freebsd-dev/sys/dev/usb/net/if_axge.c

1032 lines
25 KiB
C
Raw Normal View History

/*-
* Copyright (c) 2013-2014 Kevin Lo
* 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.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
/*
* ASIX Electronics AX88178A/AX88179 USB 2.0/3.0 gigabit ethernet driver.
*/
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/bus.h>
#include <sys/condvar.h>
#include <sys/endian.h>
#include <sys/kernel.h>
#include <sys/lock.h>
#include <sys/module.h>
#include <sys/mutex.h>
#include <sys/socket.h>
#include <sys/sysctl.h>
#include <sys/unistd.h>
#include <net/if.h>
#include <net/if_var.h>
#include <dev/usb/usb.h>
#include <dev/usb/usbdi.h>
#include <dev/usb/usbdi_util.h>
#include "usbdevs.h"
#define USB_DEBUG_VAR axge_debug
#include <dev/usb/usb_debug.h>
#include <dev/usb/usb_process.h>
#include <dev/usb/net/usb_ethernet.h>
#include <dev/usb/net/if_axgereg.h>
/*
* Various supported device vendors/products.
*/
static const STRUCT_USB_HOST_ID axge_devs[] = {
#define AXGE_DEV(v,p) { USB_VP(USB_VENDOR_##v, USB_PRODUCT_##v##_##p) }
AXGE_DEV(ASIX, AX88178A),
AXGE_DEV(ASIX, AX88179),
AXGE_DEV(DLINK, DUB1312),
AXGE_DEV(LENOVO, GIGALAN),
AXGE_DEV(SITECOMEU, LN032),
#undef AXGE_DEV
};
static const struct {
uint8_t ctrl;
uint8_t timer_l;
uint8_t timer_h;
uint8_t size;
uint8_t ifg;
} __packed axge_bulk_size[] = {
{ 7, 0x4f, 0x00, 0x12, 0xff },
{ 7, 0x20, 0x03, 0x16, 0xff },
{ 7, 0xae, 0x07, 0x18, 0xff },
{ 7, 0xcc, 0x4c, 0x18, 0x08 }
};
/* prototypes */
static device_probe_t axge_probe;
static device_attach_t axge_attach;
static device_detach_t axge_detach;
static usb_callback_t axge_bulk_read_callback;
static usb_callback_t axge_bulk_write_callback;
static miibus_readreg_t axge_miibus_readreg;
static miibus_writereg_t axge_miibus_writereg;
static miibus_statchg_t axge_miibus_statchg;
static uether_fn_t axge_attach_post;
static uether_fn_t axge_init;
static uether_fn_t axge_stop;
static uether_fn_t axge_start;
static uether_fn_t axge_tick;
Introduce axge_rxfilter() which configures RX filtering and replace axge_setmulti()/axge_setpromisc() with axge_rxfilter(). Multicast filter programming and promiscuous mode requires access to a common RX configuration register so there is no need to use separate functions with added complexity. axge_rxfilter() does not read back AXGE_RCR register since accessing a register in USB is too slow and we already have all knowledge of required configuration. Rebuilding RX filter configuration is simpler and faster than manipulating every bits after reading back the register. Note, axge_rxfilter() does not set RCR_IPE(IP header alignment on 32bit boundary) to disable extra padding bytes insertion. The extra padding wastes ethernet to USB host bandwidth as well as complicating RX handling logic. Current USB framework requires copying RX frames to mbufs so there is no need to worry about alignment. Previously axge_rx_frame() performed wrong bound check due to the extra padding and it was broken when RX checksum offloading is disabled. See added comment in axge_rx_frame () for actual RX packet layout. In axge_init(), disable WOL. It's meaningless to enable WOL in normal operation. In axge_rxeof(), use properly sized mbuf rather than blindly allocating a mbuf cluster. Use RX H/W checksum offloading only when administrator requested RX checksum offloading. Previously it always used RX H/W checksum offloading result regardless of RX checksum offloading state. Separate L4 checksum offloading validation from L3 one and properly set required offloading bits for each layer. This is to fix setting L4 checksum offloading bits for L3 packets. There are still lots of RX errors(probably RX FIFO overflows) under moderate load. Users are strongly recommended to enable ethernet flow control. Reviewed by: kevlo (initial version), hselasky
2016-08-18 06:29:07 +00:00
static uether_fn_t axge_rxfilter;
static int axge_read_mem(struct axge_softc *, uint8_t, uint16_t,
uint16_t, void *, int);
static void axge_write_mem(struct axge_softc *, uint8_t, uint16_t,
uint16_t, void *, int);
static uint8_t axge_read_cmd_1(struct axge_softc *, uint8_t, uint16_t);
static uint16_t axge_read_cmd_2(struct axge_softc *, uint8_t, uint16_t,
uint16_t);
static void axge_write_cmd_1(struct axge_softc *, uint8_t, uint16_t,
uint8_t);
static void axge_write_cmd_2(struct axge_softc *, uint8_t, uint16_t,
uint16_t, uint16_t);
static void axge_chip_init(struct axge_softc *);
static void axge_reset(struct axge_softc *);
static int axge_attach_post_sub(struct usb_ether *);
static int axge_ifmedia_upd(struct ifnet *);
static void axge_ifmedia_sts(struct ifnet *, struct ifmediareq *);
static int axge_ioctl(struct ifnet *, u_long, caddr_t);
static void axge_rx_frame(struct usb_ether *, struct usb_page_cache *, int);
static void axge_rxeof(struct usb_ether *, struct usb_page_cache *,
unsigned int, unsigned int, uint32_t);
static void axge_csum_cfg(struct usb_ether *);
#define AXGE_CSUM_FEATURES (CSUM_IP | CSUM_TCP | CSUM_UDP)
#ifdef USB_DEBUG
static int axge_debug = 0;
static SYSCTL_NODE(_hw_usb, OID_AUTO, axge, CTLFLAG_RW, 0, "USB axge");
SYSCTL_INT(_hw_usb_axge, OID_AUTO, debug, CTLFLAG_RWTUN, &axge_debug, 0,
"Debug level");
#endif
static const struct usb_config axge_config[AXGE_N_TRANSFER] = {
[AXGE_BULK_DT_WR] = {
.type = UE_BULK,
.endpoint = UE_ADDR_ANY,
.direction = UE_DIR_OUT,
.frames = AXGE_N_FRAMES,
.bufsize = AXGE_N_FRAMES * MCLBYTES,
.flags = {.pipe_bof = 1,.force_short_xfer = 1,},
.callback = axge_bulk_write_callback,
.timeout = 10000, /* 10 seconds */
},
[AXGE_BULK_DT_RD] = {
.type = UE_BULK,
.endpoint = UE_ADDR_ANY,
.direction = UE_DIR_IN,
.bufsize = 65536,
.flags = {.pipe_bof = 1,.short_xfer_ok = 1,},
.callback = axge_bulk_read_callback,
.timeout = 0, /* no timeout */
},
};
static device_method_t axge_methods[] = {
/* Device interface. */
DEVMETHOD(device_probe, axge_probe),
DEVMETHOD(device_attach, axge_attach),
DEVMETHOD(device_detach, axge_detach),
/* MII interface. */
DEVMETHOD(miibus_readreg, axge_miibus_readreg),
DEVMETHOD(miibus_writereg, axge_miibus_writereg),
DEVMETHOD(miibus_statchg, axge_miibus_statchg),
DEVMETHOD_END
};
static driver_t axge_driver = {
.name = "axge",
.methods = axge_methods,
.size = sizeof(struct axge_softc),
};
static devclass_t axge_devclass;
DRIVER_MODULE(axge, uhub, axge_driver, axge_devclass, NULL, NULL);
DRIVER_MODULE(miibus, axge, miibus_driver, miibus_devclass, NULL, NULL);
MODULE_DEPEND(axge, uether, 1, 1, 1);
MODULE_DEPEND(axge, usb, 1, 1, 1);
MODULE_DEPEND(axge, ether, 1, 1, 1);
MODULE_DEPEND(axge, miibus, 1, 1, 1);
MODULE_VERSION(axge, 1);
USB_PNP_HOST_INFO(axge_devs);
static const struct usb_ether_methods axge_ue_methods = {
.ue_attach_post = axge_attach_post,
.ue_attach_post_sub = axge_attach_post_sub,
.ue_start = axge_start,
.ue_init = axge_init,
.ue_stop = axge_stop,
.ue_tick = axge_tick,
