and/or READ/MODIFY/WRITE cycles updating one descriptor
* would have you do.
*
* This isn't debugged yet.
*/
/* #define PADDED_CELL 1 */
Add first pass of the Intel Gigabit Ethernet (wiseman) driver. This driver seems relatively functional, but could use some souping up, particularly in the performance area. This has both NetBSD and FreeBSD attachment code and a fair amount of effort has been put into making it easy to port to different *BSD platforms. The basic design is a one tfd per mbuf transmit (with no transmit related interrupts- tfds are gc'd as needed). The receive ring uses a 2K buffer per rfd with a +2 byte adjust for the ethernet header (so the payload is aligned). There's support that *almost* works for doing large packets- the rfd chaining code works, but there's some problem with getting good checksums at the IP reassembly level (ditto for doing short tfd's too). The chip has support for TCP checksums insertion for transmit and TCP checksum calculation on receive (for both you have to do some appropriate backoff && twiddling), but this isn't in place. This is nearly entirely reverse engineered from the released Intel driver, so there's a lot of "We have to do this but do not know why" stuff. There is somebody who has the chip specs who works in FreeBSD but they're being a bit standoffish about even sharing hints which is somewhat annoying. It's also apparent that all I had to work with were the first rev boards. This driver has been lightly tested on intel && alpha, but only point-to-point. There may be some issues with switches- use of boot time environment variables that override EEPROM settings (e.g., 'set wx_ilos=1' which inverts the sense of optical signal loss) may help with this. I had this out for review for three weeks, and nobody said anything negative or positive, ergo, this checkin has no 'reviewed by' field which I would have preferred.
2000-01-04 11:12:42 +00:00
/*
* Since the includes are a mess, they'll all be in if_wxvar.h
*/
#if defined(__NetBSD__)
#include <dev/pci/if_wxvar.h>
#elif defined(__FreeBSD__)
#include <pci/if_wxvar.h>
#endif
#ifdef __alpha__
#undef vtophys
#define vtophys(va) alpha_XXX_dmamap((vm_offset_t)(va))
#endif /* __alpha__ */
/*
* Function Prototpes, yadda yadda...
*/
static int wx_intr __P((void *));
static void wx_handle_link_intr __P((wx_softc_t *));
static void wx_handle_rxint __P((wx_softc_t *));
static void wx_gc __P((wx_softc_t *));
static void wx_start __P((struct ifnet *));
static int wx_ioctl __P((struct ifnet *, IOCTL_CMD_TYPE, caddr_t));
static int wx_ifmedia_upd __P((struct ifnet *));
static void wx_ifmedia_sts __P((struct ifnet *, struct ifmediareq *));
static int wx_init __P((void *));
static void wx_hw_stop __P((wx_softc_t *));
static void wx_set_addr __P((wx_softc_t *, int, u_int8_t *));
static int wx_hw_initialize __P((wx_softc_t *));
static void wx_stop __P((wx_softc_t *));
static void wx_watchdog __P((struct ifnet *));
static int wx_get_rbuf __P((wx_softc_t *, rxpkt_t *));
static void wx_rxdma_map __P((wx_softc_t *, rxpkt_t *, struct mbuf *));
static INLINE void wx_eeprom_raise_clk __P((wx_softc_t *, u_int32_t));
static INLINE void wx_eeprom_lower_clk __P((wx_softc_t *, u_int32_t));
static INLINE void wx_eeprom_sobits __P((wx_softc_t *, u_int16_t, u_int16_t));
static INLINE u_int16_t wx_eeprom_sibits __P((wx_softc_t *));
static INLINE void wx_eeprom_cleanup __P((wx_softc_t *));
static INLINE u_int16_t wx_read_eeprom_word __P((wx_softc_t *, int));
static void wx_read_eeprom __P((wx_softc_t *, u_int16_t *, int, int));
static int wx_attach_common __P((wx_softc_t *));
static void wx_stats_update __P((void *));
static INLINE void wx_mwi_whackon __P((wx_softc_t *));
static INLINE void wx_mwi_unwhack __P((wx_softc_t *));
static int wx_dring_setup __P((wx_softc_t *));
static void wx_dring_teardown __P((wx_softc_t *));
#define WX_DISABLE_INT(sc) WRITE_CSR(sc, WXREG_IMCLR, WXDISABLE)
#define WX_ENABLE_INT(sc) WRITE_CSR(sc, WXREG_IMASK, sc->wx_ienable)
#if defined(__NetBSD__)
#ifdef __BROKEN_INDIRECT_CONFIG
#define MATCHARG void *
#else
#define MATCHARG struct cfdata *
#endif
static int wx_match __P((struct device *, MATCHARG, void *));
static void wx_attach __P((struct device *, struct device *, void *));
static void wx_shutdown __P((void *));
static int wx_ether_ioctl __P((struct ifnet *, IOCTL_CMD_TYPE, caddr_t));
static int wx_mc_setup __P((wx_softc_t *));
#define ether_ioctl wx_ether_ioctl
/*
* Life *should* be simple- we only read/write 32 bit values in registers.
* Unfortunately, some platforms define bus_space functions in a fashion
* such that they cannot be used as part of a for loop, for example.
*/
static INLINE u_int32_t _read_csr __P((struct wx_softc *, u_int32_t));
static INLINE void _write_csr __P((struct wx_softc *, u_int32_t, u_int32_t));
static INLINE u_int32_t _read_csr(sc, reg)
struct wx_softc *sc;
u_int32_t reg;
{
return bus_space_read_4(sc->w.st, sc->w.sh, reg);
}
static INLINE void _write_csr(sc, reg, val)
struct wx_softc *sc;
u_int32_t reg;
u_int32_t val;
{
bus_space_write_4(sc->w.st, sc->w.sh, reg, val);
}
static wx_softc_t *wxlist;
struct cfattach wx_ca = {
sizeof (wx_softc_t), wx_match, wx_attach
};
/*
* Check if a device is an 82452.
*/
static int
wx_match(parent, match, aux)
struct device *parent;
MATCHARG match;
void *aux;
{
struct pci_attach_args *pa = aux;
if (PCI_VENDOR(pa->pa_id) != WX_VENDOR_INTEL) {
return (0);
}
if (PCI_PRODUCT(pa->pa_id) != WX_PRODUCT_82452) {
return (0);
}
return (1);
}
static void
wx_attach(parent, self, aux)
struct device *parent, *self;
void *aux;
{
wx_softc_t *tmp, *sc = (wx_softc_t *)self;
struct pci_attach_args *pa = aux;
pci_chipset_tag_t pc = pa->pa_pc;
pci_intr_handle_t ih;
const char *intrstr = NULL;
u_int32_t data;
struct ifnet *ifp;
sc->w.pci_pc = pa->pa_pc;
sc->w.pci_tag = pa->pa_tag;
/*
* Map control/status registers.
*/
if (pci_mapreg_map(pa, WX_MMBA, PCI_MAPREG_TYPE_MEM, 0,
&sc->w.st, &sc->w.sh, NULL, NULL)) {
printf(": can't map registers\n");
return;
}
printf(": Intel GigaBit Ethernet\n");
/*
* Allocate our interrupt.
*/
if (pci_intr_map(pc, pa->pa_intrtag, pa->pa_intrpin,
pa->pa_intrline, &ih)) {
printf("%s: couldn't map interrupt\n", sc->wx_name);
return;
}
intrstr = pci_intr_string(pc, ih);
sc->w.ih = pci_intr_establish(pc, ih, IPL_NET, wx_intr, sc);
if (sc->w.ih == NULL) {
printf("%s: couldn't establish interrupt", sc->wx_name);
if (intrstr != NULL)
printf(" at %s", intrstr);
printf("\n");
return;
}
printf("%s: interrupting at %s\n", sc->wx_name, intrstr);
sc->revision =
pci_conf_read(pa->pa_pc, pa->pa_tag, PCI_CLASS_REG) & 0xff;
data = pci_conf_read(pa->pa_pc, pa->pa_tag, PCI_BHLC_REG);
data &= ~(PCI_CACHELINE_MASK << PCI_CACHELINE_SHIFT);
data |= (WX_CACHELINE_SIZE << PCI_CACHELINE_SHIFT);
Add first pass of the Intel Gigabit Ethernet (wiseman) driver. This driver seems relatively functional, but could use some souping up, particularly in the performance area. This has both NetBSD and FreeBSD attachment code and a fair amount of effort has been put into making it easy to port to different *BSD platforms. The basic design is a one tfd per mbuf transmit (with no transmit related interrupts- tfds are gc'd as needed). The receive ring uses a 2K buffer per rfd with a +2 byte adjust for the ethernet header (so the payload is aligned). There's support that *almost* works for doing large packets- the rfd chaining code works, but there's some problem with getting good checksums at the IP reassembly level (ditto for doing short tfd's too). The chip has support for TCP checksums insertion for transmit and TCP checksum calculation on receive (for both you have to do some appropriate backoff && twiddling), but this isn't in place. This is nearly entirely reverse engineered from the released Intel driver, so there's a lot of "We have to do this but do not know why" stuff. There is somebody who has the chip specs who works in FreeBSD but they're being a bit standoffish about even sharing hints which is somewhat annoying. It's also apparent that all I had to work with were the first rev boards. This driver has been lightly tested on intel && alpha, but only point-to-point. There may be some issues with switches- use of boot time environment variables that override EEPROM settings (e.g., 'set wx_ilos=1' which inverts the sense of optical signal loss) may help with this. I had this out for review for three weeks, and nobody said anything negative or positive, ergo, this checkin has no 'reviewed by' field which I would have preferred.
2000-01-04 11:12:42 +00:00
pci_conf_write(pa->pa_pc, pa->pa_tag, PCI_BHLC_REG, data);
if (wx_attach_common(sc)) {
return;
}
printf("%s: Ethernet address %s\n",
sc->wx_name, ether_sprintf(sc->wx_enaddr));
ifp = &sc->w.ethercom.ec_if;
bcopy(sc->wx_name, ifp->if_xname, IFNAMSIZ);
ifp->if_softc = sc;
ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
ifp->if_ioctl = wx_ioctl;
ifp->if_start = wx_start;
ifp->if_watchdog = wx_watchdog;
/*
* Attach the interface.
*/
if_attach(ifp);
ether_ifattach(ifp, sc->wx_enaddr);
#if NBPFILTER > 0
bpfattach(&sc->w.ethercom.ec_if.if_bpf, ifp, DLT_EN10MB,
sizeof (struct ether_header));
#endif
/*
* Add shutdown hook so that DMA is disabled prior to reboot. Not
* doing do could allow DMA to corrupt kernel memory during the
* reboot before the driver initializes.
*/
shutdownhook_establish(wx_shutdown, sc);
tmp = wxlist;
if (tmp) {
while (tmp->wx_next)
tmp = tmp->wx_next;
tmp->wx_next = sc;
} else {
wxlist = sc;
}
}
/*
* Device shutdown routine. Called at system shutdown after sync. The
* main purpose of this routine is to shut off receiver DMA so that
* kernel memory doesn't get clobbered during warmboot.
*/
static void
wx_shutdown(sc)
void *sc;
{
wx_hw_stop((wx_softc_t *) sc);
}
static int
wx_ether_ioctl(ifp, cmd, data)
struct ifnet *ifp;
IOCTL_CMD_TYPE cmd;
caddr_t data;
{
struct ifaddr *ifa = (struct ifaddr *) data;
int error = 0;
wx_softc_t *sc = SOFTC_IFP(ifp);
switch (cmd) {
case SIOCSIFADDR:
ifp->if_flags |= IFF_UP;
error = wx_init(sc);
if (error) {
ifp->if_flags &= ~IFF_UP;
break;
}
switch (ifa->ifa_addr->sa_family) {
#ifdef INET
case AF_INET:
arp_ifinit(ifp, ifa);
break;
#endif
#ifdef NS
case AF_NS:
{
register struct ns_addr *ina = &IA_SNS(ifa)->sns_addr;
if (ns_nullhost(*ina))
ina->x_host = *(union ns_host *)
LLADDR(ifp->if_sadl);
else
bcopy(ina->x_host.c_host, LLADDR(ifp->if_sadl),
ifp->if_addrlen);
break;
}
#endif
default:
break;
}
break;
default:
error = EINVAL;
break;
}
return (0);
}
/*
* Program multicast addresses.
