freebsd-dev/sys/dev/et/if_et.c

2752 lines
72 KiB
C
Raw Normal View History

/*-
* SPDX-License-Identifier: BSD-3-Clause
*
* Copyright (c) 2007 Sepherosa Ziehau. All rights reserved.
*
* This code is derived from software contributed to The DragonFly Project
* by Sepherosa Ziehau <sepherosa@gmail.com>
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
*
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in
* the documentation and/or other materials provided with the
* distribution.
* 3. Neither the name of The DragonFly Project nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific, prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS 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
* COPYRIGHT HOLDERS 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.
*
* $DragonFly: src/sys/dev/netif/et/if_et.c,v 1.10 2008/05/18 07:47:14 sephe Exp $
*/
2009-11-20 20:40:34 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/endian.h>
#include <sys/kernel.h>
#include <sys/bus.h>
#include <sys/malloc.h>
#include <sys/mbuf.h>
#include <sys/proc.h>
#include <sys/rman.h>
#include <sys/module.h>
#include <sys/socket.h>
#include <sys/sockio.h>
#include <sys/sysctl.h>
#include <net/ethernet.h>
#include <net/if.h>
#include <net/if_var.h>
#include <net/if_dl.h>
#include <net/if_types.h>
#include <net/bpf.h>
#include <net/if_arp.h>
#include <net/if_media.h>
#include <net/if_vlan_var.h>
#include <machine/bus.h>
#include <dev/mii/mii.h>
#include <dev/mii/miivar.h>
#include <dev/pci/pcireg.h>
#include <dev/pci/pcivar.h>
#include <dev/et/if_etreg.h>
#include <dev/et/if_etvar.h>
#include "miibus_if.h"
MODULE_DEPEND(et, pci, 1, 1, 1);
MODULE_DEPEND(et, ether, 1, 1, 1);
MODULE_DEPEND(et, miibus, 1, 1, 1);
2009-11-19 21:45:06 +00:00
/* Tunables. */
static int msi_disable = 0;
TUNABLE_INT("hw.et.msi_disable", &msi_disable);
2009-11-19 21:45:06 +00:00
#define ET_CSUM_FEATURES (CSUM_IP | CSUM_TCP | CSUM_UDP)
static int et_probe(device_t);
static int et_attach(device_t);
static int et_detach(device_t);
static int et_shutdown(device_t);
static int et_suspend(device_t);
static int et_resume(device_t);
static int et_miibus_readreg(device_t, int, int);
static int et_miibus_writereg(device_t, int, int, int);
static void et_miibus_statchg(device_t);
static void et_init_locked(struct et_softc *);
static void et_init(void *);
static int et_ioctl(struct ifnet *, u_long, caddr_t);
static void et_start_locked(struct ifnet *);
static void et_start(struct ifnet *);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
static int et_watchdog(struct et_softc *);
static int et_ifmedia_upd_locked(struct ifnet *);
static int et_ifmedia_upd(struct ifnet *);
static void et_ifmedia_sts(struct ifnet *, struct ifmediareq *);
static uint64_t et_get_counter(struct ifnet *, ift_counter);
static void et_add_sysctls(struct et_softc *);
static int et_sysctl_rx_intr_npkts(SYSCTL_HANDLER_ARGS);
static int et_sysctl_rx_intr_delay(SYSCTL_HANDLER_ARGS);
static void et_intr(void *);
static void et_rxeof(struct et_softc *);
static void et_txeof(struct et_softc *);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
static int et_dma_alloc(struct et_softc *);
static void et_dma_free(struct et_softc *);
static void et_dma_map_addr(void *, bus_dma_segment_t *, int, int);
static int et_dma_ring_alloc(struct et_softc *, bus_size_t, bus_size_t,
bus_dma_tag_t *, uint8_t **, bus_dmamap_t *, bus_addr_t *,
const char *);
static void et_dma_ring_free(struct et_softc *, bus_dma_tag_t *, uint8_t **,
bus_dmamap_t, bus_addr_t *);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
static void et_init_tx_ring(struct et_softc *);
static int et_init_rx_ring(struct et_softc *);
static void et_free_tx_ring(struct et_softc *);
static void et_free_rx_ring(struct et_softc *);
static int et_encap(struct et_softc *, struct mbuf **);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
static int et_newbuf_cluster(struct et_rxbuf_data *, int);
static int et_newbuf_hdr(struct et_rxbuf_data *, int);
static void et_rxbuf_discard(struct et_rxbuf_data *, int);
static void et_stop(struct et_softc *);
static int et_chip_init(struct et_softc *);
static void et_chip_attach(struct et_softc *);
static void et_init_mac(struct et_softc *);
static void et_init_rxmac(struct et_softc *);
static void et_init_txmac(struct et_softc *);
static int et_init_rxdma(struct et_softc *);
static int et_init_txdma(struct et_softc *);
static int et_start_rxdma(struct et_softc *);
static int et_start_txdma(struct et_softc *);
static int et_stop_rxdma(struct et_softc *);
static int et_stop_txdma(struct et_softc *);
static void et_reset(struct et_softc *);
static int et_bus_config(struct et_softc *);
static void et_get_eaddr(device_t, uint8_t[]);
static void et_setmulti(struct et_softc *);
static void et_tick(void *);
static void et_stats_update(struct et_softc *);
static const struct et_dev {
uint16_t vid;
uint16_t did;
const char *desc;
} et_devices[] = {
{ PCI_VENDOR_LUCENT, PCI_PRODUCT_LUCENT_ET1310,
"Agere ET1310 Gigabit Ethernet" },
{ PCI_VENDOR_LUCENT, PCI_PRODUCT_LUCENT_ET1310_FAST,