Introduce axge_rxfilter() which configures RX filtering and replace axge_setmulti()/axge_setpromisc() with axge_rxfilter(). Multicast filter programming and promiscuous mode requires access to a common RX configuration register so there is no need to use separate functions with added complexity. axge_rxfilter() does not read back AXGE_RCR register since accessing a register in USB is too slow and we already have all knowledge of required configuration. Rebuilding RX filter configuration is simpler and faster than manipulating every bits after reading back the register. Note, axge_rxfilter() does not set RCR_IPE(IP header alignment on 32bit boundary) to disable extra padding bytes insertion. The extra padding wastes ethernet to USB host bandwidth as well as complicating RX handling logic. Current USB framework requires copying RX frames to mbufs so there is no need to worry about alignment. Previously axge_rx_frame() performed wrong bound check due to the extra padding and it was broken when RX checksum offloading is disabled. See added comment in axge_rx_frame () for actual RX packet layout. In axge_init(), disable WOL. It's meaningless to enable WOL in normal operation. In axge_rxeof(), use properly sized mbuf rather than blindly allocating a mbuf cluster. Use RX H/W checksum offloading only when administrator requested RX checksum offloading. Previously it always used RX H/W checksum offloading result regardless of RX checksum offloading state. Separate L4 checksum offloading validation from L3 one and properly set required offloading bits for each layer. This is to fix setting L4 checksum offloading bits for L3 packets. There are still lots of RX errors(probably RX FIFO overflows) under moderate load. Users are strongly recommended to enable ethernet flow control. Reviewed by: kevlo (initial version), hselasky
2016-08-18 06:29:07 +00:00
.ue_setmulti = axge_rxfilter,
.ue_setpromisc = axge_rxfilter,
.ue_mii_upd = axge_ifmedia_upd,
.ue_mii_sts = axge_ifmedia_sts,
};
static int
axge_read_mem(struct axge_softc *sc, uint8_t cmd, uint16_t index,
uint16_t val, void *buf, int len)
{
struct usb_device_request req;
AXGE_LOCK_ASSERT(sc, MA_OWNED);
req.bmRequestType = UT_READ_VENDOR_DEVICE;
req.bRequest = cmd;
USETW(req.wValue, val);
USETW(req.wIndex, index);
USETW(req.wLength, len);
return (uether_do_request(&sc->sc_ue, &req, buf, 1000));
}
static void
axge_write_mem(struct axge_softc *sc, uint8_t cmd, uint16_t index,
uint16_t val, void *buf, int len)
{
struct usb_device_request req;
AXGE_LOCK_ASSERT(sc, MA_OWNED);
req.bmRequestType = UT_WRITE_VENDOR_DEVICE;
req.bRequest = cmd;
USETW(req.wValue, val);
USETW(req.wIndex, index);
USETW(req.wLength, len);
if (uether_do_request(&sc->sc_ue, &req, buf, 1000)) {
/* Error ignored. */
}
}
static uint8_t
axge_read_cmd_1(struct axge_softc *sc, uint8_t cmd, uint16_t reg)
{
uint8_t val;
axge_read_mem(sc, cmd, 1, reg, &val, 1);
return (val);
}
static uint16_t
axge_read_cmd_2(struct axge_softc *sc, uint8_t cmd, uint16_t index,
uint16_t reg)
{
uint8_t val[2];
axge_read_mem(sc, cmd, index, reg, &val, 2);
return (UGETW(val));
}
static void
axge_write_cmd_1(struct axge_softc *sc, uint8_t cmd, uint16_t reg, uint8_t val)
{
axge_write_mem(sc, cmd, 1, reg, &val, 1);
}
static void
axge_write_cmd_2(struct axge_softc *sc, uint8_t cmd, uint16_t index,
uint16_t reg, uint16_t val)
{
uint8_t temp[2];
USETW(temp, val);
axge_write_mem(sc, cmd, index, reg, &temp, 2);
}
static int
axge_miibus_readreg(device_t dev, int phy, int reg)
{
struct axge_softc *sc;
uint16_t val;
int locked;
sc = device_get_softc(dev);
locked = mtx_owned(&sc->sc_mtx);
if (!locked)
AXGE_LOCK(sc);
val = axge_read_cmd_2(sc, AXGE_ACCESS_PHY, reg, phy);
if (!locked)
AXGE_UNLOCK(sc);
return (val);
}
static int
axge_miibus_writereg(device_t dev, int phy, int reg, int val)
{
struct axge_softc *sc;
int locked;
sc = device_get_softc(dev);
locked = mtx_owned(&sc->sc_mtx);
if (!locked)
AXGE_LOCK(sc);
axge_write_cmd_2(sc, AXGE_ACCESS_PHY, reg, phy, val);
if (!locked)
AXGE_UNLOCK(sc);
return (0);
}
static void
axge_miibus_statchg(device_t dev)
{
struct axge_softc *sc;
struct mii_data *mii;
struct ifnet *ifp;
uint8_t link_status, tmp[5];
uint16_t val;
int locked;
sc = device_get_softc(dev);
mii = GET_MII(sc);
locked = mtx_owned(&sc->sc_mtx);
if (!locked)
AXGE_LOCK(sc);
ifp = uether_getifp(&sc->sc_ue);
if (mii == NULL || ifp == NULL ||
(ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
goto done;
sc->sc_flags &= ~AXGE_FLAG_LINK;
if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
(IFM_ACTIVE | IFM_AVALID)) {
switch (IFM_SUBTYPE(mii->mii_media_active)) {
case IFM_10_T:
case IFM_100_TX:
case IFM_1000_T:
sc->sc_flags |= AXGE_FLAG_LINK;
break;
default:
break;
}
}
/* Lost link, do nothing. */
if ((sc->sc_flags & AXGE_FLAG_LINK) == 0)
goto done;
link_status = axge_read_cmd_1(sc, AXGE_ACCESS_MAC, AXGE_PLSR);
val = 0;
if ((IFM_OPTIONS(mii->mii_media_active) & IFM_FDX) != 0) {
val |= MSR_FD;
if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_TXPAUSE) != 0)
val |= MSR_TFC;
if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_RXPAUSE) != 0)
val |= MSR_RFC;
}
val |= MSR_RE;
switch (IFM_SUBTYPE(mii->mii_media_active)) {
case IFM_1000_T:
val |= MSR_GM | MSR_EN_125MHZ;
if (link_status & PLSR_USB_SS)
memcpy(tmp, &axge_bulk_size[0], 5);
else if (link_status & PLSR_USB_HS)
memcpy(tmp, &axge_bulk_size[1], 5);
else
memcpy(tmp, &axge_bulk_size[3], 5);
break;
case IFM_100_TX:
val |= MSR_PS;
if (link_status & (PLSR_USB_SS | PLSR_USB_HS))
memcpy(tmp, &axge_bulk_size[2], 5);
else
memcpy(tmp, &axge_bulk_size[3], 5);
break;
case IFM_10_T:
memcpy(tmp, &axge_bulk_size[3], 5);
break;
}
/* Rx bulk configuration. */
axge_write_mem(sc, AXGE_ACCESS_MAC, 5, AXGE_RX_BULKIN_QCTRL, tmp, 5);
axge_write_cmd_2(sc, AXGE_ACCESS_MAC, 2, AXGE_MSR, val);
done:
if (!locked)
AXGE_UNLOCK(sc);
}
static void
axge_chip_init(struct axge_softc *sc)
{
/* Power up ethernet PHY. */
axge_write_cmd_2(sc, AXGE_ACCESS_MAC, 2, AXGE_EPPRCR, 0);
axge_write_cmd_2(sc, AXGE_ACCESS_MAC, 2, AXGE_EPPRCR, EPPRCR_IPRL);
uether_pause(&sc->sc_ue, hz / 4);
axge_write_cmd_1(sc, AXGE_ACCESS_MAC, AXGE_CLK_SELECT,
AXGE_CLK_SELECT_ACS | AXGE_CLK_SELECT_BCS);
uether_pause(&sc->sc_ue, hz / 10);
}
static void
axge_reset(struct axge_softc *sc)
{
struct usb_config_descriptor *cd;
usb_error_t err;
cd = usbd_get_config_descriptor(sc->sc_ue.ue_udev);
err = usbd_req_set_config(sc->sc_ue.ue_udev, &sc->sc_mtx,
cd->bConfigurationValue);
if (err)
DPRINTF("reset failed (ignored)\n");
/* Wait a little while for the chip to get its brains in order. */
uether_pause(&sc->sc_ue, hz / 100);
/* Reinitialize controller to achieve full reset. */
axge_chip_init(sc);
}
static void
axge_attach_post(struct usb_ether *ue)
{
struct axge_softc *sc;
sc = uether_getsc(ue);
/* Initialize controller and get station address. */
axge_chip_init(sc);
axge_read_mem(sc, AXGE_ACCESS_MAC, ETHER_ADDR_LEN, AXGE_NIDR,
ue->ue_eaddr, ETHER_ADDR_LEN);
}
static int
axge_attach_post_sub(struct usb_ether *ue)
{
struct axge_softc *sc;
struct ifnet *ifp;
int error;
sc = uether_getsc(ue);
ifp = ue->ue_ifp;
ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
ifp->if_start = uether_start;
ifp->if_ioctl = axge_ioctl;
ifp->if_init = uether_init;
IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
ifp->if_snd.ifq_drv_maxlen = ifqmaxlen;
IFQ_SET_READY(&ifp->if_snd);
ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_TXCSUM | IFCAP_RXCSUM;
ifp->if_hwassist = AXGE_CSUM_FEATURES;
ifp->if_capenable = ifp->if_capabilities;
mtx_lock(&Giant);
error = mii_attach(ue->ue_dev, &ue->ue_miibus, ifp,
uether_ifmedia_upd, ue->ue_methods->ue_mii_sts,
BMSR_DEFCAPMASK, AXGE_PHY_ADDR, MII_OFFSET_ANY, MIIF_DOPAUSE);
mtx_unlock(&Giant);
return (error);
}
/*
* Set media options.