*
* This function must be called at splimp, but it may sleep.
*/
static int
wx_mc_setup(sc)
wx_softc_t *sc;
{
struct ifnet *ifp = &sc->wx_if;
struct ether_multistep step;
struct ether_multi *enm;
/*
* XXX: drain TX queue- use a tsleep/wakeup until done.
*/
if (sc->tactive) {
return (EBUSY);
}
wx_stop(sc);
if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
sc->all_mcasts = 1;
return (wx_init(sc));
}
ETHER_FIRST_MULTI(step, &sc->w.ethercom, enm);
while (enm != NULL) {
if (memcmp(enm->enm_addrlo, enm->enm_addrhi, 6) != 0)
continue;
if (sc->wx_nmca >= WX_RAL_TAB_SIZE-1) {
sc->wx_nmca = 0;
sc->all_mcasts = 1;
break;
}
bcopy(enm->enm_addrlo,
(void *) &sc->wx_mcaddr[sc->wx_nmca++][0], 6);
ETHER_NEXT_MULTI(step, enm);
}
return (wx_init(sc));
}
static INLINE void
wx_mwi_whackon(sc)
wx_softc_t *sc;
{
sc->wx_cmdw =
pci_conf_read(sc->w.pci_pc, sc->w.pci_tag, PCI_COMMAND_STATUS_REG);
pci_conf_write(sc->w.pci_pc, sc->w.pci_tag,
PCI_COMMAND_STATUS_REG, sc->wx_cmdw & ~MWI);
}
static INLINE void
wx_mwi_unwhack(sc)
wx_softc_t *sc;
{
if (sc->wx_cmdw & MWI) {
pci_conf_write(sc->w.pci_pc, sc->w.pci_tag,
PCI_COMMAND_STATUS_REG, sc->wx_cmdw & ~MWI);
}
}
static int
wx_dring_setup(sc)
wx_softc_t *sc;
{
size_t len;
len = sizeof (wxrd_t) * WX_MAX_RDESC;
if (len > NBPG) {
printf("%s: len (%lx) over a page for the receive ring\n",
sc->wx_name, len);
return (-1);
}
len = NBPG;
sc->rdescriptors = (wxrd_t *) WXMALLOC(len);
if (sc->rdescriptors == NULL) {
printf("%s: could not allocate rcv descriptors\n", sc->wx_name);
return (-1);
}
if (((u_long)sc->rdescriptors) & 0xfff) {
printf("%s: rcv descriptors not 4KB aligned\n", sc->wx_name);
return (-1);
}
bzero(sc->rdescriptors, len);
len = sizeof (wxtd_t) * WX_MAX_TDESC;
if (len > NBPG) {
printf("%s: len (%lx) over a page for the xmit ring\n",
sc->wx_name, len);
return (-1);
}
len = NBPG;
sc->tdescriptors = (wxtd_t *) WXMALLOC(len);
if (sc->tdescriptors == NULL) {
printf("%s: could not allocate xmt descriptors\n", sc->wx_name);
return (-1);
}
if (((u_long)sc->tdescriptors) & 0xfff) {
printf("%s: xmt descriptors not 4KB aligned\n", sc->wx_name);
return (-1);
}
bzero(sc->tdescriptors, len);
return (0);
}
static void
wx_dring_teardown(sc)
wx_softc_t *sc;
{
if (sc->rdescriptors) {
WXFREE(sc->rdescriptors);
sc->rdescriptors = NULL;
}
if (sc->tdescriptors) {
WXFREE(sc->tdescriptors);
sc->tdescriptors = NULL;
}
}
#elif defined(__FreeBSD__)
static int wx_mc_setup __P((wx_softc_t *));
/*
* Program multicast addresses.
*
* This function must be called at splimp, but it may sleep.
*/
static int
wx_mc_setup(sc)
wx_softc_t *sc;
{
struct ifnet *ifp = &sc->wx_if;
struct ifmultiaddr *ifma;
/*
* XXX: drain TX queue- use a tsleep/wakeup until done.
*/
if (sc->tactive) {
return (EBUSY);
}
wx_stop(sc);
if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
sc->all_mcasts = 1;
return (wx_init(sc));
}
for (ifma = ifp->if_multiaddrs.lh_first, sc->wx_nmca = 0;
ifma != NULL; ifma = ifma->ifma_link.le_next) {
if (ifma->ifma_addr->sa_family != AF_LINK) {
continue;
}
if (sc->wx_nmca >= WX_RAL_TAB_SIZE-1) {
sc->wx_nmca = 0;
sc->all_mcasts = 1;
break;
}
bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
(void *) &sc->wx_mcaddr[sc->wx_nmca++][0], 6);
}
return (wx_init(sc));
}
/*
* Return identification string if this is device is ours.
*/
static wx_softc_t *wxlist;
static int
wx_probe(device_t dev)
{
if ((pci_get_vendor(dev) == WX_VENDOR_INTEL) &&
(pci_get_device(dev) == WX_PRODUCT_82452)) {
device_set_desc(dev, "Intel GigaBit Ethernet");
return 0;
}
return (ENXIO);
}
static int
wx_attach(device_t dev)
{
int error = 0;
wx_softc_t *tmp, *sc = device_get_softc(dev);
struct ifnet *ifp;
int s;
u_long val;
int rid;
bzero(sc, sizeof (wx_softc_t));
callout_handle_init(&sc->w.sch);
sc->w.dev = dev;
if (getenv_int("wx_debug", &rid)) {
if (rid & (1 << device_get_unit(dev))) {
sc->wx_debug = 1;
}
}
if (getenv_int("wx_no_ilos", &rid)) {
if (rid & (1 << device_get_unit(dev))) {
sc->wx_no_ilos = 1;
}
}
if (getenv_int("wx_ilos", &rid)) {
if (rid & (1 << device_get_unit(dev))) {
sc->wx_ilos = 1;
}
}
if (getenv_int("wx_no_flow", &rid)) {
if (rid & (1 << device_get_unit(dev))) {
sc->wx_no_flow = 1;
}
}
s = splimp();
/*
* Enable bus mastering, make sure that the cache line size is right.
*/
val = pci_read_config(dev, PCIR_COMMAND, 2);
val |= (PCIM_CMD_MEMEN|PCIM_CMD_BUSMASTEREN);
pci_write_config(dev, PCIR_COMMAND, val, 2);
val = pci_read_config(dev, PCIR_CACHELNSZ, 1);
if (val != 0x10) {
pci_write_config(dev, PCIR_CACHELNSZ, 0x10, 1);
}
/*
* get revision
*/
sc->revision = pci_read_config(dev, PCIR_CLASS, 1);
/*
* Map control/status registers.
*/
rid = WX_MMBA;
sc->w.mem = bus_alloc_resource(dev, SYS_RES_MEMORY,
&rid, 0, ~0, 1, RF_ACTIVE);
if (!sc->w.mem) {
device_printf(dev, "could not map memory\n");
error = ENXIO;
goto out;
}
sc->w.st = rman_get_bustag(sc->w.mem);
sc->w.sh = rman_get_bushandle(sc->w.mem);
rid = 0;
sc->w.irq = bus_alloc_resource(dev, SYS_RES_IRQ,
&rid, 0, ~0, 1, RF_SHAREABLE | RF_ACTIVE);
if (sc->w.irq == NULL) {
device_printf(dev, "could not map interrupt\n");
error = ENXIO;
goto out;
}
error = bus_setup_intr(dev, sc->w.irq, INTR_TYPE_NET,
(void (*)(void *))wx_intr, sc, &sc->w.ih);
if (error) {
device_printf(dev, "could not setup irq\n");
goto out;
}
if (wx_attach_common(sc)) {
bus_teardown_intr(dev, sc->w.irq, sc->w.ih);
bus_release_resource(dev, SYS_RES_IRQ, 0, sc->w.irq);
bus_release_resource(dev, SYS_RES_MEMORY, WX_MMBA, sc->w.mem);
error = ENXIO;
goto out;
}
device_printf(dev, "Ethernet address %02x:%02x:%02x:%02x:%02x:%02x\n",
sc->w.arpcom.ac_enaddr[0], sc->w.arpcom.ac_enaddr[1],
sc->w.arpcom.ac_enaddr[2], sc->w.arpcom.ac_enaddr[3],
sc->w.arpcom.ac_enaddr[4], sc->w.arpcom.ac_enaddr[5]);
(void) snprintf(sc->wx_name, sizeof (sc->wx_name) - 1, "wx%d",
device_get_unit(dev));
ifp = &sc->w.arpcom.ac_if;
ifp->if_unit = device_get_unit(dev);
ifp->if_name = "wx";
ifp->if_output = ether_output;
ifp->if_baudrate = 1000000000;
ifp->if_init = (void (*)(void *))wx_init;
ifp->if_softc = sc;
ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
ifp->if_ioctl = wx_ioctl;
ifp->if_start = wx_start;
ifp->if_watchdog = wx_watchdog;
if_attach(ifp);
ifp->if_snd.ifq_maxlen = WX_MAX_TDESC - 1;
ether_ifattach(ifp);
bpfattach(ifp, DLT_EN10MB, sizeof(struct ether_header));
tmp = wxlist;
if (tmp) {
while (tmp->wx_next)
tmp = tmp->wx_next;
tmp->wx_next = sc;
} else {
wxlist = sc;
}
out:
splx(s);
return (error);
}
static int
wx_detach(device_t dev)
{
wx_softc_t *sc = device_get_softc(dev);
int s = splimp();
if_detach(&sc->w.arpcom.ac_if);
wx_stop(sc);
bus_teardown_intr(dev, sc->w.irq, sc->w.ih);
bus_release_resource(dev, SYS_RES_IRQ, 0, sc->w.irq);
bus_release_resource(dev, SYS_RES_MEMORY, WX_MMBA, sc->w.mem);
splx(s);
return (0);
}
static int
wx_shutdown(device_t dev)
{
wx_hw_stop((wx_softc_t *) device_get_softc(dev));
return (0);
}
static INLINE void
wx_mwi_whackon(sc)
wx_softc_t *sc;
{
sc->wx_cmdw = pci_read_config(sc->w.dev, PCIR_COMMAND, 2);
pci_write_config(sc->w.dev, PCIR_COMMAND, sc->wx_cmdw & ~MWI, 2);
}
static INLINE void
wx_mwi_unwhack(sc)
wx_softc_t *sc;
{
if (sc->wx_cmdw & MWI) {
pci_write_config(sc->w.dev, PCIR_COMMAND, sc->wx_cmdw, 2);
}
}
static int
wx_dring_setup(sc)
wx_softc_t *sc;
{
size_t len;
len = sizeof (wxrd_t) * WX_MAX_RDESC;
sc->rdescriptors = (wxrd_t *)
contigmalloc(len, M_DEVBUF, M_NOWAIT, 0, ~0, 4096, 0);
if (sc->rdescriptors == NULL) {
printf("%s: could not allocate rcv descriptors\n", sc->wx_name);
return (-1);
}
if (((u_long)sc->rdescriptors) & 0xfff) {
printf("%s: rcv descriptors not 4KB aligned\n", sc->wx_name);
return (-1);
}
bzero(sc->rdescriptors, len);
len = sizeof (wxtd_t) * WX_MAX_TDESC;
sc->tdescriptors = (wxtd_t *)
contigmalloc(len, M_DEVBUF, M_NOWAIT, 0, ~0, 4096, 0);
if (sc->tdescriptors == NULL) {
printf("%s: could not allocate xmt descriptors\n", sc->wx_name);
return (-1);
}
if (((u_long)sc->tdescriptors) & 0xfff) {
printf("%s: xmt descriptors not 4KB aligned\n", sc->wx_name);
return (-1);
}
bzero(sc->tdescriptors, len);
return (0);
}
static void
wx_dring_teardown(sc)
wx_softc_t *sc;
{
if (sc->rdescriptors) {
WXFREE(sc->rdescriptors);
sc->rdescriptors = NULL;
}
if (sc->tdescriptors) {
WXFREE(sc->tdescriptors);
sc->tdescriptors = NULL;
}
}
static device_method_t wx_methods[] = {
/* Device interface */
DEVMETHOD(device_probe, wx_probe),
DEVMETHOD(device_attach, wx_attach),
DEVMETHOD(device_detach, wx_detach),
DEVMETHOD(device_shutdown, wx_shutdown),
{ 0, 0 }
};
static driver_t wx_driver = {
"wx", wx_methods, sizeof(wx_softc_t),
};
static devclass_t wx_devclass;
DRIVER_MODULE(if_wx, pci, wx_driver, wx_devclass, 0, 0);
#endif
/*
* Do generic parts of attach. Our registers have been mapped
* and our interrupt registered.