"Agere ET1310 Fast Ethernet" },
{ 0, 0, NULL }
};
static device_method_t et_methods[] = {
DEVMETHOD(device_probe, et_probe),
DEVMETHOD(device_attach, et_attach),
DEVMETHOD(device_detach, et_detach),
DEVMETHOD(device_shutdown, et_shutdown),
DEVMETHOD(device_suspend, et_suspend),
DEVMETHOD(device_resume, et_resume),
DEVMETHOD(miibus_readreg, et_miibus_readreg),
DEVMETHOD(miibus_writereg, et_miibus_writereg),
DEVMETHOD(miibus_statchg, et_miibus_statchg),
DEVMETHOD_END
};
static driver_t et_driver = {
"et",
et_methods,
sizeof(struct et_softc)
};
static devclass_t et_devclass;
DRIVER_MODULE(et, pci, et_driver, et_devclass, 0, 0);
MODULE_PNP_INFO("U16:vendor;U16:device;D:#", pci, et, et_devices,
nitems(et_devices) - 1);
DRIVER_MODULE(miibus, et, miibus_driver, miibus_devclass, 0, 0);
static int et_rx_intr_npkts = 32;
static int et_rx_intr_delay = 20; /* x10 usec */
static int et_tx_intr_nsegs = 126;
static uint32_t et_timer = 1000 * 1000 * 1000; /* nanosec */
TUNABLE_INT("hw.et.timer", &et_timer);
TUNABLE_INT("hw.et.rx_intr_npkts", &et_rx_intr_npkts);
TUNABLE_INT("hw.et.rx_intr_delay", &et_rx_intr_delay);
TUNABLE_INT("hw.et.tx_intr_nsegs", &et_tx_intr_nsegs);
static int
et_probe(device_t dev)
{
const struct et_dev *d;
uint16_t did, vid;
vid = pci_get_vendor(dev);
did = pci_get_device(dev);
for (d = et_devices; d->desc != NULL; ++d) {
if (vid == d->vid && did == d->did) {
device_set_desc(dev, d->desc);
return (BUS_PROBE_DEFAULT);
}
}
2009-11-19 21:53:21 +00:00
return (ENXIO);
}
static int
et_attach(device_t dev)
{
struct et_softc *sc;
struct ifnet *ifp;
uint8_t eaddr[ETHER_ADDR_LEN];
uint32_t pmcfg;
2009-11-19 21:45:06 +00:00
int cap, error, msic;
sc = device_get_softc(dev);
sc->dev = dev;
mtx_init(&sc->sc_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
MTX_DEF);
callout_init_mtx(&sc->sc_tick, &sc->sc_mtx, 0);
ifp = sc->ifp = if_alloc(IFT_ETHER);
if (ifp == NULL) {
device_printf(dev, "can not if_alloc()\n");
error = ENOSPC;
goto fail;
}
/*
* Initialize tunables
*/
sc->sc_rx_intr_npkts = et_rx_intr_npkts;
sc->sc_rx_intr_delay = et_rx_intr_delay;
sc->sc_tx_intr_nsegs = et_tx_intr_nsegs;
sc->sc_timer = et_timer;
/* Enable bus mastering */
pci_enable_busmaster(dev);
/*
* Allocate IO memory
*/
sc->sc_mem_rid = PCIR_BAR(0);
sc->sc_mem_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
&sc->sc_mem_rid, RF_ACTIVE);
if (sc->sc_mem_res == NULL) {
device_printf(dev, "can't allocate IO memory\n");
2009-11-19 21:53:21 +00:00
return (ENXIO);
}
2009-11-19 21:45:06 +00:00
msic = 0;
if (pci_find_cap(dev, PCIY_EXPRESS, &cap) == 0) {
2009-11-19 21:45:06 +00:00
sc->sc_expcap = cap;
sc->sc_flags |= ET_FLAG_PCIE;
msic = pci_msi_count(dev);
if (bootverbose)
device_printf(dev, "MSI count: %d\n", msic);
2009-11-19 21:45:06 +00:00
}
if (msic > 0 && msi_disable == 0) {
msic = 1;
if (pci_alloc_msi(dev, &msic) == 0) {
if (msic == 1) {
device_printf(dev, "Using %d MSI message\n",
msic);
sc->sc_flags |= ET_FLAG_MSI;
} else
pci_release_msi(dev);
}
}
/*
* Allocate IRQ
*/
2009-11-19 21:45:06 +00:00
if ((sc->sc_flags & ET_FLAG_MSI) == 0) {
sc->sc_irq_rid = 0;
sc->sc_irq_res = bus_alloc_resource_any(dev, SYS_RES_IRQ,
&sc->sc_irq_rid, RF_SHAREABLE | RF_ACTIVE);
} else {
sc->sc_irq_rid = 1;
sc->sc_irq_res = bus_alloc_resource_any(dev, SYS_RES_IRQ,
&sc->sc_irq_rid, RF_ACTIVE);
}
if (sc->sc_irq_res == NULL) {
device_printf(dev, "can't allocate irq\n");
error = ENXIO;
goto fail;
}
Rework link state tracking and TX/RX MAC configuration. o Do not report link status if driver is not running. o TX/RX MAC configuration should be done with resolved speed, duplex and flow control after establishing a link so it can't be done in driver initialization routine. Move the configuration to miibus_statchg callback which will be called whenever any link state change is detected. At this moment, flow-control is not enabled yet mainly because I was not able to set correct flow control parameters to generate TX pause frames. o Now TX/RX MAC is enabled only when a valid link is detected. Rearragnge hardware initialization routine a bit to leave enabling MAC to miibus_statchg callback. In order to that, TX/RX DMA engine is enabled in et_init_locked(). o Introduce ET_FLAG_LINK flag to track current link state. o Introduce ET_FLAG_FASTETHER flag to mark whether controller is fast ethernet. This flag is checked in miibus_statchg callback to know whether PHY established a valid link. o In et_stop(), TX/RX MAC is explicitly disabled instead of relying on et_reset(). And move et_reset() from et_stop() to controller initialization. Controler reset is not required here and it would also clear critial registers(i.e station address, RX filter configuration, WOL etc) that are required to make WOL work. o Switching to current media is done in et_init_locked() after setting IFF_DRV_RUNNING flag. This should ensure reliable auto-negotiation/manual link establishment. o In et_start_locked(), check whether driver got a valid link before trying to send frames. o Remove checking a link in et_tick() as this is done by miibus_statchg callback.