*/
static int
axge_ifmedia_upd(struct ifnet *ifp)
{
struct axge_softc *sc;
struct mii_data *mii;
struct mii_softc *miisc;
int error;
sc = ifp->if_softc;
mii = GET_MII(sc);
AXGE_LOCK_ASSERT(sc, MA_OWNED);
LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
PHY_RESET(miisc);
error = mii_mediachg(mii);
return (error);
}
/*
* Report current media status.
*/
static void
axge_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
{
struct axge_softc *sc;
struct mii_data *mii;
sc = ifp->if_softc;
mii = GET_MII(sc);
AXGE_LOCK(sc);
mii_pollstat(mii);
ifmr->ifm_active = mii->mii_media_active;
ifmr->ifm_status = mii->mii_media_status;
AXGE_UNLOCK(sc);
}
/*
* Probe for a AX88179 chip.
*/
static int
axge_probe(device_t dev)
{
struct usb_attach_arg *uaa;
uaa = device_get_ivars(dev);
if (uaa->usb_mode != USB_MODE_HOST)
return (ENXIO);
if (uaa->info.bConfigIndex != AXGE_CONFIG_IDX)
return (ENXIO);
if (uaa->info.bIfaceIndex != AXGE_IFACE_IDX)
return (ENXIO);
return (usbd_lookup_id_by_uaa(axge_devs, sizeof(axge_devs), uaa));
}
/*
* Attach the interface. Allocate softc structures, do ifmedia
* setup and ethernet/BPF attach.
*/
static int
axge_attach(device_t dev)
{
struct usb_attach_arg *uaa;
struct axge_softc *sc;
struct usb_ether *ue;
uint8_t iface_index;
int error;
uaa = device_get_ivars(dev);
sc = device_get_softc(dev);
ue = &sc->sc_ue;
device_set_usb_desc(dev);
mtx_init(&sc->sc_mtx, device_get_nameunit(dev), NULL, MTX_DEF);
iface_index = AXGE_IFACE_IDX;
error = usbd_transfer_setup(uaa->device, &iface_index,
sc->sc_xfer, axge_config, AXGE_N_TRANSFER, sc, &sc->sc_mtx);
if (error) {
device_printf(dev, "allocating USB transfers failed\n");
mtx_destroy(&sc->sc_mtx);
return (ENXIO);
}
ue->ue_sc = sc;
ue->ue_dev = dev;
ue->ue_udev = uaa->device;
ue->ue_mtx = &sc->sc_mtx;
ue->ue_methods = &axge_ue_methods;
error = uether_ifattach(ue);
if (error) {
device_printf(dev, "could not attach interface\n");
goto detach;
}
return (0); /* success */
detach:
axge_detach(dev);
return (ENXIO); /* failure */
}
static int
axge_detach(device_t dev)
{
struct axge_softc *sc;
struct usb_ether *ue;
sc = device_get_softc(dev);
ue = &sc->sc_ue;
usbd_transfer_unsetup(sc->sc_xfer, AXGE_N_TRANSFER);
uether_ifdetach(ue);
mtx_destroy(&sc->sc_mtx);
return (0);
}
static void
axge_bulk_read_callback(struct usb_xfer *xfer, usb_error_t error)
{
struct axge_softc *sc;
struct usb_ether *ue;
struct usb_page_cache *pc;
int actlen;
sc = usbd_xfer_softc(xfer);
ue = &sc->sc_ue;
usbd_xfer_status(xfer, &actlen, NULL, NULL, NULL);
switch (USB_GET_STATE(xfer)) {
case USB_ST_TRANSFERRED:
pc = usbd_xfer_get_frame(xfer, 0);
axge_rx_frame(ue, pc, actlen);
/* FALLTHROUGH */
case USB_ST_SETUP:
tr_setup:
usbd_xfer_set_frame_len(xfer, 0, usbd_xfer_max_len(xfer));
usbd_transfer_submit(xfer);
uether_rxflush(ue);
break;
default:
if (error != USB_ERR_CANCELLED) {
usbd_xfer_set_stall(xfer);
goto tr_setup;
}
break;
}
}
static void
axge_bulk_write_callback(struct usb_xfer *xfer, usb_error_t error)
{
struct axge_softc *sc;
struct ifnet *ifp;
struct usb_page_cache *pc;
struct mbuf *m;
struct axge_frame_txhdr txhdr;
int nframes, pos;
sc = usbd_xfer_softc(xfer);
ifp = uether_getifp(&sc->sc_ue);
switch (USB_GET_STATE(xfer)) {
case USB_ST_TRANSFERRED:
ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
/* FALLTHROUGH */
case USB_ST_SETUP:
tr_setup:
if ((sc->sc_flags & AXGE_FLAG_LINK) == 0 ||
(ifp->if_drv_flags & IFF_DRV_OACTIVE) != 0) {
/*
* Don't send anything if there is no link or
* controller is busy.
*/
return;
}
for (nframes = 0; nframes < AXGE_N_FRAMES &&
!IFQ_DRV_IS_EMPTY(&ifp->if_snd); nframes++) {
IFQ_DRV_DEQUEUE(&ifp->if_snd, m);
if (m == NULL)
break;
usbd_xfer_set_frame_offset(xfer, nframes * MCLBYTES,
nframes);
pc = usbd_xfer_get_frame(xfer, nframes);
txhdr.mss = 0;
txhdr.len = htole32(AXGE_TXBYTES(m->m_pkthdr.len));
if ((ifp->if_capenable & IFCAP_TXCSUM) != 0 &&
(m->m_pkthdr.csum_flags & AXGE_CSUM_FEATURES) == 0)
txhdr.len |= htole32(AXGE_CSUM_DISABLE);
pos = 0;
usbd_copy_in(pc, pos, &txhdr, sizeof(txhdr));
pos += sizeof(txhdr);
usbd_m_copy_in(pc, pos, m, 0, m->m_pkthdr.len);
pos += m->m_pkthdr.len;
/*
* if there's a BPF listener, bounce a copy
* of this frame to him:
*/
BPF_MTAP(ifp, m);
m_freem(m);
/* Set frame length. */
usbd_xfer_set_frame_len(xfer, nframes, pos);
}
if (nframes != 0) {
/*
* XXX
* Update TX packet counter here. This is not
* correct way but it seems that there is no way
* to know how many packets are sent at the end
* of transfer because controller combines
* multiple writes into single one if there is
* room in TX buffer of controller.