*/
static int
wx_attach_common(sc)
wx_softc_t *sc;
{
size_t len;
u_int32_t tmp;
int ll = 0;
/*
* First, reset the chip.
*/
wx_hw_stop(sc);
/*
* Second, validate our EEPROM.
*/
/* TBD */
/*
* Third, read eeprom for our MAC address and other things.
*/
wx_read_eeprom(sc, (u_int16_t *)sc->wx_enaddr, WX_EEPROM_MAC_OFF, 3);
/*
* Fourth, establish some adapter parameters.
*/
sc->wx_txint_delay = 128;
ifmedia_init(&sc->wx_media, IFM_IMASK, wx_ifmedia_upd, wx_ifmedia_sts);
ifmedia_add(&sc->wx_media, IFM_ETHER|IFM_1000_SX, 0, NULL);
ifmedia_add(&sc->wx_media, IFM_ETHER|IFM_1000_SX|IFM_FDX, 0, NULL);
ifmedia_set(&sc->wx_media, IFM_ETHER|IFM_1000_SX|IFM_FDX);
sc->wx_media.ifm_media = sc->wx_media.ifm_cur->ifm_media;
ll += 1;
/*
* Fifth, establish a default device control register word.
*/
sc->wx_dcr = 0;
if (sc->wx_cfg1 & WX_EEPROM_CTLR1_FD)
sc->wx_dcr |= WXDCR_FD;
if (sc->wx_cfg1 & WX_EEPROM_CTLR1_ILOS)
sc->wx_dcr |= WXDCR_ILOS;
tmp = (sc->wx_cfg1 >> WX_EEPROM_CTLR1_SWDPIO_SHIFT) & WXDCR_SWDPIO_MASK;
sc->wx_dcr |= (tmp << WXDCR_SWDPIO_SHIFT);
if (sc->wx_no_ilos)
sc->wx_dcr &= ~WXDCR_ILOS;
if (sc->wx_ilos)
sc->wx_dcr |= WXDCR_ILOS;
if (sc->wx_no_flow == 0)
sc->wx_dcr |= WXDCR_RFCE | WXDCR_TFCE;
/*
* Sixth, allocate various sw structures...
*/
len = sizeof (rxpkt_t) * WX_MAX_RDESC;
sc->rbase = (rxpkt_t *) WXMALLOC(len);
if (sc->rbase == NULL) {
goto fail;
}
bzero(sc->rbase, len);
ll += 1;
len = sizeof (txpkt_t) * WX_MAX_TDESC;
sc->tbase = (txpkt_t *) WXMALLOC(len);
if (sc->tbase == NULL) {
goto fail;
}
bzero(sc->tbase, len);
ll += 1;
/*
* Seventh, allocate and dma map (platform dependent) descriptor rings.
* They have to be aligned on a 4KB boundary.
*/
if (wx_dring_setup(sc) == 0) {
return (0);
}
fail:
printf("%s: failed to do common attach (%d)\n", sc->wx_name, ll);
wx_dring_teardown(sc);
if (sc->rbase) {
WXFREE(sc->rbase);
sc->rbase = NULL;
}
if (sc->tbase) {
WXFREE(sc->tbase);
sc->tbase = NULL;
}
return (ENOMEM);
}
/*
* EEPROM functions.
*/
static INLINE void
wx_eeprom_raise_clk(sc, regval)
wx_softc_t *sc;
u_int32_t regval;
{
WRITE_CSR(sc, WXREG_EECDR, regval | WXEECD_SK);
DELAY(50);
}
static INLINE void
wx_eeprom_lower_clk(sc, regval)
wx_softc_t *sc;
u_int32_t regval;
{
WRITE_CSR(sc, WXREG_EECDR, regval & ~WXEECD_SK);
DELAY(50);
}
static INLINE void
wx_eeprom_sobits(sc, data, count)
wx_softc_t *sc;
u_int16_t data;
u_int16_t count;
{
u_int32_t regval, mask;
mask = 1 << (count - 1);
regval = READ_CSR(sc, WXREG_EECDR) & ~(WXEECD_DI|WXEECD_DO);
do {
if (data & mask)
regval |= WXEECD_DI;
else
regval &= ~WXEECD_DI;
WRITE_CSR(sc, WXREG_EECDR, regval); DELAY(50);
wx_eeprom_raise_clk(sc, regval);
wx_eeprom_lower_clk(sc, regval);
mask >>= 1;
} while (mask != 0);
WRITE_CSR(sc, WXREG_EECDR, regval & ~WXEECD_DI);
}
static INLINE u_int16_t
wx_eeprom_sibits(sc)
wx_softc_t *sc;
{
unsigned int regval, i;
u_int16_t data;
data = 0;
regval = READ_CSR(sc, WXREG_EECDR) & ~(WXEECD_DI|WXEECD_DO);
for (i = 0; i != 16; i++) {
data <<= 1;
wx_eeprom_raise_clk(sc, regval);
regval = READ_CSR(sc, WXREG_EECDR) & ~WXEECD_DI;
if (regval & WXEECD_DO) {
data |= 1;
}
wx_eeprom_lower_clk(sc, regval);
}
return (data);
}
static INLINE void
wx_eeprom_cleanup(sc)
wx_softc_t *sc;
{
u_int32_t regval;
regval = READ_CSR(sc, WXREG_EECDR) & ~(WXEECD_DI|WXEECD_CS);
WRITE_CSR(sc, WXREG_EECDR, regval); DELAY(50);
wx_eeprom_raise_clk(sc, regval);
wx_eeprom_lower_clk(sc, regval);
}
static u_int16_t INLINE
wx_read_eeprom_word(sc, offset)
wx_softc_t *sc;
int offset;
{
u_int16_t data;
WRITE_CSR(sc, WXREG_EECDR, WXEECD_CS);
wx_eeprom_sobits(sc, EEPROM_READ_OPCODE, 3);
wx_eeprom_sobits(sc, offset, 6);
data = wx_eeprom_sibits(sc);
wx_eeprom_cleanup(sc);
return (data);
}
static void
wx_read_eeprom(sc, data, offset, words)
wx_softc_t *sc;
u_int16_t *data;
int offset;
int words;
{
int i;
for (i = 0; i < words; i++) {
*data++ = wx_read_eeprom_word(sc, offset++);
}
sc->wx_cfg1 = wx_read_eeprom_word(sc, WX_EEPROM_CTLR1_OFF);
}
/*
* Start packet transmission on the interface.
*/
static void
wx_start(ifp)
struct ifnet *ifp;
{
wx_softc_t *sc = SOFTC_IFP(ifp);
u_int16_t cidx, nactv;
nactv = sc->tactive;
while (nactv < WX_MAX_TDESC) {
int ndesc;
int gctried = 0;
struct mbuf *m, *mb_head;
IF_DEQUEUE(&ifp->if_snd, mb_head);
if (mb_head == NULL) {
break;
}
sc->wx_xmitwanted++;
/*
* If we have a packet less than ethermin, pad it out.
*/
if (mb_head->m_pkthdr.len < WX_MIN_RPKT_SIZE) {
MGETHDR(m, M_DONTWAIT, MT_DATA);
if (m == NULL) {
m_freem(mb_head);
break;
}
m_copydata(mb_head, 0, mb_head->m_pkthdr.len,
mtod(m, caddr_t));
m->m_pkthdr.len = m->m_len = WX_MIN_RPKT_SIZE;
bzero(mtod(m, char *) + mb_head->m_pkthdr.len,
WX_MIN_RPKT_SIZE - mb_head->m_pkthdr.len);
sc->wx_xmitpullup++;
m_freem(mb_head);
mb_head = m;
}
again:
cidx = sc->tnxtfree;
nactv = sc->tactive;
/*
* Go through each of the mbufs in the chain and initialize
* the transmit buffer descriptors with the physical address
* and size of that mbuf. If we have a length less than our
* minimum transmit size, we bail (to do a pullup). If we run
* out of descriptors, we also bail and try and do a pullup.
*/
for (ndesc = 0, m = mb_head; m != NULL; m = m->m_next) {
vm_offset_t vptr;
wxtd_t *td;
/*
* If this mbuf has no data, skip it.
*/
if (m->m_len == 0) {
continue;
}
/*
* If this packet is too small for the chip's minimum,
* break out to to cluster it.
*/
if (m->m_len < WX_MIN_RPKT_SIZE) {
sc->wx_xmitrunt++;
break;
}
/*
* Do we have a descriptor available for this mbuf?
*/
if (++nactv == WX_MAX_TDESC) {
if (gctried++ == 0) {
sc->wx_xmitgc++;
wx_gc(sc);
goto again;
}
break;
}
sc->tbase[cidx].dptr = m;
td = &sc->tdescriptors[cidx];
td->length = m->m_len;
vptr = mtod(m, vm_offset_t);
td->address.highpart = 0;
td->address.lowpart = vtophys(vptr);
td->cso = 0;
td->status = 0;
td->special = 0;
td->cmd = 0;
td->css = 0;
/*
* Mark the last descriptor with EOP and tell the
* chip to insert a final checksum.
*/
if (m->m_next == NULL) {
td->cmd |= TXCMD_EOP|TXCMD_IFCS;
}
if (sc->wx_debug) {
printf("%s: XMIT[%d] %p vptr %lx (length %d "
"DMA addr %x) idx %d cmd %x\n", sc->wx_name,
ndesc, m, (long) vptr, td->length,
td->address.lowpart, cidx, td->cmd);
}
ndesc++;
cidx = T_NXT_IDX(cidx);
}
/*
* If we get here and m is NULL, we can send
* the the packet chain described by mb_head.
*/
if (m == NULL) {
sc->tbase[sc->tnxtfree].sidx = sc->tnxtfree;
sc->tbase[sc->tnxtfree].eidx = cidx;
sc->tbase[sc->tnxtfree].next = NULL;
if (sc->tbsyf) {
sc->tbsyl->next = &sc->tbase[sc->tnxtfree];
} else {
sc->tbsyf = &sc->tbase[sc->tnxtfree];
}
sc->tbsyl = &sc->tbase[sc->tnxtfree];
sc->tnxtfree = cidx;
sc->tactive = nactv;
ifp->if_timer = 10;
WRITE_CSR(sc, WXREG_TDT, cidx);
if (ifp->if_bpf)
bpf_mtap(WX_BPFTAP_ARG(ifp), mb_head);
continue;
}
/*
* Otherwise, we couldn't send this packet for some reason.