2011-12-07 21:29:51 +00:00
if (pci_get_device(dev) == PCI_PRODUCT_LUCENT_ET1310_FAST)
sc->sc_flags |= ET_FLAG_FASTETHER;
error = et_bus_config(sc);
if (error)
goto fail;
et_get_eaddr(dev, eaddr);
/* Take PHY out of COMA and enable clocks. */
pmcfg = ET_PM_SYSCLK_GATE | ET_PM_TXCLK_GATE | ET_PM_RXCLK_GATE;
if ((sc->sc_flags & ET_FLAG_FASTETHER) == 0)
pmcfg |= EM_PM_GIGEPHY_ENB;
CSR_WRITE_4(sc, ET_PM, pmcfg);
et_reset(sc);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
error = et_dma_alloc(sc);
if (error)
goto fail;
ifp->if_softc = sc;
if_initname(ifp, device_get_name(dev), device_get_unit(dev));
ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
ifp->if_init = et_init;
ifp->if_ioctl = et_ioctl;
ifp->if_start = et_start;
ifp->if_get_counter = et_get_counter;
ifp->if_capabilities = IFCAP_TXCSUM | IFCAP_VLAN_MTU;
ifp->if_capenable = ifp->if_capabilities;
ifp->if_snd.ifq_drv_maxlen = ET_TX_NDESC - 1;
IFQ_SET_MAXLEN(&ifp->if_snd, ET_TX_NDESC - 1);
IFQ_SET_READY(&ifp->if_snd);
et_chip_attach(sc);
error = mii_attach(dev, &sc->sc_miibus, ifp, et_ifmedia_upd,
et_ifmedia_sts, BMSR_DEFCAPMASK, MII_PHY_ANY, MII_OFFSET_ANY,
MIIF_DOPAUSE);
if (error) {
device_printf(dev, "attaching PHYs failed\n");
goto fail;
}
ether_ifattach(ifp, eaddr);
/* Tell the upper layer(s) we support long frames. */
ifp->if_hdrlen = sizeof(struct ether_vlan_header);
error = bus_setup_intr(dev, sc->sc_irq_res, INTR_TYPE_NET | INTR_MPSAFE,
2009-11-19 21:45:06 +00:00
NULL, et_intr, sc, &sc->sc_irq_handle);
if (error) {
ether_ifdetach(ifp);
device_printf(dev, "can't setup intr\n");
goto fail;
}
et_add_sysctls(sc);
2009-11-19 21:53:21 +00:00
return (0);
fail:
et_detach(dev);
2009-11-19 21:53:21 +00:00
return (error);
}
static int
et_detach(device_t dev)
{
2012-01-10 20:52:02 +00:00
struct et_softc *sc;
2012-01-10 20:52:02 +00:00
sc = device_get_softc(dev);
if (device_is_attached(dev)) {
ether_ifdetach(sc->ifp);
ET_LOCK(sc);
et_stop(sc);
ET_UNLOCK(sc);
callout_drain(&sc->sc_tick);
}
if (sc->sc_miibus != NULL)
device_delete_child(dev, sc->sc_miibus);
bus_generic_detach(dev);
if (sc->sc_irq_handle != NULL)
bus_teardown_intr(dev, sc->sc_irq_res, sc->sc_irq_handle);
if (sc->sc_irq_res != NULL)
bus_release_resource(dev, SYS_RES_IRQ,
rman_get_rid(sc->sc_irq_res), sc->sc_irq_res);
2009-11-19 21:45:06 +00:00
if ((sc->sc_flags & ET_FLAG_MSI) != 0)
pci_release_msi(dev);
if (sc->sc_mem_res != NULL)
bus_release_resource(dev, SYS_RES_MEMORY,
rman_get_rid(sc->sc_mem_res), sc->sc_mem_res);
if (sc->ifp != NULL)
if_free(sc->ifp);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
et_dma_free(sc);
2009-11-19 21:39:43 +00:00
mtx_destroy(&sc->sc_mtx);
2009-11-19 21:53:21 +00:00
return (0);
}
static int
et_shutdown(device_t dev)
{
2012-01-10 20:52:02 +00:00
struct et_softc *sc;
2012-01-10 20:52:02 +00:00
sc = device_get_softc(dev);
ET_LOCK(sc);
et_stop(sc);
ET_UNLOCK(sc);
2009-11-19 21:53:21 +00:00
return (0);
}
static int
et_miibus_readreg(device_t dev, int phy, int reg)
{
2012-01-10 20:52:02 +00:00
struct et_softc *sc;
uint32_t val;
int i, ret;
2012-01-10 20:52:02 +00:00
sc = device_get_softc(dev);
/* Stop any pending operations */
CSR_WRITE_4(sc, ET_MII_CMD, 0);
val = (phy << ET_MII_ADDR_PHY_SHIFT) & ET_MII_ADDR_PHY_MASK;
val |= (reg << ET_MII_ADDR_REG_SHIFT) & ET_MII_ADDR_REG_MASK;
CSR_WRITE_4(sc, ET_MII_ADDR, val);
/* Start reading */
CSR_WRITE_4(sc, ET_MII_CMD, ET_MII_CMD_READ);
#define NRETRY 50
for (i = 0; i < NRETRY; ++i) {
val = CSR_READ_4(sc, ET_MII_IND);
if ((val & (ET_MII_IND_BUSY | ET_MII_IND_INVALID)) == 0)
break;
DELAY(50);
}
if (i == NRETRY) {
if_printf(sc->ifp,
"read phy %d, reg %d timed out\n", phy, reg);
ret = 0;
goto back;
}
#undef NRETRY
val = CSR_READ_4(sc, ET_MII_STAT);
ret = val & ET_MII_STAT_VALUE_MASK;
back:
/* Make sure that the current operation is stopped */
CSR_WRITE_4(sc, ET_MII_CMD, 0);
2009-11-19 21:53:21 +00:00
return (ret);
}
static int
et_miibus_writereg(device_t dev, int phy, int reg, int val0)
{
2012-01-10 20:52:02 +00:00
struct et_softc *sc;
uint32_t val;
int i;
2012-01-10 20:52:02 +00:00
sc = device_get_softc(dev);
/* Stop any pending operations */
CSR_WRITE_4(sc, ET_MII_CMD, 0);
val = (phy << ET_MII_ADDR_PHY_SHIFT) & ET_MII_ADDR_PHY_MASK;
val |= (reg << ET_MII_ADDR_REG_SHIFT) & ET_MII_ADDR_REG_MASK;
CSR_WRITE_4(sc, ET_MII_ADDR, val);
/* Start writing */
CSR_WRITE_4(sc, ET_MII_CTRL,
(val0 << ET_MII_CTRL_VALUE_SHIFT) & ET_MII_CTRL_VALUE_MASK);
#define NRETRY 100
for (i = 0; i < NRETRY; ++i) {
val = CSR_READ_4(sc, ET_MII_IND);
if ((val & ET_MII_IND_BUSY) == 0)
break;
DELAY(50);
}
if (i == NRETRY) {
if_printf(sc->ifp,
"write phy %d, reg %d timed out\n", phy, reg);
et_miibus_readreg(dev, phy, reg);
}
#undef NRETRY
/* Make sure that the current operation is stopped */
CSR_WRITE_4(sc, ET_MII_CMD, 0);
2009-11-19 21:53:21 +00:00
return (0);
}
static void
et_miibus_statchg(device_t dev)
{
Rework link state tracking and TX/RX MAC configuration. o Do not report link status if driver is not running. o TX/RX MAC configuration should be done with resolved speed, duplex and flow control after establishing a link so it can't be done in driver initialization routine. Move the configuration to miibus_statchg callback which will be called whenever any link state change is detected. At this moment, flow-control is not enabled yet mainly because I was not able to set correct flow control parameters to generate TX pause frames. o Now TX/RX MAC is enabled only when a valid link is detected. Rearragnge hardware initialization routine a bit to leave enabling MAC to miibus_statchg callback. In order to that, TX/RX DMA engine is enabled in et_init_locked(). o Introduce ET_FLAG_LINK flag to track current link state. o Introduce ET_FLAG_FASTETHER flag to mark whether controller is fast ethernet. This flag is checked in miibus_statchg callback to know whether PHY established a valid link. o In et_stop(), TX/RX MAC is explicitly disabled instead of relying on et_reset(). And move et_reset() from et_stop() to controller initialization. Controler reset is not required here and it would also clear critial registers(i.e station address, RX filter configuration, WOL etc) that are required to make WOL work. o Switching to current media is done in et_init_locked() after setting IFF_DRV_RUNNING flag. This should ensure reliable auto-negotiation/manual link establishment. o In et_start_locked(), check whether driver got a valid link before trying to send frames. o Remove checking a link in et_tick() as this is done by miibus_statchg callback.