*/
if_inc_counter(ifp, IFCOUNTER_OPACKETS, nframes);
usbd_xfer_set_frames(xfer, nframes);
usbd_transfer_submit(xfer);
ifp->if_drv_flags |= IFF_DRV_OACTIVE;
}
return;
/* NOTREACHED */
default:
if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
if (error != USB_ERR_CANCELLED) {
usbd_xfer_set_stall(xfer);
goto tr_setup;
}
return;
}
}
static void
axge_tick(struct usb_ether *ue)
{
struct axge_softc *sc;
struct mii_data *mii;
sc = uether_getsc(ue);
mii = GET_MII(sc);
AXGE_LOCK_ASSERT(sc, MA_OWNED);
mii_tick(mii);
}
static void
Introduce axge_rxfilter() which configures RX filtering and replace axge_setmulti()/axge_setpromisc() with axge_rxfilter(). Multicast filter programming and promiscuous mode requires access to a common RX configuration register so there is no need to use separate functions with added complexity. axge_rxfilter() does not read back AXGE_RCR register since accessing a register in USB is too slow and we already have all knowledge of required configuration. Rebuilding RX filter configuration is simpler and faster than manipulating every bits after reading back the register. Note, axge_rxfilter() does not set RCR_IPE(IP header alignment on 32bit boundary) to disable extra padding bytes insertion. The extra padding wastes ethernet to USB host bandwidth as well as complicating RX handling logic. Current USB framework requires copying RX frames to mbufs so there is no need to worry about alignment. Previously axge_rx_frame() performed wrong bound check due to the extra padding and it was broken when RX checksum offloading is disabled. See added comment in axge_rx_frame () for actual RX packet layout. In axge_init(), disable WOL. It's meaningless to enable WOL in normal operation. In axge_rxeof(), use properly sized mbuf rather than blindly allocating a mbuf cluster. Use RX H/W checksum offloading only when administrator requested RX checksum offloading. Previously it always used RX H/W checksum offloading result regardless of RX checksum offloading state. Separate L4 checksum offloading validation from L3 one and properly set required offloading bits for each layer. This is to fix setting L4 checksum offloading bits for L3 packets. There are still lots of RX errors(probably RX FIFO overflows) under moderate load. Users are strongly recommended to enable ethernet flow control. Reviewed by: kevlo (initial version), hselasky
2016-08-18 06:29:07 +00:00
axge_rxfilter(struct usb_ether *ue)
{
struct axge_softc *sc;
struct ifnet *ifp;
struct ifmultiaddr *ifma;
uint32_t h;
uint16_t rxmode;
uint8_t hashtbl[8] = { 0, 0, 0, 0, 0, 0, 0, 0 };
sc = uether_getsc(ue);
ifp = uether_getifp(ue);
h = 0;
AXGE_LOCK_ASSERT(sc, MA_OWNED);
Introduce axge_rxfilter() which configures RX filtering and replace axge_setmulti()/axge_setpromisc() with axge_rxfilter(). Multicast filter programming and promiscuous mode requires access to a common RX configuration register so there is no need to use separate functions with added complexity. axge_rxfilter() does not read back AXGE_RCR register since accessing a register in USB is too slow and we already have all knowledge of required configuration. Rebuilding RX filter configuration is simpler and faster than manipulating every bits after reading back the register. Note, axge_rxfilter() does not set RCR_IPE(IP header alignment on 32bit boundary) to disable extra padding bytes insertion. The extra padding wastes ethernet to USB host bandwidth as well as complicating RX handling logic. Current USB framework requires copying RX frames to mbufs so there is no need to worry about alignment. Previously axge_rx_frame() performed wrong bound check due to the extra padding and it was broken when RX checksum offloading is disabled. See added comment in axge_rx_frame () for actual RX packet layout. In axge_init(), disable WOL. It's meaningless to enable WOL in normal operation. In axge_rxeof(), use properly sized mbuf rather than blindly allocating a mbuf cluster. Use RX H/W checksum offloading only when administrator requested RX checksum offloading. Previously it always used RX H/W checksum offloading result regardless of RX checksum offloading state. Separate L4 checksum offloading validation from L3 one and properly set required offloading bits for each layer. This is to fix setting L4 checksum offloading bits for L3 packets. There are still lots of RX errors(probably RX FIFO overflows) under moderate load. Users are strongly recommended to enable ethernet flow control. Reviewed by: kevlo (initial version), hselasky
2016-08-18 06:29:07 +00:00
/*
* Configure RX settings.
* Don't set RCR_IPE(IP header alignment on 32bit boundary) to disable
* inserting extra padding bytes. This wastes ethernet to USB host
* bandwidth as well as complicating RX handling logic. Current USB
* framework requires copying RX frames to mbufs so there is no need
* to worry about alignment.
*/
rxmode = RCR_DROP_CRCERR | RCR_START;
if (ifp->if_flags & IFF_BROADCAST)
rxmode |= RCR_ACPT_BCAST;
if (ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) {
Introduce axge_rxfilter() which configures RX filtering and replace axge_setmulti()/axge_setpromisc() with axge_rxfilter(). Multicast filter programming and promiscuous mode requires access to a common RX configuration register so there is no need to use separate functions with added complexity. axge_rxfilter() does not read back AXGE_RCR register since accessing a register in USB is too slow and we already have all knowledge of required configuration. Rebuilding RX filter configuration is simpler and faster than manipulating every bits after reading back the register. Note, axge_rxfilter() does not set RCR_IPE(IP header alignment on 32bit boundary) to disable extra padding bytes insertion. The extra padding wastes ethernet to USB host bandwidth as well as complicating RX handling logic. Current USB framework requires copying RX frames to mbufs so there is no need to worry about alignment. Previously axge_rx_frame() performed wrong bound check due to the extra padding and it was broken when RX checksum offloading is disabled. See added comment in axge_rx_frame () for actual RX packet layout. In axge_init(), disable WOL. It's meaningless to enable WOL in normal operation. In axge_rxeof(), use properly sized mbuf rather than blindly allocating a mbuf cluster. Use RX H/W checksum offloading only when administrator requested RX checksum offloading. Previously it always used RX H/W checksum offloading result regardless of RX checksum offloading state. Separate L4 checksum offloading validation from L3 one and properly set required offloading bits for each layer. This is to fix setting L4 checksum offloading bits for L3 packets. There are still lots of RX errors(probably RX FIFO overflows) under moderate load. Users are strongly recommended to enable ethernet flow control. Reviewed by: kevlo (initial version), hselasky
2016-08-18 06:29:07 +00:00
if (ifp->if_flags & IFF_PROMISC)
rxmode |= RCR_PROMISC;
rxmode |= RCR_ACPT_ALL_MCAST;
axge_write_cmd_2(sc, AXGE_ACCESS_MAC, 2, AXGE_RCR, rxmode);
return;
}
Introduce axge_rxfilter() which configures RX filtering and replace axge_setmulti()/axge_setpromisc() with axge_rxfilter(). Multicast filter programming and promiscuous mode requires access to a common RX configuration register so there is no need to use separate functions with added complexity. axge_rxfilter() does not read back AXGE_RCR register since accessing a register in USB is too slow and we already have all knowledge of required configuration. Rebuilding RX filter configuration is simpler and faster than manipulating every bits after reading back the register. Note, axge_rxfilter() does not set RCR_IPE(IP header alignment on 32bit boundary) to disable extra padding bytes insertion. The extra padding wastes ethernet to USB host bandwidth as well as complicating RX handling logic. Current USB framework requires copying RX frames to mbufs so there is no need to worry about alignment. Previously axge_rx_frame() performed wrong bound check due to the extra padding and it was broken when RX checksum offloading is disabled. See added comment in axge_rx_frame () for actual RX packet layout. In axge_init(), disable WOL. It's meaningless to enable WOL in normal operation. In axge_rxeof(), use properly sized mbuf rather than blindly allocating a mbuf cluster. Use RX H/W checksum offloading only when administrator requested RX checksum offloading. Previously it always used RX H/W checksum offloading result regardless of RX checksum offloading state. Separate L4 checksum offloading validation from L3 one and properly set required offloading bits for each layer. This is to fix setting L4 checksum offloading bits for L3 packets. There are still lots of RX errors(probably RX FIFO overflows) under moderate load. Users are strongly recommended to enable ethernet flow control. Reviewed by: kevlo (initial version), hselasky
2016-08-18 06:29:07 +00:00
rxmode |= RCR_ACPT_MCAST;
if_maddr_rlock(ifp);
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;
hashtbl[h / 8] |= 1 << (h % 8);
}
if_maddr_runlock(ifp);
axge_write_mem(sc, AXGE_ACCESS_MAC, 8, AXGE_MFA, (void *)&hashtbl, 8);
axge_write_cmd_2(sc, AXGE_ACCESS_MAC, 2, AXGE_RCR, rxmode);
}
static void
axge_start(struct usb_ether *ue)
{
struct axge_softc *sc;
sc = uether_getsc(ue);
/*
* Start the USB transfers, if not already started.
*/
usbd_transfer_start(sc->sc_xfer[AXGE_BULK_DT_RD]);
usbd_transfer_start(sc->sc_xfer[AXGE_BULK_DT_WR]);
}
static void
axge_init(struct usb_ether *ue)
{
struct axge_softc *sc;
struct ifnet *ifp;
sc = uether_getsc(ue);
ifp = uether_getifp(ue);
AXGE_LOCK_ASSERT(sc, MA_OWNED);
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
return;
/*
* Cancel pending I/O and free all RX/TX buffers.