*
* If don't have a descriptor available, and this is a
* single mbuf packet, freeze output so that later we
* can restart when we have more room. Otherwise, we'll
* try and cluster the request. We've already tried to
* garbage collect completed descriptors.
*/
if (nactv == WX_MAX_TDESC && mb_head->m_next == NULL) {
sc->wx_xmitputback++;
ifp->if_flags |= IFF_OACTIVE;
IF_PREPEND(&ifp->if_snd, mb_head);
break;
}
/*
* Otherwise, it's either a fragment length somewhere in the
* chain that isn't at least WX_MIN_XPKT_SIZE in length or
* the number of fragments exceeds the number of descriptors
* available.
*
* We could try a variety of strategies here- if this is
* a length problem for single mbuf packet or a length problem
* for the last mbuf in a chain (we could just try and adjust
* it), but it's just simpler to try and cluster it.
*/
MGETHDR(m, M_DONTWAIT, MT_DATA);
if (m == NULL) {
m_freem(mb_head);
break;
}
MCLGET(m, M_DONTWAIT);
if ((m->m_flags & M_EXT) == 0) {
m_freem(m);
m_freem(mb_head);
break;
}
m_copydata(mb_head, 0, mb_head->m_pkthdr.len, mtod(m, caddr_t));
m->m_pkthdr.len = m->m_len = mb_head->m_pkthdr.len;
m_freem(mb_head);
mb_head = m;
sc->wx_xmitcluster++;
goto again;
}
if (sc->tactive == WX_MAX_TDESC) {
sc->wx_xmitblocked++;
ifp->if_flags |= IFF_OACTIVE;
}
}
/*
* Process interface interrupts.
*/
static int
wx_intr(arg)
void *arg;
{
wx_softc_t *sc = arg;
int claimed = 0;
/*
* Read interrupt cause register. Reading it clears bits.
*/
sc->wx_icr = READ_CSR(sc, WXREG_ICR);
if (sc->wx_icr) {
claimed++;
WX_DISABLE_INT(sc);
sc->wx_intr++;
if (sc->wx_icr & (WXISR_LSC|WXISR_RXSEQ)) {
wx_handle_link_intr(sc);
}
wx_handle_rxint(sc);
if (sc->tactive) {
Add first pass of the Intel Gigabit Ethernet (wiseman) driver. This driver seems relatively functional, but could use some souping up, particularly in the performance area. This has both NetBSD and FreeBSD attachment code and a fair amount of effort has been put into making it easy to port to different *BSD platforms. The basic design is a one tfd per mbuf transmit (with no transmit related interrupts- tfds are gc'd as needed). The receive ring uses a 2K buffer per rfd with a +2 byte adjust for the ethernet header (so the payload is aligned). There's support that *almost* works for doing large packets- the rfd chaining code works, but there's some problem with getting good checksums at the IP reassembly level (ditto for doing short tfd's too). The chip has support for TCP checksums insertion for transmit and TCP checksum calculation on receive (for both you have to do some appropriate backoff && twiddling), but this isn't in place. This is nearly entirely reverse engineered from the released Intel driver, so there's a lot of "We have to do this but do not know why" stuff. There is somebody who has the chip specs who works in FreeBSD but they're being a bit standoffish about even sharing hints which is somewhat annoying. It's also apparent that all I had to work with were the first rev boards. This driver has been lightly tested on intel && alpha, but only point-to-point. There may be some issues with switches- use of boot time environment variables that override EEPROM settings (e.g., 'set wx_ilos=1' which inverts the sense of optical signal loss) may help with this. I had this out for review for three weeks, and nobody said anything negative or positive, ergo, this checkin has no 'reviewed by' field which I would have preferred.
2000-01-04 11:12:42 +00:00
wx_gc(sc);
}
if (sc->wx_if.if_snd.ifq_head != NULL) {
wx_start(&sc->wx_if);
}
WX_ENABLE_INT(sc);
}
return (claimed);
}
static void
wx_handle_link_intr(sc)
wx_softc_t *sc;
{
sc->wx_linkintr++;
printf("%s: link intr 0x%x\n", sc->wx_name, sc->wx_icr);
}
static void
wx_handle_rxint(sc)
wx_softc_t *sc;
{
struct ether_header *eh;
struct mbuf *m0, *mb, *pending[WX_MAX_RDESC];
Add first pass of the Intel Gigabit Ethernet (wiseman) driver. This driver seems relatively functional, but could use some souping up, particularly in the performance area. This has both NetBSD and FreeBSD attachment code and a fair amount of effort has been put into making it easy to port to different *BSD platforms. The basic design is a one tfd per mbuf transmit (with no transmit related interrupts- tfds are gc'd as needed). The receive ring uses a 2K buffer per rfd with a +2 byte adjust for the ethernet header (so the payload is aligned). There's support that *almost* works for doing large packets- the rfd chaining code works, but there's some problem with getting good checksums at the IP reassembly level (ditto for doing short tfd's too). The chip has support for TCP checksums insertion for transmit and TCP checksum calculation on receive (for both you have to do some appropriate backoff && twiddling), but this isn't in place. This is nearly entirely reverse engineered from the released Intel driver, so there's a lot of "We have to do this but do not know why" stuff. There is somebody who has the chip specs who works in FreeBSD but they're being a bit standoffish about even sharing hints which is somewhat annoying. It's also apparent that all I had to work with were the first rev boards. This driver has been lightly tested on intel && alpha, but only point-to-point. There may be some issues with switches- use of boot time environment variables that override EEPROM settings (e.g., 'set wx_ilos=1' which inverts the sense of optical signal loss) may help with this. I had this out for review for three weeks, and nobody said anything negative or positive, ergo, this checkin has no 'reviewed by' field which I would have preferred.
2000-01-04 11:12:42 +00:00
struct ifnet *ifp = &sc->wx_if;
int npkts, ndesc, lidx, idx, tlen;
Add first pass of the Intel Gigabit Ethernet (wiseman) driver. This driver seems relatively functional, but could use some souping up, particularly in the performance area. This has both NetBSD and FreeBSD attachment code and a fair amount of effort has been put into making it easy to port to different *BSD platforms. The basic design is a one tfd per mbuf transmit (with no transmit related interrupts- tfds are gc'd as needed). The receive ring uses a 2K buffer per rfd with a +2 byte adjust for the ethernet header (so the payload is aligned). There's support that *almost* works for doing large packets- the rfd chaining code works, but there's some problem with getting good checksums at the IP reassembly level (ditto for doing short tfd's too). The chip has support for TCP checksums insertion for transmit and TCP checksum calculation on receive (for both you have to do some appropriate backoff && twiddling), but this isn't in place. This is nearly entirely reverse engineered from the released Intel driver, so there's a lot of "We have to do this but do not know why" stuff. There is somebody who has the chip specs who works in FreeBSD but they're being a bit standoffish about even sharing hints which is somewhat annoying. It's also apparent that all I had to work with were the first rev boards. This driver has been lightly tested on intel && alpha, but only point-to-point. There may be some issues with switches- use of boot time environment variables that override EEPROM settings (e.g., 'set wx_ilos=1' which inverts the sense of optical signal loss) may help with this. I had this out for review for three weeks, and nobody said anything negative or positive, ergo, this checkin has no 'reviewed by' field which I would have preferred.
2000-01-04 11:12:42 +00:00
for (m0 = sc->rpending, tlen = ndesc = npkts = 0, idx = sc->rnxt,
lidx = R_PREV_IDX(idx); ndesc < WX_MAX_RDESC;
ndesc++, lidx = idx, idx = R_NXT_IDX(idx)) {
Add first pass of the Intel Gigabit Ethernet (wiseman) driver. This driver seems relatively functional, but could use some souping up, particularly in the performance area. This has both NetBSD and FreeBSD attachment code and a fair amount of effort has been put into making it easy to port to different *BSD platforms. The basic design is a one tfd per mbuf transmit (with no transmit related interrupts- tfds are gc'd as needed). The receive ring uses a 2K buffer per rfd with a +2 byte adjust for the ethernet header (so the payload is aligned). There's support that *almost* works for doing large packets- the rfd chaining code works, but there's some problem with getting good checksums at the IP reassembly level (ditto for doing short tfd's too). The chip has support for TCP checksums insertion for transmit and TCP checksum calculation on receive (for both you have to do some appropriate backoff && twiddling), but this isn't in place. This is nearly entirely reverse engineered from the released Intel driver, so there's a lot of "We have to do this but do not know why" stuff. There is somebody who has the chip specs who works in FreeBSD but they're being a bit standoffish about even sharing hints which is somewhat annoying. It's also apparent that all I had to work with were the first rev boards. This driver has been lightly tested on intel && alpha, but only point-to-point. There may be some issues with switches- use of boot time environment variables that override EEPROM settings (e.g., 'set wx_ilos=1' which inverts the sense of optical signal loss) may help with this. I had this out for review for three weeks, and nobody said anything negative or positive, ergo, this checkin has no 'reviewed by' field which I would have preferred.
2000-01-04 11:12:42 +00:00
wxrd_t *rd;
rxpkt_t *rxpkt;
int length, offset, lastframe;
rd = &sc->rdescriptors[idx];
if ((rd->status & RDSTAT_DD) == 0) {
if (m0) {
if (sc->rpending == NULL) {
m0->m_pkthdr.len = tlen;
sc->rpending = m0;
} else {
m_freem(m0);
}
Add first pass of the Intel Gigabit Ethernet (wiseman) driver. This driver seems relatively functional, but could use some souping up, particularly in the performance area. This has both NetBSD and FreeBSD attachment code and a fair amount of effort has been put into making it easy to port to different *BSD platforms. The basic design is a one tfd per mbuf transmit (with no transmit related interrupts- tfds are gc'd as needed). The receive ring uses a 2K buffer per rfd with a +2 byte adjust for the ethernet header (so the payload is aligned). There's support that *almost* works for doing large packets- the rfd chaining code works, but there's some problem with getting good checksums at the IP reassembly level (ditto for doing short tfd's too). The chip has support for TCP checksums insertion for transmit and TCP checksum calculation on receive (for both you have to do some appropriate backoff && twiddling), but this isn't in place. This is nearly entirely reverse engineered from the released Intel driver, so there's a lot of "We have to do this but do not know why" stuff. There is somebody who has the chip specs who works in FreeBSD but they're being a bit standoffish about even sharing hints which is somewhat annoying. It's also apparent that all I had to work with were the first rev boards. This driver has been lightly tested on intel && alpha, but only point-to-point. There may be some issues with switches- use of boot time environment variables that override EEPROM settings (e.g., 'set wx_ilos=1' which inverts the sense of optical signal loss) may help with this. I had this out for review for three weeks, and nobody said anything negative or positive, ergo, this checkin has no 'reviewed by' field which I would have preferred.