2011-12-07 21:29:51 +00:00
struct et_softc *sc;
struct mii_data *mii;
struct ifnet *ifp;
uint32_t cfg1, cfg2, ctrl;
int i;
sc = device_get_softc(dev);
mii = device_get_softc(sc->sc_miibus);
ifp = sc->ifp;
if (mii == NULL || ifp == NULL ||
(ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
return;
sc->sc_flags &= ~ET_FLAG_LINK;
if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
(IFM_ACTIVE | IFM_AVALID)) {
switch (IFM_SUBTYPE(mii->mii_media_active)) {
case IFM_10_T:
case IFM_100_TX:
sc->sc_flags |= ET_FLAG_LINK;
break;
case IFM_1000_T:
if ((sc->sc_flags & ET_FLAG_FASTETHER) == 0)
sc->sc_flags |= ET_FLAG_LINK;
break;
}
}
/* XXX Stop TX/RX MAC? */
if ((sc->sc_flags & ET_FLAG_LINK) == 0)
return;
/* Program MACs with resolved speed/duplex/flow-control. */
ctrl = CSR_READ_4(sc, ET_MAC_CTRL);
ctrl &= ~(ET_MAC_CTRL_GHDX | ET_MAC_CTRL_MODE_MII);
cfg1 = CSR_READ_4(sc, ET_MAC_CFG1);
cfg1 &= ~(ET_MAC_CFG1_TXFLOW | ET_MAC_CFG1_RXFLOW |
ET_MAC_CFG1_LOOPBACK);
cfg2 = CSR_READ_4(sc, ET_MAC_CFG2);
cfg2 &= ~(ET_MAC_CFG2_MODE_MII | ET_MAC_CFG2_MODE_GMII |
ET_MAC_CFG2_FDX | ET_MAC_CFG2_BIGFRM);
cfg2 |= ET_MAC_CFG2_LENCHK | ET_MAC_CFG2_CRC | ET_MAC_CFG2_PADCRC |
((7 << ET_MAC_CFG2_PREAMBLE_LEN_SHIFT) &
ET_MAC_CFG2_PREAMBLE_LEN_MASK);
if (IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_T)
cfg2 |= ET_MAC_CFG2_MODE_GMII;
else {
cfg2 |= ET_MAC_CFG2_MODE_MII;
ctrl |= ET_MAC_CTRL_MODE_MII;
}
if (IFM_OPTIONS(mii->mii_media_active) & IFM_FDX) {
cfg2 |= ET_MAC_CFG2_FDX;
/*
* Controller lacks automatic TX pause frame
* generation so it should be handled by driver.
* Even though driver can send pause frame with
* arbitrary pause time, controller does not
* provide a way that tells how many free RX
* buffers are available in controller. This
* limitation makes it hard to generate XON frame
* in time on driver side so don't enable TX flow
* control.
*/
Rework link state tracking and TX/RX MAC configuration. o Do not report link status if driver is not running. o TX/RX MAC configuration should be done with resolved speed, duplex and flow control after establishing a link so it can't be done in driver initialization routine. Move the configuration to miibus_statchg callback which will be called whenever any link state change is detected. At this moment, flow-control is not enabled yet mainly because I was not able to set correct flow control parameters to generate TX pause frames. o Now TX/RX MAC is enabled only when a valid link is detected. Rearragnge hardware initialization routine a bit to leave enabling MAC to miibus_statchg callback. In order to that, TX/RX DMA engine is enabled in et_init_locked(). o Introduce ET_FLAG_LINK flag to track current link state. o Introduce ET_FLAG_FASTETHER flag to mark whether controller is fast ethernet. This flag is checked in miibus_statchg callback to know whether PHY established a valid link. o In et_stop(), TX/RX MAC is explicitly disabled instead of relying on et_reset(). And move et_reset() from et_stop() to controller initialization. Controler reset is not required here and it would also clear critial registers(i.e station address, RX filter configuration, WOL etc) that are required to make WOL work. o Switching to current media is done in et_init_locked() after setting IFF_DRV_RUNNING flag. This should ensure reliable auto-negotiation/manual link establishment. o In et_start_locked(), check whether driver got a valid link before trying to send frames. o Remove checking a link in et_tick() as this is done by miibus_statchg callback.
2011-12-07 21:29:51 +00:00
#ifdef notyet
if (IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_TXPAUSE)
cfg1 |= ET_MAC_CFG1_TXFLOW;
#endif
Rework link state tracking and TX/RX MAC configuration. o Do not report link status if driver is not running. o TX/RX MAC configuration should be done with resolved speed, duplex and flow control after establishing a link so it can't be done in driver initialization routine. Move the configuration to miibus_statchg callback which will be called whenever any link state change is detected. At this moment, flow-control is not enabled yet mainly because I was not able to set correct flow control parameters to generate TX pause frames. o Now TX/RX MAC is enabled only when a valid link is detected. Rearragnge hardware initialization routine a bit to leave enabling MAC to miibus_statchg callback. In order to that, TX/RX DMA engine is enabled in et_init_locked(). o Introduce ET_FLAG_LINK flag to track current link state. o Introduce ET_FLAG_FASTETHER flag to mark whether controller is fast ethernet. This flag is checked in miibus_statchg callback to know whether PHY established a valid link. o In et_stop(), TX/RX MAC is explicitly disabled instead of relying on et_reset(). And move et_reset() from et_stop() to controller initialization. Controler reset is not required here and it would also clear critial registers(i.e station address, RX filter configuration, WOL etc) that are required to make WOL work. o Switching to current media is done in et_init_locked() after setting IFF_DRV_RUNNING flag. This should ensure reliable auto-negotiation/manual link establishment. o In et_start_locked(), check whether driver got a valid link before trying to send frames. o Remove checking a link in et_tick() as this is done by miibus_statchg callback.