*/
axge_stop(ue);
axge_reset(sc);
/* Set MAC address. */
axge_write_mem(sc, AXGE_ACCESS_MAC, ETHER_ADDR_LEN, AXGE_NIDR,
IF_LLADDR(ifp), ETHER_ADDR_LEN);
axge_write_cmd_1(sc, AXGE_ACCESS_MAC, AXGE_PWLLR, 0x34);
axge_write_cmd_1(sc, AXGE_ACCESS_MAC, AXGE_PWLHR, 0x52);
/* Configure TX/RX checksum offloading. */
axge_csum_cfg(ue);
Introduce axge_rxfilter() which configures RX filtering and replace axge_setmulti()/axge_setpromisc() with axge_rxfilter(). Multicast filter programming and promiscuous mode requires access to a common RX configuration register so there is no need to use separate functions with added complexity. axge_rxfilter() does not read back AXGE_RCR register since accessing a register in USB is too slow and we already have all knowledge of required configuration. Rebuilding RX filter configuration is simpler and faster than manipulating every bits after reading back the register. Note, axge_rxfilter() does not set RCR_IPE(IP header alignment on 32bit boundary) to disable extra padding bytes insertion. The extra padding wastes ethernet to USB host bandwidth as well as complicating RX handling logic. Current USB framework requires copying RX frames to mbufs so there is no need to worry about alignment. Previously axge_rx_frame() performed wrong bound check due to the extra padding and it was broken when RX checksum offloading is disabled. See added comment in axge_rx_frame () for actual RX packet layout. In axge_init(), disable WOL. It's meaningless to enable WOL in normal operation. In axge_rxeof(), use properly sized mbuf rather than blindly allocating a mbuf cluster. Use RX H/W checksum offloading only when administrator requested RX checksum offloading. Previously it always used RX H/W checksum offloading result regardless of RX checksum offloading state. Separate L4 checksum offloading validation from L3 one and properly set required offloading bits for each layer. This is to fix setting L4 checksum offloading bits for L3 packets. There are still lots of RX errors(probably RX FIFO overflows) under moderate load. Users are strongly recommended to enable ethernet flow control. Reviewed by: kevlo (initial version), hselasky
2016-08-18 06:29:07 +00:00
/* Configure RX filters. */
axge_rxfilter(ue);
Introduce axge_rxfilter() which configures RX filtering and replace axge_setmulti()/axge_setpromisc() with axge_rxfilter(). Multicast filter programming and promiscuous mode requires access to a common RX configuration register so there is no need to use separate functions with added complexity. axge_rxfilter() does not read back AXGE_RCR register since accessing a register in USB is too slow and we already have all knowledge of required configuration. Rebuilding RX filter configuration is simpler and faster than manipulating every bits after reading back the register. Note, axge_rxfilter() does not set RCR_IPE(IP header alignment on 32bit boundary) to disable extra padding bytes insertion. The extra padding wastes ethernet to USB host bandwidth as well as complicating RX handling logic. Current USB framework requires copying RX frames to mbufs so there is no need to worry about alignment. Previously axge_rx_frame() performed wrong bound check due to the extra padding and it was broken when RX checksum offloading is disabled. See added comment in axge_rx_frame () for actual RX packet layout. In axge_init(), disable WOL. It's meaningless to enable WOL in normal operation. In axge_rxeof(), use properly sized mbuf rather than blindly allocating a mbuf cluster. Use RX H/W checksum offloading only when administrator requested RX checksum offloading. Previously it always used RX H/W checksum offloading result regardless of RX checksum offloading state. Separate L4 checksum offloading validation from L3 one and properly set required offloading bits for each layer. This is to fix setting L4 checksum offloading bits for L3 packets. There are still lots of RX errors(probably RX FIFO overflows) under moderate load. Users are strongly recommended to enable ethernet flow control. Reviewed by: kevlo (initial version), hselasky
2016-08-18 06:29:07 +00:00
/*
* XXX
* Controller supports wakeup on link change detection,
* magic packet and wakeup frame recpetion. But it seems
* there is no framework for USB ethernet suspend/wakeup.
* Disable all wakeup functions.
*/
axge_write_cmd_1(sc, AXGE_ACCESS_MAC, AXGE_MMSR, 0);
(void)axge_read_cmd_1(sc, AXGE_ACCESS_MAC, AXGE_MMSR);
Introduce axge_rxfilter() which configures RX filtering and replace axge_setmulti()/axge_setpromisc() with axge_rxfilter(). Multicast filter programming and promiscuous mode requires access to a common RX configuration register so there is no need to use separate functions with added complexity. axge_rxfilter() does not read back AXGE_RCR register since accessing a register in USB is too slow and we already have all knowledge of required configuration. Rebuilding RX filter configuration is simpler and faster than manipulating every bits after reading back the register. Note, axge_rxfilter() does not set RCR_IPE(IP header alignment on 32bit boundary) to disable extra padding bytes insertion. The extra padding wastes ethernet to USB host bandwidth as well as complicating RX handling logic. Current USB framework requires copying RX frames to mbufs so there is no need to worry about alignment. Previously axge_rx_frame() performed wrong bound check due to the extra padding and it was broken when RX checksum offloading is disabled. See added comment in axge_rx_frame () for actual RX packet layout. In axge_init(), disable WOL. It's meaningless to enable WOL in normal operation. In axge_rxeof(), use properly sized mbuf rather than blindly allocating a mbuf cluster. Use RX H/W checksum offloading only when administrator requested RX checksum offloading. Previously it always used RX H/W checksum offloading result regardless of RX checksum offloading state. Separate L4 checksum offloading validation from L3 one and properly set required offloading bits for each layer. This is to fix setting L4 checksum offloading bits for L3 packets. There are still lots of RX errors(probably RX FIFO overflows) under moderate load. Users are strongly recommended to enable ethernet flow control. Reviewed by: kevlo (initial version), hselasky
2016-08-18 06:29:07 +00:00
/* Configure default medium type. */
axge_write_cmd_2(sc, AXGE_ACCESS_MAC, 2, AXGE_MSR, MSR_GM | MSR_FD |
MSR_RFC | MSR_TFC | MSR_RE);
usbd_xfer_set_stall(sc->sc_xfer[AXGE_BULK_DT_WR]);
ifp->if_drv_flags |= IFF_DRV_RUNNING;
/* Switch to selected media. */
axge_ifmedia_upd(ifp);
}
static void
axge_stop(struct usb_ether *ue)
{
struct axge_softc *sc;
struct ifnet *ifp;
sc = uether_getsc(ue);
ifp = uether_getifp(ue);
AXGE_LOCK_ASSERT(sc, MA_OWNED);
ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
sc->sc_flags &= ~AXGE_FLAG_LINK;
/*
* Stop all the transfers, if not already stopped:
*/
usbd_transfer_stop(sc->sc_xfer[AXGE_BULK_DT_WR]);
usbd_transfer_stop(sc->sc_xfer[AXGE_BULK_DT_RD]);
}
static int
axge_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
{
struct usb_ether *ue;
struct axge_softc *sc;
struct ifreq *ifr;
int error, mask, reinit;
ue = ifp->if_softc;
sc = uether_getsc(ue);
ifr = (struct ifreq *)data;
error = 0;
reinit = 0;
if (cmd == SIOCSIFCAP) {
AXGE_LOCK(sc);
mask = ifr->ifr_reqcap ^ ifp->if_capenable;
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 |= AXGE_CSUM_FEATURES;
else
ifp->if_hwassist &= ~AXGE_CSUM_FEATURES;
reinit++;
}
if ((mask & IFCAP_RXCSUM) != 0 &&
(ifp->if_capabilities & IFCAP_RXCSUM) != 0) {
ifp->if_capenable ^= IFCAP_RXCSUM;
reinit++;
}
if (reinit > 0 && ifp->if_drv_flags & IFF_DRV_RUNNING)
ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
else
reinit = 0;
AXGE_UNLOCK(sc);
if (reinit > 0)
uether_init(ue);
} else
error = uether_ioctl(ifp, cmd, data);
return (error);
}
static void
axge_rx_frame(struct usb_ether *ue, struct usb_page_cache *pc, int actlen)
{
Introduce axge_rxfilter() which configures RX filtering and replace axge_setmulti()/axge_setpromisc() with axge_rxfilter(). Multicast filter programming and promiscuous mode requires access to a common RX configuration register so there is no need to use separate functions with added complexity. axge_rxfilter() does not read back AXGE_RCR register since accessing a register in USB is too slow and we already have all knowledge of required configuration. Rebuilding RX filter configuration is simpler and faster than manipulating every bits after reading back the register. Note, axge_rxfilter() does not set RCR_IPE(IP header alignment on 32bit boundary) to disable extra padding bytes insertion. The extra padding wastes ethernet to USB host bandwidth as well as complicating RX handling logic. Current USB framework requires copying RX frames to mbufs so there is no need to worry about alignment. Previously axge_rx_frame() performed wrong bound check due to the extra padding and it was broken when RX checksum offloading is disabled. See added comment in axge_rx_frame () for actual RX packet layout. In axge_init(), disable WOL. It's meaningless to enable WOL in normal operation. In axge_rxeof(), use properly sized mbuf rather than blindly allocating a mbuf cluster. Use RX H/W checksum offloading only when administrator requested RX checksum offloading. Previously it always used RX H/W checksum offloading result regardless of RX checksum offloading state. Separate L4 checksum offloading validation from L3 one and properly set required offloading bits for each layer. This is to fix setting L4 checksum offloading bits for L3 packets. There are still lots of RX errors(probably RX FIFO overflows) under moderate load. Users are strongly recommended to enable ethernet flow control. Reviewed by: kevlo (initial version), hselasky