2000-01-04 11:12:42 +00:00
m0 = NULL;
}
if (sc->wx_debug) {
printf("WXRX: ndesc %d idx %d lidx %d\n",
ndesc, idx, lidx);
}
Add first pass of the Intel Gigabit Ethernet (wiseman) driver. This driver seems relatively functional, but could use some souping up, particularly in the performance area. This has both NetBSD and FreeBSD attachment code and a fair amount of effort has been put into making it easy to port to different *BSD platforms. The basic design is a one tfd per mbuf transmit (with no transmit related interrupts- tfds are gc'd as needed). The receive ring uses a 2K buffer per rfd with a +2 byte adjust for the ethernet header (so the payload is aligned). There's support that *almost* works for doing large packets- the rfd chaining code works, but there's some problem with getting good checksums at the IP reassembly level (ditto for doing short tfd's too). The chip has support for TCP checksums insertion for transmit and TCP checksum calculation on receive (for both you have to do some appropriate backoff && twiddling), but this isn't in place. This is nearly entirely reverse engineered from the released Intel driver, so there's a lot of "We have to do this but do not know why" stuff. There is somebody who has the chip specs who works in FreeBSD but they're being a bit standoffish about even sharing hints which is somewhat annoying. It's also apparent that all I had to work with were the first rev boards. This driver has been lightly tested on intel && alpha, but only point-to-point. There may be some issues with switches- use of boot time environment variables that override EEPROM settings (e.g., 'set wx_ilos=1' which inverts the sense of optical signal loss) may help with this. I had this out for review for three weeks, and nobody said anything negative or positive, ergo, this checkin has no 'reviewed by' field which I would have preferred.
2000-01-04 11:12:42 +00:00
break;
}
if (rd->errors != 0) {
printf("%s: packet with errors (%x)\n",
sc->wx_name, rd->errors);
rd->status = 0;
ifp->if_ierrors++;
if (m0) {
m_freem(m0);
m0 = NULL;
if (sc->rpending) {
m_freem(sc->rpending);
sc->rpending = NULL;
}
Add first pass of the Intel Gigabit Ethernet (wiseman) driver. This driver seems relatively functional, but could use some souping up, particularly in the performance area. This has both NetBSD and FreeBSD attachment code and a fair amount of effort has been put into making it easy to port to different *BSD platforms. The basic design is a one tfd per mbuf transmit (with no transmit related interrupts- tfds are gc'd as needed). The receive ring uses a 2K buffer per rfd with a +2 byte adjust for the ethernet header (so the payload is aligned). There's support that *almost* works for doing large packets- the rfd chaining code works, but there's some problem with getting good checksums at the IP reassembly level (ditto for doing short tfd's too). The chip has support for TCP checksums insertion for transmit and TCP checksum calculation on receive (for both you have to do some appropriate backoff && twiddling), but this isn't in place. This is nearly entirely reverse engineered from the released Intel driver, so there's a lot of "We have to do this but do not know why" stuff. There is somebody who has the chip specs who works in FreeBSD but they're being a bit standoffish about even sharing hints which is somewhat annoying. It's also apparent that all I had to work with were the first rev boards. This driver has been lightly tested on intel && alpha, but only point-to-point. There may be some issues with switches- use of boot time environment variables that override EEPROM settings (e.g., 'set wx_ilos=1' which inverts the sense of optical signal loss) may help with this. I had this out for review for three weeks, and nobody said anything negative or positive, ergo, this checkin has no 'reviewed by' field which I would have preferred.
2000-01-04 11:12:42 +00:00
}
continue;
}
rxpkt = &sc->rbase[idx];
mb = rxpkt->dptr;
if (mb == NULL) {
printf("%s: receive descriptor with no mbuf\n",
sc->wx_name);
(void) wx_get_rbuf(sc, rxpkt);
rd->status = 0;
ifp->if_ierrors++;
if (m0) {
m_freem(m0);
m0 = NULL;
if (sc->rpending) {
m_freem(sc->rpending);
sc->rpending = NULL;
}
Add first pass of the Intel Gigabit Ethernet (wiseman) driver. This driver seems relatively functional, but could use some souping up, particularly in the performance area. This has both NetBSD and FreeBSD attachment code and a fair amount of effort has been put into making it easy to port to different *BSD platforms. The basic design is a one tfd per mbuf transmit (with no transmit related interrupts- tfds are gc'd as needed). The receive ring uses a 2K buffer per rfd with a +2 byte adjust for the ethernet header (so the payload is aligned). There's support that *almost* works for doing large packets- the rfd chaining code works, but there's some problem with getting good checksums at the IP reassembly level (ditto for doing short tfd's too). The chip has support for TCP checksums insertion for transmit and TCP checksum calculation on receive (for both you have to do some appropriate backoff && twiddling), but this isn't in place. This is nearly entirely reverse engineered from the released Intel driver, so there's a lot of "We have to do this but do not know why" stuff. There is somebody who has the chip specs who works in FreeBSD but they're being a bit standoffish about even sharing hints which is somewhat annoying. It's also apparent that all I had to work with were the first rev boards. This driver has been lightly tested on intel && alpha, but only point-to-point. There may be some issues with switches- use of boot time environment variables that override EEPROM settings (e.g., 'set wx_ilos=1' which inverts the sense of optical signal loss) may help with this. I had this out for review for three weeks, and nobody said anything negative or positive, ergo, this checkin has no 'reviewed by' field which I would have preferred.
2000-01-04 11:12:42 +00:00
}
continue;
}
/* XXX: Flush DMA for rxpkt */
if (wx_get_rbuf(sc, rxpkt)) {
sc->wx_rxnobuf++;
wx_rxdma_map(sc, rxpkt, mb);
ifp->if_ierrors++;
rd->status = 0;
if (m0) {
m_freem(m0);
m0 = NULL;
if (sc->rpending) {
m_freem(sc->rpending);
sc->rpending = NULL;
}
Add first pass of the Intel Gigabit Ethernet (wiseman) driver. This driver seems relatively functional, but could use some souping up, particularly in the performance area. This has both NetBSD and FreeBSD attachment code and a fair amount of effort has been put into making it easy to port to different *BSD platforms. The basic design is a one tfd per mbuf transmit (with no transmit related interrupts- tfds are gc'd as needed). The receive ring uses a 2K buffer per rfd with a +2 byte adjust for the ethernet header (so the payload is aligned). There's support that *almost* works for doing large packets- the rfd chaining code works, but there's some problem with getting good checksums at the IP reassembly level (ditto for doing short tfd's too). The chip has support for TCP checksums insertion for transmit and TCP checksum calculation on receive (for both you have to do some appropriate backoff && twiddling), but this isn't in place. This is nearly entirely reverse engineered from the released Intel driver, so there's a lot of "We have to do this but do not know why" stuff. There is somebody who has the chip specs who works in FreeBSD but they're being a bit standoffish about even sharing hints which is somewhat annoying. It's also apparent that all I had to work with were the first rev boards. This driver has been lightly tested on intel && alpha, but only point-to-point. There may be some issues with switches- use of boot time environment variables that override EEPROM settings (e.g., 'set wx_ilos=1' which inverts the sense of optical signal loss) may help with this. I had this out for review for three weeks, and nobody said anything negative or positive, ergo, this checkin has no 'reviewed by' field which I would have preferred.
2000-01-04 11:12:42 +00:00
}
continue;
}
/*
* Save the completing packet's offset value and length
* and install the new one into the descriptor.
*/
lastframe = (rd->status & RDSTAT_EOP) != 0;
length = rd->length;
offset = rd->address.lowpart & 0xff;
bzero (rd, sizeof (*rd));
rd->address.lowpart = rxpkt->dma_addr + WX_RX_OFFSET_VALUE;
mb->m_len = length;
mb->m_data += offset;
mb->m_next = NULL;
if (m0 == NULL) {
m0 = mb;
tlen = length;
} else if (m0 == sc->rpending) {
/*
* Pick up where we left off before. If
* we have an offset (we're assuming the
* first frame has an offset), then we've
* lost sync somewhere along the line.
*/
if (offset) {
printf("%s: lost sync with partial packet\n",
sc->wx_name);
m_freem(sc->rpending);
sc->rpending = NULL;
m0 = mb;
tlen = length;
} else {
sc->rpending = NULL;
tlen = m0->m_pkthdr.len;
}
Add first pass of the Intel Gigabit Ethernet (wiseman) driver. This driver seems relatively functional, but could use some souping up, particularly in the performance area. This has both NetBSD and FreeBSD attachment code and a fair amount of effort has been put into making it easy to port to different *BSD platforms. The basic design is a one tfd per mbuf transmit (with no transmit related interrupts- tfds are gc'd as needed). The receive ring uses a 2K buffer per rfd with a +2 byte adjust for the ethernet header (so the payload is aligned). There's support that *almost* works for doing large packets- the rfd chaining code works, but there's some problem with getting good checksums at the IP reassembly level (ditto for doing short tfd's too). The chip has support for TCP checksums insertion for transmit and TCP checksum calculation on receive (for both you have to do some appropriate backoff && twiddling), but this isn't in place. This is nearly entirely reverse engineered from the released Intel driver, so there's a lot of "We have to do this but do not know why" stuff. There is somebody who has the chip specs who works in FreeBSD but they're being a bit standoffish about even sharing hints which is somewhat annoying. It's also apparent that all I had to work with were the first rev boards. This driver has been lightly tested on intel && alpha, but only point-to-point. There may be some issues with switches- use of boot time environment variables that override EEPROM settings (e.g., 'set wx_ilos=1' which inverts the sense of optical signal loss) may help with this. I had this out for review for three weeks, and nobody said anything negative or positive, ergo, this checkin has no 'reviewed by' field which I would have preferred.
2000-01-04 11:12:42 +00:00
} else {
tlen += length;
Add first pass of the Intel Gigabit Ethernet (wiseman) driver. This driver seems relatively functional, but could use some souping up, particularly in the performance area. This has both NetBSD and FreeBSD attachment code and a fair amount of effort has been put into making it easy to port to different *BSD platforms. The basic design is a one tfd per mbuf transmit (with no transmit related interrupts- tfds are gc'd as needed). The receive ring uses a 2K buffer per rfd with a +2 byte adjust for the ethernet header (so the payload is aligned). There's support that *almost* works for doing large packets- the rfd chaining code works, but there's some problem with getting good checksums at the IP reassembly level (ditto for doing short tfd's too). The chip has support for TCP checksums insertion for transmit and TCP checksum calculation on receive (for both you have to do some appropriate backoff && twiddling), but this isn't in place. This is nearly entirely reverse engineered from the released Intel driver, so there's a lot of "We have to do this but do not know why" stuff. There is somebody who has the chip specs who works in FreeBSD but they're being a bit standoffish about even sharing hints which is somewhat annoying. It's also apparent that all I had to work with were the first rev boards. This driver has been lightly tested on intel && alpha, but only point-to-point. There may be some issues with switches- use of boot time environment variables that override EEPROM settings (e.g., 'set wx_ilos=1' which inverts the sense of optical signal loss) may help with this. I had this out for review for three weeks, and nobody said anything negative or positive, ergo, this checkin has no 'reviewed by' field which I would have preferred.
2000-01-04 11:12:42 +00:00
}
if (sc->wx_debug) {
printf("%s: RDESC[%d] len %d off %d lastframe %d\n",
sc->wx_name, idx, mb->m_len, offset, lastframe);
#ifdef WX_EXTREME_DEBUGGING
{
int i, l = mb->m_len;
u_int8_t *p = mtod(mb, u_int8_t *);
printf("0000");
for (i = 0, l = mb->m_len; l != 0; l--, i++) {
printf(" %02x", (*(p++)) & 0xff);
if (((i + 1) & 0x7) == 0 &&
((i+1) & 0xf)) {
printf(" ");
}
if (((i + 1) & 0xf) == 0) {
printf("\n%04x", i+1);
}
}
printf("\n");
}
#endif
Add first pass of the Intel Gigabit Ethernet (wiseman) driver. This driver seems relatively functional, but could use some souping up, particularly in the performance area. This has both NetBSD and FreeBSD attachment code and a fair amount of effort has been put into making it easy to port to different *BSD platforms. The basic design is a one tfd per mbuf transmit (with no transmit related interrupts- tfds are gc'd as needed). The receive ring uses a 2K buffer per rfd with a +2 byte adjust for the ethernet header (so the payload is aligned). There's support that *almost* works for doing large packets- the rfd chaining code works, but there's some problem with getting good checksums at the IP reassembly level (ditto for doing short tfd's too). The chip has support for TCP checksums insertion for transmit and TCP checksum calculation on receive (for both you have to do some appropriate backoff && twiddling), but this isn't in place. This is nearly entirely reverse engineered from the released Intel driver, so there's a lot of "We have to do this but do not know why" stuff. There is somebody who has the chip specs who works in FreeBSD but they're being a bit standoffish about even sharing hints which is somewhat annoying. It's also apparent that all I had to work with were the first rev boards. This driver has been lightly tested on intel && alpha, but only point-to-point. There may be some issues with switches- use of boot time environment variables that override EEPROM settings (e.g., 'set wx_ilos=1' which inverts the sense of optical signal loss) may help with this. I had this out for review for three weeks, and nobody said anything negative or positive, ergo, this checkin has no 'reviewed by' field which I would have preferred.