2011-12-07 21:29:51 +00:00
if (IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_RXPAUSE)
cfg1 |= ET_MAC_CFG1_RXFLOW;
} else
ctrl |= ET_MAC_CTRL_GHDX;
CSR_WRITE_4(sc, ET_MAC_CTRL, ctrl);
CSR_WRITE_4(sc, ET_MAC_CFG2, cfg2);
cfg1 |= ET_MAC_CFG1_TXEN | ET_MAC_CFG1_RXEN;
CSR_WRITE_4(sc, ET_MAC_CFG1, cfg1);
#define NRETRY 50
for (i = 0; i < NRETRY; ++i) {
cfg1 = CSR_READ_4(sc, ET_MAC_CFG1);
if ((cfg1 & (ET_MAC_CFG1_SYNC_TXEN | ET_MAC_CFG1_SYNC_RXEN)) ==
(ET_MAC_CFG1_SYNC_TXEN | ET_MAC_CFG1_SYNC_RXEN))
break;
DELAY(100);
}
if (i == NRETRY)
if_printf(ifp, "can't enable RX/TX\n");
sc->sc_flags |= ET_FLAG_TXRX_ENABLED;
#undef NRETRY
}
static int
et_ifmedia_upd_locked(struct ifnet *ifp)
{
2012-01-10 20:52:02 +00:00
struct et_softc *sc;
struct mii_data *mii;
- Remove attempts to implement setting of BMCR_LOOP/MIIF_NOLOOP (reporting IFM_LOOP based on BMCR_LOOP is left in place though as it might provide useful for debugging). For most mii(4) drivers it was unclear whether the PHYs driven by them actually support loopback or not. Moreover, typically loopback mode also needs to be activated on the MAC, which none of the Ethernet drivers using mii(4) implements. Given that loopback media has no real use (and obviously hardly had a chance to actually work) besides for driver development (which just loopback mode should be sufficient for though, i.e one doesn't necessary need support for loopback media) support for it is just dropped as both NetBSD and OpenBSD already did quite some time ago. - Let mii_phy_add_media() also announce the support of IFM_NONE. - Restructure the PHY entry points to use a structure of entry points instead of discrete function pointers, and extend this to include a "reset" entry point. Make sure any PHY-specific reset routine is always used, and provide one for lxtphy(4) which disables MII interrupts (as is done for a few other PHYs we have drivers for). This includes changing NIC drivers which previously just called the generic mii_phy_reset() to now actually call the PHY-specific reset routine, which might be crucial in some cases. While at it, the redundant checks in these NIC drivers for mii->mii_instance not being zero before calling the reset routines were removed because as soon as one PHY driver attaches mii->mii_instance is incremented and we hardly can end up in their media change callbacks etc if no PHY driver has attached as mii_attach() would have failed in that case and not attach a miibus(4) instance. Consequently, NIC drivers now no longer should call mii_phy_reset() directly, so it was removed from EXPORT_SYMS. - Add a mii_phy_dev_attach() as a companion helper to mii_phy_dev_probe(). The purpose of that function is to perform the common steps to attach a PHY driver instance and to hook it up to the miibus(4) instance and to optionally also handle the probing, addition and initialization of the supported media. So all a PHY driver without any special requirements has to do in its bus attach method is to call mii_phy_dev_attach() along with PHY-specific MIIF_* flags, a pointer to its PHY functions and the add_media set to one. All PHY drivers were updated to take advantage of mii_phy_dev_attach() as appropriate. Along with these changes the capability mask was added to the mii_softc structure so PHY drivers taking advantage of mii_phy_dev_attach() but still handling media on their own do not need to fiddle with the MII attach arguments anyway. - Keep track of the PHY offset in the mii_softc structure. This is done for compatibility with NetBSD/OpenBSD. - Keep track of the PHY's OUI, model and revision in the mii_softc structure. Several PHY drivers require this information also after attaching and previously had to wrap their own softc around mii_softc. NetBSD/OpenBSD also keep track of the model and revision on their mii_softc structure. All PHY drivers were updated to take advantage as appropriate. - Convert the mebers of the MII data structure to unsigned where appropriate. This is partly inspired by NetBSD/OpenBSD. - According to IEEE 802.3-2002 the bits actually have to be reversed when mapping an OUI to the MII ID registers. All PHY drivers and miidevs where changed as necessary. Actually this now again allows to largely share miidevs with NetBSD, which fixed this problem already 9 years ago. Consequently miidevs was synced as far as possible. - Add MIIF_NOMANPAUSE and mii_phy_flowstatus() calls to drivers that weren't explicitly converted to support flow control before. It's unclear whether flow control actually works with these but typically it should and their net behavior should be more correct with these changes in place than without if the MAC driver sets MIIF_DOPAUSE. Obtained from: NetBSD (partially) Reviewed by: yongari (earlier version), silence on arch@ and net@
2011-05-03 19:51:29 +00:00
struct mii_softc *miisc;
2012-01-10 20:52:02 +00:00
sc = ifp->if_softc;
mii = device_get_softc(sc->sc_miibus);
- Remove attempts to implement setting of BMCR_LOOP/MIIF_NOLOOP (reporting IFM_LOOP based on BMCR_LOOP is left in place though as it might provide useful for debugging). For most mii(4) drivers it was unclear whether the PHYs driven by them actually support loopback or not. Moreover, typically loopback mode also needs to be activated on the MAC, which none of the Ethernet drivers using mii(4) implements. Given that loopback media has no real use (and obviously hardly had a chance to actually work) besides for driver development (which just loopback mode should be sufficient for though, i.e one doesn't necessary need support for loopback media) support for it is just dropped as both NetBSD and OpenBSD already did quite some time ago. - Let mii_phy_add_media() also announce the support of IFM_NONE. - Restructure the PHY entry points to use a structure of entry points instead of discrete function pointers, and extend this to include a "reset" entry point. Make sure any PHY-specific reset routine is always used, and provide one for lxtphy(4) which disables MII interrupts (as is done for a few other PHYs we have drivers for). This includes changing NIC drivers which previously just called the generic mii_phy_reset() to now actually call the PHY-specific reset routine, which might be crucial in some cases. While at it, the redundant checks in these NIC drivers for mii->mii_instance not being zero before calling the reset routines were removed because as soon as one PHY driver attaches mii->mii_instance is incremented and we hardly can end up in their media change callbacks etc if no PHY driver has attached as mii_attach() would have failed in that case and not attach a miibus(4) instance. Consequently, NIC drivers now no longer should call mii_phy_reset() directly, so it was removed from EXPORT_SYMS. - Add a mii_phy_dev_attach() as a companion helper to mii_phy_dev_probe(). The purpose of that function is to perform the common steps to attach a PHY driver instance and to hook it up to the miibus(4) instance and to optionally also handle the probing, addition and initialization of the supported media. So all a PHY driver without any special requirements has to do in its bus attach method is to call mii_phy_dev_attach() along with PHY-specific MIIF_* flags, a pointer to its PHY functions and