2016-08-18 06:29:07 +00:00
struct axge_frame_rxhdr pkt_hdr;
uint32_t rxhdr;
Introduce axge_rxfilter() which configures RX filtering and replace axge_setmulti()/axge_setpromisc() with axge_rxfilter(). Multicast filter programming and promiscuous mode requires access to a common RX configuration register so there is no need to use separate functions with added complexity. axge_rxfilter() does not read back AXGE_RCR register since accessing a register in USB is too slow and we already have all knowledge of required configuration. Rebuilding RX filter configuration is simpler and faster than manipulating every bits after reading back the register. Note, axge_rxfilter() does not set RCR_IPE(IP header alignment on 32bit boundary) to disable extra padding bytes insertion. The extra padding wastes ethernet to USB host bandwidth as well as complicating RX handling logic. Current USB framework requires copying RX frames to mbufs so there is no need to worry about alignment. Previously axge_rx_frame() performed wrong bound check due to the extra padding and it was broken when RX checksum offloading is disabled. See added comment in axge_rx_frame () for actual RX packet layout. In axge_init(), disable WOL. It's meaningless to enable WOL in normal operation. In axge_rxeof(), use properly sized mbuf rather than blindly allocating a mbuf cluster. Use RX H/W checksum offloading only when administrator requested RX checksum offloading. Previously it always used RX H/W checksum offloading result regardless of RX checksum offloading state. Separate L4 checksum offloading validation from L3 one and properly set required offloading bits for each layer. This is to fix setting L4 checksum offloading bits for L3 packets. There are still lots of RX errors(probably RX FIFO overflows) under moderate load. Users are strongly recommended to enable ethernet flow control. Reviewed by: kevlo (initial version), hselasky
2016-08-18 06:29:07 +00:00
uint32_t pos;
uint32_t pkt_cnt, pkt_end;
uint32_t hdr_off;
Introduce axge_rxfilter() which configures RX filtering and replace axge_setmulti()/axge_setpromisc() with axge_rxfilter(). Multicast filter programming and promiscuous mode requires access to a common RX configuration register so there is no need to use separate functions with added complexity. axge_rxfilter() does not read back AXGE_RCR register since accessing a register in USB is too slow and we already have all knowledge of required configuration. Rebuilding RX filter configuration is simpler and faster than manipulating every bits after reading back the register. Note, axge_rxfilter() does not set RCR_IPE(IP header alignment on 32bit boundary) to disable extra padding bytes insertion. The extra padding wastes ethernet to USB host bandwidth as well as complicating RX handling logic. Current USB framework requires copying RX frames to mbufs so there is no need to worry about alignment. Previously axge_rx_frame() performed wrong bound check due to the extra padding and it was broken when RX checksum offloading is disabled. See added comment in axge_rx_frame () for actual RX packet layout. In axge_init(), disable WOL. It's meaningless to enable WOL in normal operation. In axge_rxeof(), use properly sized mbuf rather than blindly allocating a mbuf cluster. Use RX H/W checksum offloading only when administrator requested RX checksum offloading. Previously it always used RX H/W checksum offloading result regardless of RX checksum offloading state. Separate L4 checksum offloading validation from L3 one and properly set required offloading bits for each layer. This is to fix setting L4 checksum offloading bits for L3 packets. There are still lots of RX errors(probably RX FIFO overflows) under moderate load. Users are strongly recommended to enable ethernet flow control. Reviewed by: kevlo (initial version), hselasky
2016-08-18 06:29:07 +00:00
uint32_t pktlen;
/* verify we have enough data */
if (actlen < (int)sizeof(rxhdr))
return;
pos = 0;
usbd_copy_out(pc, actlen - sizeof(rxhdr), &rxhdr, sizeof(rxhdr));
rxhdr = le32toh(rxhdr);
Introduce axge_rxfilter() which configures RX filtering and replace axge_setmulti()/axge_setpromisc() with axge_rxfilter(). Multicast filter programming and promiscuous mode requires access to a common RX configuration register so there is no need to use separate functions with added complexity. axge_rxfilter() does not read back AXGE_RCR register since accessing a register in USB is too slow and we already have all knowledge of required configuration. Rebuilding RX filter configuration is simpler and faster than manipulating every bits after reading back the register. Note, axge_rxfilter() does not set RCR_IPE(IP header alignment on 32bit boundary) to disable extra padding bytes insertion. The extra padding wastes ethernet to USB host bandwidth as well as complicating RX handling logic. Current USB framework requires copying RX frames to mbufs so there is no need to worry about alignment. Previously axge_rx_frame() performed wrong bound check due to the extra padding and it was broken when RX checksum offloading is disabled. See added comment in axge_rx_frame () for actual RX packet layout. In axge_init(), disable WOL. It's meaningless to enable WOL in normal operation. In axge_rxeof(), use properly sized mbuf rather than blindly allocating a mbuf cluster. Use RX H/W checksum offloading only when administrator requested RX checksum offloading. Previously it always used RX H/W checksum offloading result regardless of RX checksum offloading state. Separate L4 checksum offloading validation from L3 one and properly set required offloading bits for each layer. This is to fix setting L4 checksum offloading bits for L3 packets. There are still lots of RX errors(probably RX FIFO overflows) under moderate load. Users are strongly recommended to enable ethernet flow control. Reviewed by: kevlo (initial version), hselasky
2016-08-18 06:29:07 +00:00
pkt_cnt = rxhdr & 0xFFFF;
hdr_off = pkt_end = (rxhdr >> 16) & 0xFFFF;
Introduce axge_rxfilter() which configures RX filtering and replace axge_setmulti()/axge_setpromisc() with axge_rxfilter(). Multicast filter programming and promiscuous mode requires access to a common RX configuration register so there is no need to use separate functions with added complexity. axge_rxfilter() does not read back AXGE_RCR register since accessing a register in USB is too slow and we already have all knowledge of required configuration. Rebuilding RX filter configuration is simpler and faster than manipulating every bits after reading back the register. Note, axge_rxfilter() does not set RCR_IPE(IP header alignment on 32bit boundary) to disable extra padding bytes insertion. The extra padding wastes ethernet to USB host bandwidth as well as complicating RX handling logic. Current USB framework requires copying RX frames to mbufs so there is no need to worry about alignment. Previously axge_rx_frame() performed wrong bound check due to the extra padding and it was broken when RX checksum offloading is disabled. See added comment in axge_rx_frame () for actual RX packet layout. In axge_init(), disable WOL. It's meaningless to enable WOL in normal operation. In axge_rxeof(), use properly sized mbuf rather than blindly allocating a mbuf cluster. Use RX H/W checksum offloading only when administrator requested RX checksum offloading. Previously it always used RX H/W checksum offloading result regardless of RX checksum offloading state. Separate L4 checksum offloading validation from L3 one and properly set required offloading bits for each layer. This is to fix setting L4 checksum offloading bits for L3 packets. There are still lots of RX errors(probably RX FIFO overflows) under moderate load. Users are strongly recommended to enable ethernet flow control. Reviewed by: kevlo (initial version), hselasky
2016-08-18 06:29:07 +00:00
/*
* <----------------------- actlen ------------------------>
* [frame #0]...[frame #N][pkt_hdr #0]...[pkt_hdr #N][rxhdr]
* Each RX frame would be aligned on 8 bytes boundary. If
* RCR_IPE bit is set in AXGE_RCR register, there would be 2
* padding bytes and 6 dummy bytes(as the padding also should
* be aligned on 8 bytes boundary) for each RX frame to align
* IP header on 32bits boundary. Driver don't set RCR_IPE bit
* of AXGE_RCR register, so there should be no padding bytes
* which simplifies RX logic a lot.