2000-01-04 11:12:42 +00:00
}
if (m0 != mb)
m_cat(m0, mb);
Add first pass of the Intel Gigabit Ethernet (wiseman) driver. This driver seems relatively functional, but could use some souping up, particularly in the performance area. This has both NetBSD and FreeBSD attachment code and a fair amount of effort has been put into making it easy to port to different *BSD platforms. The basic design is a one tfd per mbuf transmit (with no transmit related interrupts- tfds are gc'd as needed). The receive ring uses a 2K buffer per rfd with a +2 byte adjust for the ethernet header (so the payload is aligned). There's support that *almost* works for doing large packets- the rfd chaining code works, but there's some problem with getting good checksums at the IP reassembly level (ditto for doing short tfd's too). The chip has support for TCP checksums insertion for transmit and TCP checksum calculation on receive (for both you have to do some appropriate backoff && twiddling), but this isn't in place. This is nearly entirely reverse engineered from the released Intel driver, so there's a lot of "We have to do this but do not know why" stuff. There is somebody who has the chip specs who works in FreeBSD but they're being a bit standoffish about even sharing hints which is somewhat annoying. It's also apparent that all I had to work with were the first rev boards. This driver has been lightly tested on intel && alpha, but only point-to-point. There may be some issues with switches- use of boot time environment variables that override EEPROM settings (e.g., 'set wx_ilos=1' which inverts the sense of optical signal loss) may help with this. I had this out for review for three weeks, and nobody said anything negative or positive, ergo, this checkin has no 'reviewed by' field which I would have preferred.
2000-01-04 11:12:42 +00:00
if (lastframe == 0) {
continue;
}
m0->m_pkthdr.rcvif = ifp;
m0->m_pkthdr.len = tlen - WX_CRC_LENGTH;
mb->m_len -= WX_CRC_LENGTH;
Add first pass of the Intel Gigabit Ethernet (wiseman) driver. This driver seems relatively functional, but could use some souping up, particularly in the performance area. This has both NetBSD and FreeBSD attachment code and a fair amount of effort has been put into making it easy to port to different *BSD platforms. The basic design is a one tfd per mbuf transmit (with no transmit related interrupts- tfds are gc'd as needed). The receive ring uses a 2K buffer per rfd with a +2 byte adjust for the ethernet header (so the payload is aligned). There's support that *almost* works for doing large packets- the rfd chaining code works, but there's some problem with getting good checksums at the IP reassembly level (ditto for doing short tfd's too). The chip has support for TCP checksums insertion for transmit and TCP checksum calculation on receive (for both you have to do some appropriate backoff && twiddling), but this isn't in place. This is nearly entirely reverse engineered from the released Intel driver, so there's a lot of "We have to do this but do not know why" stuff. There is somebody who has the chip specs who works in FreeBSD but they're being a bit standoffish about even sharing hints which is somewhat annoying. It's also apparent that all I had to work with were the first rev boards. This driver has been lightly tested on intel && alpha, but only point-to-point. There may be some issues with switches- use of boot time environment variables that override EEPROM settings (e.g., 'set wx_ilos=1' which inverts the sense of optical signal loss) may help with this. I had this out for review for three weeks, and nobody said anything negative or positive, ergo, this checkin has no 'reviewed by' field which I would have preferred.
2000-01-04 11:12:42 +00:00
eh = mtod(m0, struct ether_header *);
if ((ifp->if_flags & IFF_PROMISC) &&
(bcmp(eh->ether_dhost, sc->wx_enaddr, ETHER_ADDR_LEN) &&
(eh->ether_dhost[0] & 1) == 0)) {
m_freem(m0);
if (sc->rpending) {
m_freem(sc->rpending);
sc->rpending = NULL;
}
Add first pass of the Intel Gigabit Ethernet (wiseman) driver. This driver seems relatively functional, but could use some souping up, particularly in the performance area. This has both NetBSD and FreeBSD attachment code and a fair amount of effort has been put into making it easy to port to different *BSD platforms. The basic design is a one tfd per mbuf transmit (with no transmit related interrupts- tfds are gc'd as needed). The receive ring uses a 2K buffer per rfd with a +2 byte adjust for the ethernet header (so the payload is aligned). There's support that *almost* works for doing large packets- the rfd chaining code works, but there's some problem with getting good checksums at the IP reassembly level (ditto for doing short tfd's too). The chip has support for TCP checksums insertion for transmit and TCP checksum calculation on receive (for both you have to do some appropriate backoff && twiddling), but this isn't in place. This is nearly entirely reverse engineered from the released Intel driver, so there's a lot of "We have to do this but do not know why" stuff. There is somebody who has the chip specs who works in FreeBSD but they're being a bit standoffish about even sharing hints which is somewhat annoying. It's also apparent that all I had to work with were the first rev boards. This driver has been lightly tested on intel && alpha, but only point-to-point. There may be some issues with switches- use of boot time environment variables that override EEPROM settings (e.g., 'set wx_ilos=1' which inverts the sense of optical signal loss) may help with this. I had this out for review for three weeks, and nobody said anything negative or positive, ergo, this checkin has no 'reviewed by' field which I would have preferred.
2000-01-04 11:12:42 +00:00
} else {
pending[npkts++] = m0;
}
m0 = NULL;
tlen = 0;
}
if (ndesc) {
WRITE_CSR(sc, WXREG_RDT0, lidx);
sc->rnxt = idx;
Add first pass of the Intel Gigabit Ethernet (wiseman) driver. This driver seems relatively functional, but could use some souping up, particularly in the performance area. This has both NetBSD and FreeBSD attachment code and a fair amount of effort has been put into making it easy to port to different *BSD platforms. The basic design is a one tfd per mbuf transmit (with no transmit related interrupts- tfds are gc'd as needed). The receive ring uses a 2K buffer per rfd with a +2 byte adjust for the ethernet header (so the payload is aligned). There's support that *almost* works for doing large packets- the rfd chaining code works, but there's some problem with getting good checksums at the IP reassembly level (ditto for doing short tfd's too). The chip has support for TCP checksums insertion for transmit and TCP checksum calculation on receive (for both you have to do some appropriate backoff && twiddling), but this isn't in place. This is nearly entirely reverse engineered from the released Intel driver, so there's a lot of "We have to do this but do not know why" stuff. There is somebody who has the chip specs who works in FreeBSD but they're being a bit standoffish about even sharing hints which is somewhat annoying. It's also apparent that all I had to work with were the first rev boards. This driver has been lightly tested on intel && alpha, but only point-to-point. There may be some issues with switches- use of boot time environment variables that override EEPROM settings (e.g., 'set wx_ilos=1' which inverts the sense of optical signal loss) may help with this. I had this out for review for three weeks, and nobody said anything negative or positive, ergo, this checkin has no 'reviewed by' field which I would have preferred.
2000-01-04 11:12:42 +00:00
}
if (npkts) {
sc->wx_rxintr++;
}
for (idx = 0; idx < npkts; idx++) {
mb = pending[idx];
if (ifp->if_bpf) {
bpf_mtap(WX_BPFTAP_ARG(ifp), mb);
}
ifp->if_ipackets++;
if (sc->wx_debug) {
printf("%s: RECV packet length %d\n",
sc->wx_name, mb->m_pkthdr.len);
}
#ifdef __FreeBSD__
eh = mtod(mb, struct ether_header *);
m_adj(mb, sizeof (struct ether_header));
#ifdef WX_EXTREME_DEBUGGING
{
struct mbuf *mx = mb;
printf("pktlen %d\n", mb->m_pkthdr.len);
while (mx) {
int i, l = mx->m_len;
u_int8_t *p = mtod(mx, u_int8_t *);
printf("m_len %d\n", mx->m_len);
printf("0000");
for (i = 0, l = mx->m_len; l != 0; l--, i++) {
printf(" %02x", (*(p++)) & 0xff);
if (((i + 1) & 0x7) == 0 &&
((i+1) & 0xf)) {
printf(" ");
}
if (((i + 1) & 0xf) == 0) {
printf("\n%04x", i+1);
}
}
printf("\n");
mx = mx->m_next;
if (mx)
printf("\n");
}
}
#endif
Add first pass of the Intel Gigabit Ethernet (wiseman) driver. This driver seems relatively functional, but could use some souping up, particularly in the performance area. This has both NetBSD and FreeBSD attachment code and a fair amount of effort has been put into making it easy to port to different *BSD platforms. The basic design is a one tfd per mbuf transmit (with no transmit related interrupts- tfds are gc'd as needed). The receive ring uses a 2K buffer per rfd with a +2 byte adjust for the ethernet header (so the payload is aligned). There's support that *almost* works for doing large packets- the rfd chaining code works, but there's some problem with getting good checksums at the IP reassembly level (ditto for doing short tfd's too). The chip has support for TCP checksums insertion for transmit and TCP checksum calculation on receive (for both you have to do some appropriate backoff && twiddling), but this isn't in place. This is nearly entirely reverse engineered from the released Intel driver, so there's a lot of "We have to do this but do not know why" stuff. There is somebody who has the chip specs who works in FreeBSD but they're being a bit standoffish about even sharing hints which is somewhat annoying. It's also apparent that all I had to work with were the first rev boards. This driver has been lightly tested on intel && alpha, but only point-to-point. There may be some issues with switches- use of boot time environment variables that override EEPROM settings (e.g., 'set wx_ilos=1' which inverts the sense of optical signal loss) may help with this. I had this out for review for three weeks, and nobody said anything negative or positive, ergo, this checkin has no 'reviewed by' field which I would have preferred.
2000-01-04 11:12:42 +00:00
ether_input(ifp, eh, mb);
#else
(*ifp->if_input)(ifp, mb);
#endif
}
}
static void
wx_gc(sc)
wx_softc_t *sc;
{
struct ifnet *ifp = &sc->wx_if;
txpkt_t *txpkt = sc->tbsyf;
u_int32_t tdh = READ_CSR(sc, WXREG_TDH);
int s;
s = splimp();
while (txpkt != NULL) {
u_int32_t end = txpkt->eidx, cidx = tdh;
/*
* Normalize start..end indices to 2 *
* WX_MAX_TDESC range to eliminate wrap.
*/
if (txpkt->eidx < txpkt->sidx) {
end += WX_MAX_TDESC;
}
/*
* Normalize current chip index to 2 *
* WX_MAX_TDESC range to eliminate wrap.
*/
if (cidx < txpkt->sidx) {
cidx += WX_MAX_TDESC;
}
/*
* If the current chip index is between low and
* high indices for this packet, it's not finished
* transmitting yet. Because transmits are done FIFO,
* this means we're done garbage collecting too.