the add_media set to one. All PHY drivers were updated to take advantage of mii_phy_dev_attach() as appropriate. Along with these changes the capability mask was added to the mii_softc structure so PHY drivers taking advantage of mii_phy_dev_attach() but still handling media on their own do not need to fiddle with the MII attach arguments anyway. - Keep track of the PHY offset in the mii_softc structure. This is done for compatibility with NetBSD/OpenBSD. - Keep track of the PHY's OUI, model and revision in the mii_softc structure. Several PHY drivers require this information also after attaching and previously had to wrap their own softc around mii_softc. NetBSD/OpenBSD also keep track of the model and revision on their mii_softc structure. All PHY drivers were updated to take advantage as appropriate. - Convert the mebers of the MII data structure to unsigned where appropriate. This is partly inspired by NetBSD/OpenBSD. - According to IEEE 802.3-2002 the bits actually have to be reversed when mapping an OUI to the MII ID registers. All PHY drivers and miidevs where changed as necessary. Actually this now again allows to largely share miidevs with NetBSD, which fixed this problem already 9 years ago. Consequently miidevs was synced as far as possible. - Add MIIF_NOMANPAUSE and mii_phy_flowstatus() calls to drivers that weren't explicitly converted to support flow control before. It's unclear whether flow control actually works with these but typically it should and their net behavior should be more correct with these changes in place than without if the MAC driver sets MIIF_DOPAUSE. Obtained from: NetBSD (partially) Reviewed by: yongari (earlier version), silence on arch@ and net@
2011-05-03 19:51:29 +00:00
LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
PHY_RESET(miisc);
return (mii_mediachg(mii));
}
static int
et_ifmedia_upd(struct ifnet *ifp)
{
2012-01-10 20:52:02 +00:00
struct et_softc *sc;
int res;
2012-01-10 20:52:02 +00:00
sc = ifp->if_softc;
ET_LOCK(sc);
res = et_ifmedia_upd_locked(ifp);
ET_UNLOCK(sc);
2009-11-19 21:53:21 +00:00
return (res);
}
static void
et_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
{
Rework link state tracking and TX/RX MAC configuration. o Do not report link status if driver is not running. o TX/RX MAC configuration should be done with resolved speed, duplex and flow control after establishing a link so it can't be done in driver initialization routine. Move the configuration to miibus_statchg callback which will be called whenever any link state change is detected. At this moment, flow-control is not enabled yet mainly because I was not able to set correct flow control parameters to generate TX pause frames. o Now TX/RX MAC is enabled only when a valid link is detected. Rearragnge hardware initialization routine a bit to leave enabling MAC to miibus_statchg callback. In order to that, TX/RX DMA engine is enabled in et_init_locked(). o Introduce ET_FLAG_LINK flag to track current link state. o Introduce ET_FLAG_FASTETHER flag to mark whether controller is fast ethernet. This flag is checked in miibus_statchg callback to know whether PHY established a valid link. o In et_stop(), TX/RX MAC is explicitly disabled instead of relying on et_reset(). And move et_reset() from et_stop() to controller initialization. Controler reset is not required here and it would also clear critial registers(i.e station address, RX filter configuration, WOL etc) that are required to make WOL work. o Switching to current media is done in et_init_locked() after setting IFF_DRV_RUNNING flag. This should ensure reliable auto-negotiation/manual link establishment. o In et_start_locked(), check whether driver got a valid link before trying to send frames. o Remove checking a link in et_tick() as this is done by miibus_statchg callback.
2011-12-07 21:29:51 +00:00
struct et_softc *sc;
struct mii_data *mii;
Rework link state tracking and TX/RX MAC configuration. o Do not report link status if driver is not running. o TX/RX MAC configuration should be done with resolved speed, duplex and flow control after establishing a link so it can't be done in driver initialization routine. Move the configuration to miibus_statchg callback which will be called whenever any link state change is detected. At this moment, flow-control is not enabled yet mainly because I was not able to set correct flow control parameters to generate TX pause frames. o Now TX/RX MAC is enabled only when a valid link is detected. Rearragnge hardware initialization routine a bit to leave enabling MAC to miibus_statchg callback. In order to that, TX/RX DMA engine is enabled in et_init_locked(). o Introduce ET_FLAG_LINK flag to track current link state. o Introduce ET_FLAG_FASTETHER flag to mark whether controller is fast ethernet. This flag is checked in miibus_statchg callback to know whether PHY established a valid link. o In et_stop(), TX/RX MAC is explicitly disabled instead of relying on et_reset(). And move et_reset() from et_stop() to controller initialization. Controler reset is not required here and it would also clear critial registers(i.e station address, RX filter configuration, WOL etc) that are required to make WOL work. o Switching to current media is done in et_init_locked() after setting IFF_DRV_RUNNING flag. This should ensure reliable auto-negotiation/manual link establishment. o In et_start_locked(), check whether driver got a valid link before trying to send frames. o Remove checking a link in et_tick() as this is done by miibus_statchg callback.
2011-12-07 21:29:51 +00:00
sc = ifp->if_softc;
ET_LOCK(sc);
Rework link state tracking and TX/RX MAC configuration. o Do not report link status if driver is not running. o TX/RX MAC configuration should be done with resolved speed, duplex and flow control after establishing a link so it can't be done in driver initialization routine. Move the configuration to miibus_statchg callback which will be called whenever any link state change is detected. At this moment, flow-control is not enabled yet mainly because I was not able to set correct flow control parameters to generate TX pause frames. o Now TX/RX MAC is enabled only when a valid link is detected. Rearragnge hardware initialization routine a bit to leave enabling MAC to miibus_statchg callback. In order to that, TX/RX DMA engine is enabled in et_init_locked(). o Introduce ET_FLAG_LINK flag to track current link state. o Introduce ET_FLAG_FASTETHER flag to mark whether controller is fast ethernet. This flag is checked in miibus_statchg callback to know whether PHY established a valid link. o In et_stop(), TX/RX MAC is explicitly disabled instead of relying on et_reset(). And move et_reset() from et_stop() to controller initialization. Controler reset is not required here and it would also clear critial registers(i.e station address, RX filter configuration, WOL etc) that are required to make WOL work. o Switching to current media is done in et_init_locked() after setting IFF_DRV_RUNNING flag. This should ensure reliable auto-negotiation/manual link establishment. o In et_start_locked(), check whether driver got a valid link before trying to send frames. o Remove checking a link in et_tick() as this is done by miibus_statchg callback.