*/
while (pkt_cnt--) {
/* verify the header offset */
if ((int)(hdr_off + sizeof(pkt_hdr)) > actlen) {
DPRINTF("End of packet headers\n");
break;
}
Introduce axge_rxfilter() which configures RX filtering and replace axge_setmulti()/axge_setpromisc() with axge_rxfilter(). Multicast filter programming and promiscuous mode requires access to a common RX configuration register so there is no need to use separate functions with added complexity. axge_rxfilter() does not read back AXGE_RCR register since accessing a register in USB is too slow and we already have all knowledge of required configuration. Rebuilding RX filter configuration is simpler and faster than manipulating every bits after reading back the register. Note, axge_rxfilter() does not set RCR_IPE(IP header alignment on 32bit boundary) to disable extra padding bytes insertion. The extra padding wastes ethernet to USB host bandwidth as well as complicating RX handling logic. Current USB framework requires copying RX frames to mbufs so there is no need to worry about alignment. Previously axge_rx_frame() performed wrong bound check due to the extra padding and it was broken when RX checksum offloading is disabled. See added comment in axge_rx_frame () for actual RX packet layout. In axge_init(), disable WOL. It's meaningless to enable WOL in normal operation. In axge_rxeof(), use properly sized mbuf rather than blindly allocating a mbuf cluster. Use RX H/W checksum offloading only when administrator requested RX checksum offloading. Previously it always used RX H/W checksum offloading result regardless of RX checksum offloading state. Separate L4 checksum offloading validation from L3 one and properly set required offloading bits for each layer. This is to fix setting L4 checksum offloading bits for L3 packets. There are still lots of RX errors(probably RX FIFO overflows) under moderate load. Users are strongly recommended to enable ethernet flow control. Reviewed by: kevlo (initial version), hselasky
2016-08-18 06:29:07 +00:00
usbd_copy_out(pc, hdr_off, &pkt_hdr, sizeof(pkt_hdr));
pkt_hdr.status = le32toh(pkt_hdr.status);
pktlen = AXGE_RXBYTES(pkt_hdr.status);
if (pos + pktlen > pkt_end) {
DPRINTF("Data position reached end\n");
break;
}
Introduce axge_rxfilter() which configures RX filtering and replace axge_setmulti()/axge_setpromisc() with axge_rxfilter(). Multicast filter programming and promiscuous mode requires access to a common RX configuration register so there is no need to use separate functions with added complexity. axge_rxfilter() does not read back AXGE_RCR register since accessing a register in USB is too slow and we already have all knowledge of required configuration. Rebuilding RX filter configuration is simpler and faster than manipulating every bits after reading back the register. Note, axge_rxfilter() does not set RCR_IPE(IP header alignment on 32bit boundary) to disable extra padding bytes insertion. The extra padding wastes ethernet to USB host bandwidth as well as complicating RX handling logic. Current USB framework requires copying RX frames to mbufs so there is no need to worry about alignment. Previously axge_rx_frame() performed wrong bound check due to the extra padding and it was broken when RX checksum offloading is disabled. See added comment in axge_rx_frame () for actual RX packet layout. In axge_init(), disable WOL. It's meaningless to enable WOL in normal operation. In axge_rxeof(), use properly sized mbuf rather than blindly allocating a mbuf cluster. Use RX H/W checksum offloading only when administrator requested RX checksum offloading. Previously it always used RX H/W checksum offloading result regardless of RX checksum offloading state. Separate L4 checksum offloading validation from L3 one and properly set required offloading bits for each layer. This is to fix setting L4 checksum offloading bits for L3 packets. There are still lots of RX errors(probably RX FIFO overflows) under moderate load. Users are strongly recommended to enable ethernet flow control. Reviewed by: kevlo (initial version), hselasky
2016-08-18 06:29:07 +00:00
if (AXGE_RX_ERR(pkt_hdr.status) != 0) {
DPRINTF("Dropped a packet\n");
if_inc_counter(ue->ue_ifp, IFCOUNTER_IERRORS, 1);
Introduce axge_rxfilter() which configures RX filtering and replace axge_setmulti()/axge_setpromisc() with axge_rxfilter(). Multicast filter programming and promiscuous mode requires access to a common RX configuration register so there is no need to use separate functions with added complexity. axge_rxfilter() does not read back AXGE_RCR register since accessing a register in USB is too slow and we already have all knowledge of required configuration. Rebuilding RX filter configuration is simpler and faster than manipulating every bits after reading back the register. Note, axge_rxfilter() does not set RCR_IPE(IP header alignment on 32bit boundary) to disable extra padding bytes insertion. The extra padding wastes ethernet to USB host bandwidth as well as complicating RX handling logic. Current USB framework requires copying RX frames to mbufs so there is no need to worry about alignment. Previously axge_rx_frame() performed wrong bound check due to the extra padding and it was broken when RX checksum offloading is disabled. See added comment in axge_rx_frame () for actual RX packet layout. In axge_init(), disable WOL. It's meaningless to enable WOL in normal operation. In axge_rxeof(), use properly sized mbuf rather than blindly allocating a mbuf cluster. Use RX H/W checksum offloading only when administrator requested RX checksum offloading. Previously it always used RX H/W checksum offloading result regardless of RX checksum offloading state. Separate L4 checksum offloading validation from L3 one and properly set required offloading bits for each layer. This is to fix setting L4 checksum offloading bits for L3 packets. There are still lots of RX errors(probably RX FIFO overflows) under moderate load. Users are strongly recommended to enable ethernet flow control. Reviewed by: kevlo (initial version), hselasky
2016-08-18 06:29:07 +00:00
} else
axge_rxeof(ue, pc, pos, pktlen, pkt_hdr.status);
pos += (pktlen + 7) & ~7;
hdr_off += sizeof(pkt_hdr);
}
}
static void
Introduce axge_rxfilter() which configures RX filtering and replace axge_setmulti()/axge_setpromisc() with axge_rxfilter(). Multicast filter programming and promiscuous mode requires access to a common RX configuration register so there is no need to use separate functions with added complexity. axge_rxfilter() does not read back AXGE_RCR register since accessing a register in USB is too slow and we already have all knowledge of required configuration. Rebuilding RX filter configuration is simpler and faster than manipulating every bits after reading back the register. Note, axge_rxfilter() does not set RCR_IPE(IP header alignment on 32bit boundary) to disable extra padding bytes insertion. The extra padding wastes ethernet to USB host bandwidth as well as complicating RX handling logic. Current USB framework requires copying RX frames to mbufs so there is no need to worry about alignment. Previously axge_rx_frame() performed wrong bound check due to the extra padding and it was broken when RX checksum offloading is disabled. See added comment in axge_rx_frame () for actual RX packet layout. In axge_init(), disable WOL. It's meaningless to enable WOL in normal operation. In axge_rxeof(), use properly sized mbuf rather than blindly allocating a mbuf cluster. Use RX H/W checksum offloading only when administrator requested RX checksum offloading. Previously it always used RX H/W checksum offloading result regardless of RX checksum offloading state. Separate L4 checksum offloading validation from L3 one and properly set required offloading bits for each layer. This is to fix setting L4 checksum offloading bits for L3 packets. There are still lots of RX errors(probably RX FIFO overflows) under moderate load. Users are strongly recommended to enable ethernet flow control. Reviewed by: kevlo (initial version), hselasky
2016-08-18 06:29:07 +00:00
axge_rxeof(struct usb_ether *ue, struct usb_page_cache *pc, unsigned int offset,
unsigned int len, uint32_t status)
{
struct ifnet *ifp;
struct mbuf *m;
ifp = ue->ue_ifp;
if (len < ETHER_HDR_LEN || len > MCLBYTES - ETHER_ALIGN) {
if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
return;
}
Introduce axge_rxfilter() which configures RX filtering and replace axge_setmulti()/axge_setpromisc() with axge_rxfilter(). Multicast filter programming and promiscuous mode requires access to a common RX configuration register so there is no need to use separate functions with added complexity. axge_rxfilter() does not read back AXGE_RCR register since accessing a register in USB is too slow and we already have all knowledge of required configuration. Rebuilding RX filter configuration is simpler and faster than manipulating every bits after reading back the register. Note, axge_rxfilter() does not set RCR_IPE(IP header alignment on 32bit boundary) to disable extra padding bytes insertion. The extra padding wastes ethernet to USB host bandwidth as well as complicating RX handling logic. Current USB framework requires copying RX frames to mbufs so there is no need to worry about alignment. Previously axge_rx_frame() performed wrong bound check due to the extra padding and it was broken when RX checksum offloading is disabled. See added comment in axge_rx_frame () for actual RX packet layout. In axge_init(), disable WOL. It's meaningless to enable WOL in normal operation. In axge_rxeof(), use properly sized mbuf rather than blindly allocating a mbuf cluster. Use RX H/W checksum offloading only when administrator requested RX checksum offloading. Previously it always used RX H/W checksum offloading result regardless of RX checksum offloading state. Separate L4 checksum offloading validation from L3 one and properly set required offloading bits for each layer. This is to fix setting L4 checksum offloading bits for L3 packets. There are still lots of RX errors(probably RX FIFO overflows) under moderate load. Users are strongly recommended to enable ethernet flow control. Reviewed by: kevlo (initial version), hselasky