*/
if (txpkt->sidx <= cidx && cidx < txpkt->eidx) {
if (sc->wx_debug) {
printf("%s: TXGC %d..%d TDH %d\n", sc->wx_name,
txpkt->sidx, txpkt->eidx, tdh);
}
break;
}
ifp->if_opackets++;
if (txpkt->dptr) {
(void) m_freem(txpkt->dptr);
} else {
printf("%s: null mbuf in gc\n", sc->wx_name);
}
for (cidx = txpkt->sidx; cidx != txpkt->eidx;
cidx = T_NXT_IDX(cidx)) {
txpkt_t *tmp;
wxtd_t *td;
td = &sc->tdescriptors[cidx];
if (td->status & TXSTS_EC) {
printf("%s: excess collisions\n", sc->wx_name);
ifp->if_collisions++;
ifp->if_oerrors++;
}
if (td->status & TXSTS_LC) {
printf("%s: lost carrier\n", sc->wx_name);
ifp->if_oerrors++;
}
tmp = &sc->tbase[cidx];
if (sc->wx_debug) {
printf("%s: TXGC[%d] %p %d..%d done nact %d "
"TDH %d\n", sc->wx_name, cidx, tmp->dptr,
txpkt->sidx, txpkt->eidx, sc->tactive, tdh);
}
tmp->dptr = NULL;
if (sc->tactive == 0) {
printf("%s: nactive < 0?\n", sc->wx_name);
} else {
sc->tactive -= 1;
}
bzero(td, sizeof (*td));
}
sc->tbsyf = txpkt->next;
txpkt = sc->tbsyf;
}
if (sc->tactive < WX_MAX_TDESC) {
ifp->if_timer = 0;
ifp->if_flags &= ~IFF_OACTIVE;
}
splx(s);
}
/*
* Update packet in/out/collision statistics.
*/
static void
wx_stats_update(arg)
void *arg;
{
wx_softc_t *sc = arg;
int s;
s = splimp();
wx_gc(sc);
splx(s);
/*
* Schedule another timeout one second from now.
*/
TIMEOUT(sc, wx_stats_update, sc, hz);
}
/*
* Stop and reinitialize the hardware
*/
static void
wx_hw_stop(sc)
wx_softc_t *sc;
{
u_int32_t icr;
if (sc->revision == 2) {
wx_mwi_whackon(sc);
}
WRITE_CSR(sc, WXREG_DCR, WXDCR_RST);
DELAY(20 * 1000);
WRITE_CSR(sc, WXREG_IMASK, ~0);
icr = READ_CSR(sc, WXREG_ICR);
if (sc->revision == 2) {
wx_mwi_unwhack(sc);
}
WX_DISABLE_INT(sc);
}
static void
wx_set_addr(sc, idx, mac)
wx_softc_t *sc;
int idx;
u_int8_t *mac;
{
u_int32_t t0, t1;
t0 = (mac[0]) | (mac[1] << 8) | (mac[2] << 16) | (mac[3] << 24);
t1 = (mac[4] << 0) | (mac[5] << 8);
t1 |= WX_RAL_AV;
WRITE_CSR(sc, WXREG_RAL_LO(idx), t0);
WRITE_CSR(sc, WXREG_RAL_HI(idx), t1);
}
static int
wx_hw_initialize(sc)
wx_softc_t *sc;
{
int i;
WRITE_CSR(sc, WXREG_VET, 0);
for (i = 0; i < (WX_VLAN_TAB_SIZE << 2); i += 4) {
WRITE_CSR(sc, (WXREG_VFTA + i), 0);
}
if (sc->revision == 2) {
wx_mwi_whackon(sc);
WRITE_CSR(sc, WXREG_RCTL, WXRCTL_RST);
DELAY(5 * 1000);
}
/*
* Load the first receiver address with our MAC address,
* and load as many multicast addresses as can fit into
* the receive address array.
*/
wx_set_addr(sc, 0, sc->wx_enaddr);
for (i = 1; i <= sc->wx_nmca; i++) {
if (i >= WX_RAL_TAB_SIZE) {
break;
} else {
wx_set_addr(sc, i, sc->wx_mcaddr[i-1]);
}
}
while (i < WX_RAL_TAB_SIZE) {
WRITE_CSR(sc, WXREG_RAL_LO(i), 0);
WRITE_CSR(sc, WXREG_RAL_HI(i), 0);
i++;
}
if (sc->revision == 2) {
WRITE_CSR(sc, WXREG_RCTL, 0);
DELAY(1 * 1000);
wx_mwi_unwhack(sc);
}
/*
* Clear out the hashed multicast table array.
*/
for (i = 0; i < WX_MC_TAB_SIZE; i++) {
WRITE_CSR(sc, WXREG_MTA + (sizeof (u_int32_t) * 4), 0);
}
if (sc->wx_dcr & (WXDCR_RFCE|WXDCR_TFCE)) {
WRITE_CSR(sc, WXREG_FCAL, FC_FRM_CONST_LO);
WRITE_CSR(sc, WXREG_FCAH, FC_FRM_CONST_HI);
WRITE_CSR(sc, WXREG_FCT, FC_TYP_CONST);
} else {
WRITE_CSR(sc, WXREG_FCAL, 0);
WRITE_CSR(sc, WXREG_FCAH, 0);
WRITE_CSR(sc, WXREG_FCT, 0);
}
WRITE_CSR(sc, WXREG_FLOW_XTIMER, WX_XTIMER_DFLT);
if (sc->revision == 2) {
WRITE_CSR(sc, WXREG_FLOW_RCV_HI, 0);
WRITE_CSR(sc, WXREG_FLOW_RCV_LO, 0);
} else {
WRITE_CSR(sc, WXREG_FLOW_RCV_HI, WX_RCV_FLOW_HI_DFLT);
WRITE_CSR(sc, WXREG_FLOW_RCV_LO, WX_RCV_FLOW_LO_DFLT);
}
WRITE_CSR(sc, WXREG_XMIT_CFGW, WXTXCW_DEFAULT);
WRITE_CSR(sc, WXREG_DCR, sc->wx_dcr);
DELAY(50 * 1000);
/*
* The pin stuff is all FM from the Linux driver.
*/
if ((READ_CSR(sc, WXREG_DCR) & WXDCR_SWDPIN1) == 0) {
for (i = 0; i < 500; i++) {
DELAY(10 * 1000);
if (READ_CSR(sc, WXREG_DSR) & WXDSR_LU) {
sc->linkup = 1;
break;
}
}
} else {
printf("%s: swdpio did not clear\n", sc->wx_name);
return (-1);
}
if (sc->linkup == 0) {
printf("%s: link never came up\n", sc->wx_name);
return (-1);
}
sc->wx_ienable = WXIENABLE_DEFAULT;
return (0);
}
/*
* Stop the interface. Cancels the statistics updater and resets the interface.
*/
static void
wx_stop(sc)
wx_softc_t *sc;
{
txpkt_t *txp;
rxpkt_t *rxp;
struct ifnet *ifp = &sc->wx_if;
/*
* Cancel stats updater.
*/
UNTIMEOUT(wx_stats_update, sc, sc);
/*
* Reset the chip
*/
wx_hw_stop(sc);
/*
* Release any xmit buffers.
*/
for (txp = sc->tbase; txp && txp < &sc->tbase[WX_MAX_TDESC]; txp++) {
if (txp->dptr) {
m_free(txp->dptr);
txp->dptr = NULL;
}
}
/*
* Free all the receive buffers.
*/
for (rxp = sc->rbase; rxp && rxp < &sc->rbase[WX_MAX_RDESC]; rxp++) {
if (rxp->dptr) {
m_free(rxp->dptr);
rxp->dptr = NULL;
}
}
if (sc->rpending) {
m_freem(rxp->dptr);
sc->rpending = NULL;
}
Add first pass of the Intel Gigabit Ethernet (wiseman) driver. This driver seems relatively functional, but could use some souping up, particularly in the performance area. This has both NetBSD and FreeBSD attachment code and a fair amount of effort has been put into making it easy to port to different *BSD platforms. The basic design is a one tfd per mbuf transmit (with no transmit related interrupts- tfds are gc'd as needed). The receive ring uses a 2K buffer per rfd with a +2 byte adjust for the ethernet header (so the payload is aligned). There's support that *almost* works for doing large packets- the rfd chaining code works, but there's some problem with getting good checksums at the IP reassembly level (ditto for doing short tfd's too). The chip has support for TCP checksums insertion for transmit and TCP checksum calculation on receive (for both you have to do some appropriate backoff && twiddling), but this isn't in place. This is nearly entirely reverse engineered from the released Intel driver, so there's a lot of "We have to do this but do not know why" stuff. There is somebody who has the chip specs who works in FreeBSD but they're being a bit standoffish about even sharing hints which is somewhat annoying. It's also apparent that all I had to work with were the first rev boards. This driver has been lightly tested on intel && alpha, but only point-to-point. There may be some issues with switches- use of boot time environment variables that override EEPROM settings (e.g., 'set wx_ilos=1' which inverts the sense of optical signal loss) may help with this. I had this out for review for three weeks, and nobody said anything negative or positive, ergo, this checkin has no 'reviewed by' field which I would have preferred.
2000-01-04 11:12:42 +00:00
/*
* And we're outta here...
*/
ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
ifp->if_timer = 0;
}
/*
* Watchdog/transmission transmit timeout handler.
*/
static void
wx_watchdog(ifp)
struct ifnet *ifp;
{
wx_softc_t *sc = SOFTC_IFP(ifp);
printf("%s: device timeout\n", sc->wx_name);
ifp->if_oerrors++;
if (wx_init(sc)) {
printf("%s: could not re-init device\n", sc->wx_name);
VTIMEOUT(sc, (void (*)(void *))wx_init, sc, hz);
}
}
static int
wx_init(xsc)
void *xsc;
{
struct ifmedia *ifm;
Add first pass of the Intel Gigabit Ethernet (wiseman) driver. This driver seems relatively functional, but could use some souping up, particularly in the performance area. This has both NetBSD and FreeBSD attachment code and a fair amount of effort has been put into making it easy to port to different *BSD platforms. The basic design is a one tfd per mbuf transmit (with no transmit related interrupts- tfds are gc'd as needed). The receive ring uses a 2K buffer per rfd with a +2 byte adjust for the ethernet header (so the payload is aligned). There's support that *almost* works for doing large packets- the rfd chaining code works, but there's some problem with getting good checksums at the IP reassembly level (ditto for doing short tfd's too). The chip has support for TCP checksums insertion for transmit and TCP checksum calculation on receive (for both you have to do some appropriate backoff && twiddling), but this isn't in place. This is nearly entirely reverse engineered from the released Intel driver, so there's a lot of "We have to do this but do not know why" stuff. There is somebody who has the chip specs who works in FreeBSD but they're being a bit standoffish about even sharing hints which is somewhat annoying. It's also apparent that all I had to work with were the first rev boards. This driver has been lightly tested on intel && alpha, but only point-to-point. There may be some issues with switches- use of boot time environment variables that override EEPROM settings (e.g., 'set wx_ilos=1' which inverts the sense of optical signal loss) may help with this. I had this out for review for three weeks, and nobody said anything negative or positive, ergo, this checkin has no 'reviewed by' field which I would have preferred.
2000-01-04 11:12:42 +00:00
wx_softc_t *sc = xsc;
struct ifnet *ifp = &sc->wx_if;
rxpkt_t *rxpkt;
wxrd_t *rd;
size_t len;
int s, i;
s = splimp();
/*
* Cancel any pending I/O by resetting things.
* wx_stop will free any allocated mbufs.
*/
wx_stop(sc);
/*
* Reset the hardware. All network addresses loaded here, but
* neither the receiver nor the transmitter are enabled.
*/
if (wx_hw_initialize(sc)) {
return (EIO);
}
/*
* Set up the receive ring stuff.