2011-12-07 21:29:51 +00:00
if ((ifp->if_flags & IFF_UP) == 0) {
ET_UNLOCK(sc);
return;
}
mii = device_get_softc(sc->sc_miibus);
mii_pollstat(mii);
ifmr->ifm_active = mii->mii_media_active;
ifmr->ifm_status = mii->mii_media_status;
ET_UNLOCK(sc);
}
static void
et_stop(struct et_softc *sc)
{
2012-01-10 20:52:02 +00:00
struct ifnet *ifp;
ET_LOCK_ASSERT(sc);
2012-01-10 20:52:02 +00:00
ifp = sc->ifp;
callout_stop(&sc->sc_tick);
/* Disable interrupts. */
CSR_WRITE_4(sc, ET_INTR_MASK, 0xffffffff);
Rework link state tracking and TX/RX MAC configuration. o Do not report link status if driver is not running. o TX/RX MAC configuration should be done with resolved speed, duplex and flow control after establishing a link so it can't be done in driver initialization routine. Move the configuration to miibus_statchg callback which will be called whenever any link state change is detected. At this moment, flow-control is not enabled yet mainly because I was not able to set correct flow control parameters to generate TX pause frames. o Now TX/RX MAC is enabled only when a valid link is detected. Rearragnge hardware initialization routine a bit to leave enabling MAC to miibus_statchg callback. In order to that, TX/RX DMA engine is enabled in et_init_locked(). o Introduce ET_FLAG_LINK flag to track current link state. o Introduce ET_FLAG_FASTETHER flag to mark whether controller is fast ethernet. This flag is checked in miibus_statchg callback to know whether PHY established a valid link. o In et_stop(), TX/RX MAC is explicitly disabled instead of relying on et_reset(). And move et_reset() from et_stop() to controller initialization. Controler reset is not required here and it would also clear critial registers(i.e station address, RX filter configuration, WOL etc) that are required to make WOL work. o Switching to current media is done in et_init_locked() after setting IFF_DRV_RUNNING flag. This should ensure reliable auto-negotiation/manual link establishment. o In et_start_locked(), check whether driver got a valid link before trying to send frames. o Remove checking a link in et_tick() as this is done by miibus_statchg callback.
2011-12-07 21:29:51 +00:00
CSR_WRITE_4(sc, ET_MAC_CFG1, CSR_READ_4(sc, ET_MAC_CFG1) & ~(
ET_MAC_CFG1_TXEN | ET_MAC_CFG1_RXEN));
DELAY(100);
et_stop_rxdma(sc);
et_stop_txdma(sc);
et_stats_update(sc);
et_free_tx_ring(sc);
et_free_rx_ring(sc);
sc->sc_tx = 0;
sc->sc_tx_intr = 0;
sc->sc_flags &= ~ET_FLAG_TXRX_ENABLED;
sc->watchdog_timer = 0;
ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
}
static int
et_bus_config(struct et_softc *sc)
{
uint32_t val, max_plsz;
uint16_t ack_latency, replay_timer;
/*
* Test whether EEPROM is valid
* NOTE: Read twice to get the correct value
*/
pci_read_config(sc->dev, ET_PCIR_EEPROM_STATUS, 1);
val = pci_read_config(sc->dev, ET_PCIR_EEPROM_STATUS, 1);
if (val & ET_PCIM_EEPROM_STATUS_ERROR) {
device_printf(sc->dev, "EEPROM status error 0x%02x\n", val);
2009-11-19 21:53:21 +00:00
return (ENXIO);
}
/* TODO: LED */
if ((sc->sc_flags & ET_FLAG_PCIE) == 0)
return (0);
/*
* Configure ACK latency and replay timer according to
* max playload size
*/
val = pci_read_config(sc->dev,
sc->sc_expcap + PCIER_DEVICE_CAP, 4);
max_plsz = val & PCIEM_CAP_MAX_PAYLOAD;
switch (max_plsz) {
case ET_PCIV_DEVICE_CAPS_PLSZ_128:
ack_latency = ET_PCIV_ACK_LATENCY_128;
replay_timer = ET_PCIV_REPLAY_TIMER_128;
break;
case ET_PCIV_DEVICE_CAPS_PLSZ_256:
ack_latency = ET_PCIV_ACK_LATENCY_256;
replay_timer = ET_PCIV_REPLAY_TIMER_256;
break;
default:
ack_latency = pci_read_config(sc->dev, ET_PCIR_ACK_LATENCY, 2);
replay_timer = pci_read_config(sc->dev,
ET_PCIR_REPLAY_TIMER, 2);
device_printf(sc->dev, "ack latency %u, replay timer %u\n",
ack_latency, replay_timer);
break;
}
if (ack_latency != 0) {
pci_write_config(sc->dev, ET_PCIR_ACK_LATENCY, ack_latency, 2);
pci_write_config(sc->dev, ET_PCIR_REPLAY_TIMER, replay_timer,
2);
}
/*
* Set L0s and L1 latency timer to 2us
*/
val = pci_read_config(sc->dev, ET_PCIR_L0S_L1_LATENCY, 4);
val &= ~(PCIEM_LINK_CAP_L0S_EXIT | PCIEM_LINK_CAP_L1_EXIT);
/* L0s exit latency : 2us */
val |= 0x00005000;
/* L1 exit latency : 2us */
val |= 0x00028000;
pci_write_config(sc->dev, ET_PCIR_L0S_L1_LATENCY, val, 4);
/*
* Set max read request size to 2048 bytes
*/
pci_set_max_read_req(sc->dev, 2048);
2009-11-19 21:53:21 +00:00
return (0);
}
static void
et_get_eaddr(device_t dev, uint8_t eaddr[])
{
uint32_t val;
int i;
val = pci_read_config(dev, ET_PCIR_MAC_ADDR0, 4);
for (i = 0; i < 4; ++i)
eaddr[i] = (val >> (8 * i)) & 0xff;
val = pci_read_config(dev, ET_PCIR_MAC_ADDR1, 2);
for (; i < ETHER_ADDR_LEN; ++i)
eaddr[i] = (val >> (8 * (i - 4))) & 0xff;
}
static void
et_reset(struct et_softc *sc)
{
2012-01-10 20:52:02 +00:00
CSR_WRITE_4(sc, ET_MAC_CFG1,
ET_MAC_CFG1_RST_TXFUNC | ET_MAC_CFG1_RST_RXFUNC |
ET_MAC_CFG1_RST_TXMC | ET_MAC_CFG1_RST_RXMC |
ET_MAC_CFG1_SIM_RST | ET_MAC_CFG1_SOFT_RST);
CSR_WRITE_4(sc, ET_SWRST,
ET_SWRST_TXDMA | ET_SWRST_RXDMA |
ET_SWRST_TXMAC | ET_SWRST_RXMAC |
ET_SWRST_MAC | ET_SWRST_MAC_STAT | ET_SWRST_MMC);
CSR_WRITE_4(sc, ET_MAC_CFG1,