2016-08-18 06:29:07 +00:00
if (len > MHLEN - ETHER_ALIGN)
m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
else
m = m_gethdr(M_NOWAIT, MT_DATA);
if (m == NULL) {
if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
return;
}
m->m_pkthdr.rcvif = ifp;
Introduce axge_rxfilter() which configures RX filtering and replace axge_setmulti()/axge_setpromisc() with axge_rxfilter(). Multicast filter programming and promiscuous mode requires access to a common RX configuration register so there is no need to use separate functions with added complexity. axge_rxfilter() does not read back AXGE_RCR register since accessing a register in USB is too slow and we already have all knowledge of required configuration. Rebuilding RX filter configuration is simpler and faster than manipulating every bits after reading back the register. Note, axge_rxfilter() does not set RCR_IPE(IP header alignment on 32bit boundary) to disable extra padding bytes insertion. The extra padding wastes ethernet to USB host bandwidth as well as complicating RX handling logic. Current USB framework requires copying RX frames to mbufs so there is no need to worry about alignment. Previously axge_rx_frame() performed wrong bound check due to the extra padding and it was broken when RX checksum offloading is disabled. See added comment in axge_rx_frame () for actual RX packet layout. In axge_init(), disable WOL. It's meaningless to enable WOL in normal operation. In axge_rxeof(), use properly sized mbuf rather than blindly allocating a mbuf cluster. Use RX H/W checksum offloading only when administrator requested RX checksum offloading. Previously it always used RX H/W checksum offloading result regardless of RX checksum offloading state. Separate L4 checksum offloading validation from L3 one and properly set required offloading bits for each layer. This is to fix setting L4 checksum offloading bits for L3 packets. There are still lots of RX errors(probably RX FIFO overflows) under moderate load. Users are strongly recommended to enable ethernet flow control. Reviewed by: kevlo (initial version), hselasky
2016-08-18 06:29:07 +00:00
m->m_len = m->m_pkthdr.len = len;
m->m_data += ETHER_ALIGN;
usbd_copy_out(pc, offset, mtod(m, uint8_t *), len);
Introduce axge_rxfilter() which configures RX filtering and replace axge_setmulti()/axge_setpromisc() with axge_rxfilter(). Multicast filter programming and promiscuous mode requires access to a common RX configuration register so there is no need to use separate functions with added complexity. axge_rxfilter() does not read back AXGE_RCR register since accessing a register in USB is too slow and we already have all knowledge of required configuration. Rebuilding RX filter configuration is simpler and faster than manipulating every bits after reading back the register. Note, axge_rxfilter() does not set RCR_IPE(IP header alignment on 32bit boundary) to disable extra padding bytes insertion. The extra padding wastes ethernet to USB host bandwidth as well as complicating RX handling logic. Current USB framework requires copying RX frames to mbufs so there is no need to worry about alignment. Previously axge_rx_frame() performed wrong bound check due to the extra padding and it was broken when RX checksum offloading is disabled. See added comment in axge_rx_frame () for actual RX packet layout. In axge_init(), disable WOL. It's meaningless to enable WOL in normal operation. In axge_rxeof(), use properly sized mbuf rather than blindly allocating a mbuf cluster. Use RX H/W checksum offloading only when administrator requested RX checksum offloading. Previously it always used RX H/W checksum offloading result regardless of RX checksum offloading state. Separate L4 checksum offloading validation from L3 one and properly set required offloading bits for each layer. This is to fix setting L4 checksum offloading bits for L3 packets. There are still lots of RX errors(probably RX FIFO overflows) under moderate load. Users are strongly recommended to enable ethernet flow control. Reviewed by: kevlo (initial version), hselasky
2016-08-18 06:29:07 +00:00
if ((ifp->if_capenable & IFCAP_RXCSUM) != 0) {
if ((status & AXGE_RX_L3_CSUM_ERR) == 0 &&
(status & AXGE_RX_L3_TYPE_MASK) == AXGE_RX_L3_TYPE_IPV4)
m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED |
CSUM_IP_VALID;
if ((status & AXGE_RX_L4_CSUM_ERR) == 0 &&
((status & AXGE_RX_L4_TYPE_MASK) == AXGE_RX_L4_TYPE_UDP ||
(status & AXGE_RX_L4_TYPE_MASK) == AXGE_RX_L4_TYPE_TCP)) {
m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
Introduce axge_rxfilter() which configures RX filtering and replace axge_setmulti()/axge_setpromisc() with axge_rxfilter(). Multicast filter programming and promiscuous mode requires access to a common RX configuration register so there is no need to use separate functions with added complexity. axge_rxfilter() does not read back AXGE_RCR register since accessing a register in USB is too slow and we already have all knowledge of required configuration. Rebuilding RX filter configuration is simpler and faster than manipulating every bits after reading back the register. Note, axge_rxfilter() does not set RCR_IPE(IP header alignment on 32bit boundary) to disable extra padding bytes insertion. The extra padding wastes ethernet to USB host bandwidth as well as complicating RX handling logic. Current USB framework requires copying RX frames to mbufs so there is no need to worry about alignment. Previously axge_rx_frame() performed wrong bound check due to the extra padding and it was broken when RX checksum offloading is disabled. See added comment in axge_rx_frame () for actual RX packet layout. In axge_init(), disable WOL. It's meaningless to enable WOL in normal operation. In axge_rxeof(), use properly sized mbuf rather than blindly allocating a mbuf cluster. Use RX H/W checksum offloading only when administrator requested RX checksum offloading. Previously it always used RX H/W checksum offloading result regardless of RX checksum offloading state. Separate L4 checksum offloading validation from L3 one and properly set required offloading bits for each layer. This is to fix setting L4 checksum offloading bits for L3 packets. There are still lots of RX errors(probably RX FIFO overflows) under moderate load. Users are strongly recommended to enable ethernet flow control. Reviewed by: kevlo (initial version), hselasky
2016-08-18 06:29:07 +00:00
CSUM_PSEUDO_HDR;
m->m_pkthdr.csum_data = 0xffff;
}
}
Introduce axge_rxfilter() which configures RX filtering and replace axge_setmulti()/axge_setpromisc() with axge_rxfilter(). Multicast filter programming and promiscuous mode requires access to a common RX configuration register so there is no need to use separate functions with added complexity. axge_rxfilter() does not read back AXGE_RCR register since accessing a register in USB is too slow and we already have all knowledge of required configuration. Rebuilding RX filter configuration is simpler and faster than manipulating every bits after reading back the register. Note, axge_rxfilter() does not set RCR_IPE(IP header alignment on 32bit boundary) to disable extra padding bytes insertion. The extra padding wastes ethernet to USB host bandwidth as well as complicating RX handling logic. Current USB framework requires copying RX frames to mbufs so there is no need to worry about alignment. Previously axge_rx_frame() performed wrong bound check due to the extra padding and it was broken when RX checksum offloading is disabled. See added comment in axge_rx_frame () for actual RX packet layout. In axge_init(), disable WOL. It's meaningless to enable WOL in normal operation. In axge_rxeof(), use properly sized mbuf rather than blindly allocating a mbuf cluster. Use RX H/W checksum offloading only when administrator requested RX checksum offloading. Previously it always used RX H/W checksum offloading result regardless of RX checksum offloading state. Separate L4 checksum offloading validation from L3 one and properly set required offloading bits for each layer. This is to fix setting L4 checksum offloading bits for L3 packets. There are still lots of RX errors(probably RX FIFO overflows) under moderate load. Users are strongly recommended to enable ethernet flow control. Reviewed by: kevlo (initial version), hselasky
2016-08-18 06:29:07 +00:00
if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
_IF_ENQUEUE(&ue->ue_rxq, m);
}
static void
axge_csum_cfg(struct usb_ether *ue)
{
struct axge_softc *sc;
struct ifnet *ifp;
uint8_t csum;
sc = uether_getsc(ue);
AXGE_LOCK_ASSERT(sc, MA_OWNED);
ifp = uether_getifp(ue);
csum = 0;
if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
csum |= CTCR_IP | CTCR_TCP | CTCR_UDP;
axge_write_cmd_1(sc, AXGE_ACCESS_MAC, AXGE_CTCR, csum);
csum = 0;
if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
csum |= CRCR_IP | CRCR_TCP | CRCR_UDP;
axge_write_cmd_1(sc, AXGE_ACCESS_MAC, AXGE_CRCR, csum);
}