*/
len = sizeof (wxrd_t) * WX_MAX_RDESC;
bzero(sc->rdescriptors, len);
for (rxpkt = sc->rbase, i = 0; rxpkt != NULL && i < WX_MAX_RDESC;
i += RXINCR, rxpkt++) {
Add first pass of the Intel Gigabit Ethernet (wiseman) driver. This driver seems relatively functional, but could use some souping up, particularly in the performance area. This has both NetBSD and FreeBSD attachment code and a fair amount of effort has been put into making it easy to port to different *BSD platforms. The basic design is a one tfd per mbuf transmit (with no transmit related interrupts- tfds are gc'd as needed). The receive ring uses a 2K buffer per rfd with a +2 byte adjust for the ethernet header (so the payload is aligned). There's support that *almost* works for doing large packets- the rfd chaining code works, but there's some problem with getting good checksums at the IP reassembly level (ditto for doing short tfd's too). The chip has support for TCP checksums insertion for transmit and TCP checksum calculation on receive (for both you have to do some appropriate backoff && twiddling), but this isn't in place. This is nearly entirely reverse engineered from the released Intel driver, so there's a lot of "We have to do this but do not know why" stuff. There is somebody who has the chip specs who works in FreeBSD but they're being a bit standoffish about even sharing hints which is somewhat annoying. It's also apparent that all I had to work with were the first rev boards. This driver has been lightly tested on intel && alpha, but only point-to-point. There may be some issues with switches- use of boot time environment variables that override EEPROM settings (e.g., 'set wx_ilos=1' which inverts the sense of optical signal loss) may help with this. I had this out for review for three weeks, and nobody said anything negative or positive, ergo, this checkin has no 'reviewed by' field which I would have preferred.
2000-01-04 11:12:42 +00:00
rd = &sc->rdescriptors[i];
if (wx_get_rbuf(sc, rxpkt)) {
break;
}
rd->address.lowpart = rxpkt->dma_addr + WX_RX_OFFSET_VALUE;
}
if (i != WX_MAX_RDESC) {
printf("%s: could not set up rbufs\n", sc->wx_name);
wx_stop(sc);
return (ENOMEM);
}
/*
* Set up transmit parameters and enable the transmitter.
*/
sc->tnxtfree = sc->tactive = 0;
sc->tbsyf = sc->tbsyl = NULL;
WRITE_CSR(sc, WXREG_TCTL, 0);
DELAY(5 * 1000);
WRITE_CSR(sc, WXREG_TDBA_LO,
vtophys((vm_offset_t)&sc->tdescriptors[0]));
WRITE_CSR(sc, WXREG_TDBA_HI, 0);
WRITE_CSR(sc, WXREG_TDLEN, WX_MAX_TDESC * sizeof (wxtd_t));
WRITE_CSR(sc, WXREG_TDH, 0);
WRITE_CSR(sc, WXREG_TDT, 0);
WRITE_CSR(sc, WXREG_TQSA_HI, 0);
WRITE_CSR(sc, WXREG_TQSA_LO, 0);
WRITE_CSR(sc, WXREG_TIPG, WX_TIPG_DFLT);
WRITE_CSR(sc, WXREG_TIDV, sc->wx_txint_delay);
WRITE_CSR(sc, WXREG_TCTL, (WXTCTL_CT(WX_COLLISION_THRESHOLD) |
WXTCTL_COLD(WX_FDX_COLLISION_DX) | WXTCTL_EN));
/*
* Set up receive parameters and enable the receiver.
*/
sc->rnxt = 0;
WRITE_CSR(sc, WXREG_RCTL, 0);
DELAY(5 * 1000);
WRITE_CSR(sc, WXREG_RDTR0, WXRDTR_FPD);
WRITE_CSR(sc, WXREG_RDBA0_LO,
vtophys((vm_offset_t)&sc->rdescriptors[0]));
WRITE_CSR(sc, WXREG_RDBA0_HI, 0);
WRITE_CSR(sc, WXREG_RDLEN0, WX_MAX_RDESC * sizeof (wxrd_t));
WRITE_CSR(sc, WXREG_RDH0, 0);
WRITE_CSR(sc, WXREG_RDT0, (WX_MAX_RDESC - RXINCR));
Add first pass of the Intel Gigabit Ethernet (wiseman) driver. This driver seems relatively functional, but could use some souping up, particularly in the performance area. This has both NetBSD and FreeBSD attachment code and a fair amount of effort has been put into making it easy to port to different *BSD platforms. The basic design is a one tfd per mbuf transmit (with no transmit related interrupts- tfds are gc'd as needed). The receive ring uses a 2K buffer per rfd with a +2 byte adjust for the ethernet header (so the payload is aligned). There's support that *almost* works for doing large packets- the rfd chaining code works, but there's some problem with getting good checksums at the IP reassembly level (ditto for doing short tfd's too). The chip has support for TCP checksums insertion for transmit and TCP checksum calculation on receive (for both you have to do some appropriate backoff && twiddling), but this isn't in place. This is nearly entirely reverse engineered from the released Intel driver, so there's a lot of "We have to do this but do not know why" stuff. There is somebody who has the chip specs who works in FreeBSD but they're being a bit standoffish about even sharing hints which is somewhat annoying. It's also apparent that all I had to work with were the first rev boards. This driver has been lightly tested on intel && alpha, but only point-to-point. There may be some issues with switches- use of boot time environment variables that override EEPROM settings (e.g., 'set wx_ilos=1' which inverts the sense of optical signal loss) may help with this. I had this out for review for three weeks, and nobody said anything negative or positive, ergo, this checkin has no 'reviewed by' field which I would have preferred.
2000-01-04 11:12:42 +00:00
WRITE_CSR(sc, WXREG_RDTR1, 0);
WRITE_CSR(sc, WXREG_RDBA1_LO, 0);
WRITE_CSR(sc, WXREG_RDBA1_HI, 0);
WRITE_CSR(sc, WXREG_RDLEN1, 0);
WRITE_CSR(sc, WXREG_RDH1, 0);
WRITE_CSR(sc, WXREG_RDT1, 0);
#if 0
WRITE_CSR(sc, WXREG_RCTL, WXRCTL_EN | WXRCTL_LPE | WXRCTL_512BRBUF |
((ifp->if_flags & IFF_BROADCAST) ? WXRCTL_BAM : 0) |
((ifp->if_flags & IFF_PROMISC) ? WXRCTL_UPE : 0) |
((sc->all_mcasts) ? WXRCTL_MPE : 0));
WRITE_CSR(sc, WXREG_RCTL, WXRCTL_EN | WXRCTL_LPE | WXRCTL_2KRBUF |
((ifp->if_flags & IFF_BROADCAST) ? WXRCTL_BAM : 0) |
((ifp->if_flags & IFF_PROMISC) ? WXRCTL_UPE : 0) |
((sc->all_mcasts) ? WXRCTL_MPE : 0));
#endif
WRITE_CSR(sc, WXREG_RCTL, WXRCTL_EN | WXRCTL_2KRBUF |
((ifp->if_flags & IFF_BROADCAST) ? WXRCTL_BAM : 0) |
((ifp->if_flags & IFF_PROMISC) ? WXRCTL_UPE : 0) |
((sc->all_mcasts) ? WXRCTL_MPE : 0));
/*
* Enable Interrupts
*/
WX_ENABLE_INT(sc);
/*
* Mark that we're up and running...
*/
ifp->if_flags |= IFF_RUNNING;
ifp->if_flags &= ~IFF_OACTIVE;
ifm = &sc->wx_media;
i = ifm->ifm_media;
ifm->ifm_media = ifm->ifm_cur->ifm_media;
wx_ifmedia_upd(ifp);
ifm->ifm_media = i;
splx(s);
/*
* Start stats updater.
*/
TIMEOUT(sc, wx_stats_update, sc, hz);
/*
* And we're outta here...
*/
return (0);
}
/*
* Get a receive buffer for our use (and dma map the data area).
*
* This chip can have buffers be 256, 512, 1024 or 2048 bytes in size.
* It wants them aligned on 256 byte boundaries, but can actually cope
* with an offset in the first 255 bytes of the head of a receive frame.
*
* We'll allocate a MCLBYTE sized cluster but *not* adjust the data pointer
* by any alignment value. Instead, we'll tell the chip to offset by any
* alignment and we'll catch the alignment on the backend at interrupt time.
*/
static void
wx_rxdma_map(sc, rxpkt, mb)
wx_softc_t *sc;
rxpkt_t *rxpkt;
struct mbuf *mb;
{
rxpkt->dptr = mb;
rxpkt->dma_addr = vtophys(mtod(mb, vm_offset_t));
}
static int
wx_get_rbuf(sc, rxpkt)
wx_softc_t *sc;
rxpkt_t *rxpkt;
{
struct mbuf *mb;
MGETHDR(mb, M_DONTWAIT, MT_DATA);
if (mb == NULL) {
rxpkt->dptr = NULL;
return (-1);
}
MCLGET(mb, M_DONTWAIT);
if ((mb->m_flags & M_EXT) == 0) {
m_freem(mb);
rxpkt->dptr = NULL;
return (-1);
}
wx_rxdma_map(sc, rxpkt, mb);
return (0);
}
static int
wx_ioctl(ifp, command, data)
struct ifnet *ifp;
IOCTL_CMD_TYPE command;
caddr_t data;
{
wx_softc_t *sc = SOFTC_IFP(ifp);
struct ifreq *ifr = (struct ifreq *) data;
int s, error = 0;
s = splimp();
switch (command) {
case SIOCSIFADDR:
#if !defined(__NetBSD__)
case SIOCGIFADDR:
#ifdef SIOCSIFMTU
case SIOCSIFMTU:
#endif
#endif
error = ether_ioctl(ifp, command, data);
break;
case SIOCSIFFLAGS:
sc->all_mcasts = (ifp->if_flags & IFF_ALLMULTI) ? 1 : 0;
/*
* If interface is marked up and not running, then start it.
* If it is marked down and running, stop it.
* If it's up then re-initialize it. This is so flags
* such as IFF_PROMISC are handled.
*/
if (ifp->if_flags & IFF_UP) {
error = wx_init(sc);
} else {
if (ifp->if_flags & IFF_RUNNING) {
wx_stop(sc);
}
}
break;
#ifdef SIOCADDMULTI
case SIOCADDMULTI:
case SIOCDELMULTI:
#if defined(__NetBSD__)
{
int all_mc_change = (sc->all_mcasts ==
((ifp->if_flags & IFF_ALLMULTI) ? 1 : 0));
error = (command == SIOCADDMULTI) ?
ether_addmulti(ifr, &sc->w.ethercom) :
ether_delmulti(ifr, &sc->w.ethercom);
if (error != ENETRESET && all_mc_change == 0) {
break;
}
}
#endif
sc->all_mcasts = (ifp->if_flags & IFF_ALLMULTI) ? 1 : 0;
error = wx_mc_setup(sc);
break;
#endif
#ifdef SIOCGIFMEDIA
case SIOCGIFMEDIA:
case SIOCSIFMEDIA:
error = ifmedia_ioctl(ifp, ifr, &sc->wx_media, command);
break;
#endif
default:
error = EINVAL;
}
(void) splx(s);
return (error);
}
static int
wx_ifmedia_upd(ifp)
struct ifnet *ifp;
{
struct wx_softc *sc = SOFTC_IFP(ifp);
struct ifmedia *ifm = &sc->wx_media;
if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
return (EINVAL);
return (0);
}
static void
wx_ifmedia_sts(ifp, ifmr)
struct ifnet *ifp;
struct ifmediareq *ifmr;
{
struct wx_softc *sc = SOFTC_IFP(ifp);
ifmr->ifm_status = IFM_AVALID;
ifmr->ifm_active = IFM_ETHER;
if (sc->linkup == 0)
return;
ifmr->ifm_status |= IFM_ACTIVE|IFM_1000_SX;
if (READ_CSR(sc, WXREG_DSR) & WXDSR_FD)
ifmr->ifm_active |= IFM_FDX;
}