ET_MAC_CFG1_RST_TXFUNC | ET_MAC_CFG1_RST_RXFUNC |
ET_MAC_CFG1_RST_TXMC | ET_MAC_CFG1_RST_RXMC);
CSR_WRITE_4(sc, ET_MAC_CFG1, 0);
/* Disable interrupts. */
CSR_WRITE_4(sc, ET_INTR_MASK, 0xffffffff);
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
struct et_dmamap_arg {
bus_addr_t et_busaddr;
};
static void
et_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error)
{
struct et_dmamap_arg *ctx;
if (error)
return;
KASSERT(nseg == 1, ("%s: %d segments returned!", __func__, nseg));
ctx = arg;
ctx->et_busaddr = segs->ds_addr;
}
static int
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
et_dma_ring_alloc(struct et_softc *sc, bus_size_t alignment, bus_size_t maxsize,
bus_dma_tag_t *tag, uint8_t **ring, bus_dmamap_t *map, bus_addr_t *paddr,
const char *msg)
{
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
struct et_dmamap_arg ctx;
int error;
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
error = bus_dma_tag_create(sc->sc_dtag, alignment, 0, BUS_SPACE_MAXADDR,
BUS_SPACE_MAXADDR, NULL, NULL, maxsize, 1, maxsize, 0, NULL, NULL,
tag);
if (error != 0) {
device_printf(sc->dev, "could not create %s dma tag\n", msg);
2009-11-19 21:53:21 +00:00
return (error);
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* Allocate DMA'able memory for ring. */
error = bus_dmamem_alloc(*tag, (void **)ring,
BUS_DMA_NOWAIT | BUS_DMA_ZERO | BUS_DMA_COHERENT, map);
if (error != 0) {
device_printf(sc->dev,
"could not allocate DMA'able memory for %s\n", msg);
2009-11-19 21:53:21 +00:00
return (error);
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* Load the address of the ring. */
ctx.et_busaddr = 0;
error = bus_dmamap_load(*tag, *map, *ring, maxsize, et_dma_map_addr,
&ctx, BUS_DMA_NOWAIT);
if (error != 0) {
device_printf(sc->dev,
"could not load DMA'able memory for %s\n", msg);
2009-11-19 21:53:21 +00:00
return (error);
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
*paddr = ctx.et_busaddr;
return (0);
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
static void
et_dma_ring_free(struct et_softc *sc, bus_dma_tag_t *tag, uint8_t **ring,
bus_dmamap_t map, bus_addr_t *paddr)
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
{
if (*paddr != 0) {
bus_dmamap_unload(*tag, map);
*paddr = 0;
}
if (*ring != NULL) {
bus_dmamem_free(*tag, *ring, map);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
*ring = NULL;
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
if (*tag) {
bus_dma_tag_destroy(*tag);
*tag = NULL;
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
static int
et_dma_alloc(struct et_softc *sc)
{
struct et_txdesc_ring *tx_ring;
struct et_rxdesc_ring *rx_ring;
struct et_rxstat_ring *rxst_ring;
struct et_rxstatus_data *rxsd;
struct et_rxbuf_data *rbd;
struct et_txbuf_data *tbd;
struct et_txstatus_data *txsd;
int i, error;
error = bus_dma_tag_create(bus_get_dma_tag(sc->dev), 1, 0,
BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0, NULL, NULL,
&sc->sc_dtag);
if (error != 0) {
device_printf(sc->dev, "could not allocate parent dma tag\n");
2009-11-19 21:53:21 +00:00
return (error);
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* TX ring. */
tx_ring = &sc->sc_tx_ring;
error = et_dma_ring_alloc(sc, ET_RING_ALIGN, ET_TX_RING_SIZE,
&tx_ring->tr_dtag, (uint8_t **)&tx_ring->tr_desc, &tx_ring->tr_dmap,
&tx_ring->tr_paddr, "TX ring");
if (error)
2009-11-19 21:53:21 +00:00
return (error);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* TX status block. */
txsd = &sc->sc_tx_status;
error = et_dma_ring_alloc(sc, ET_STATUS_ALIGN, sizeof(uint32_t),
&txsd->txsd_dtag, (uint8_t **)&txsd->txsd_status, &txsd->txsd_dmap,
&txsd->txsd_paddr, "TX status block");
if (error)
return (error);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* RX ring 0, used as to recive small sized frames. */
rx_ring = &sc->sc_rx_ring[0];
error = et_dma_ring_alloc(sc, ET_RING_ALIGN, ET_RX_RING_SIZE,
&rx_ring->rr_dtag, (uint8_t **)&rx_ring->rr_desc, &rx_ring->rr_dmap,
&rx_ring->rr_paddr, "RX ring 0");
rx_ring->rr_posreg = ET_RX_RING0_POS;
if (error)
return (error);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* RX ring 1, used as to store normal sized frames. */
rx_ring = &sc->sc_rx_ring[1];
error = et_dma_ring_alloc(sc, ET_RING_ALIGN, ET_RX_RING_SIZE,
&rx_ring->rr_dtag, (uint8_t **)&rx_ring->rr_desc, &rx_ring->rr_dmap,
&rx_ring->rr_paddr, "RX ring 1");
rx_ring->rr_posreg = ET_RX_RING1_POS;
if (error)
return (error);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* RX stat ring. */
rxst_ring = &sc->sc_rxstat_ring;
error = et_dma_ring_alloc(sc, ET_RING_ALIGN, ET_RXSTAT_RING_SIZE,
&rxst_ring->rsr_dtag, (uint8_t **)&rxst_ring->rsr_stat,
&rxst_ring->rsr_dmap, &rxst_ring->rsr_paddr, "RX stat ring");
if (error)
return (error);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* RX status block. */
rxsd = &sc->sc_rx_status;
error = et_dma_ring_alloc(sc, ET_STATUS_ALIGN,
sizeof(struct et_rxstatus), &rxsd->rxsd_dtag,
(uint8_t **)&rxsd->rxsd_status, &rxsd->rxsd_dmap,
&rxsd->rxsd_paddr, "RX status block");
if (error)
return (error);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* Create parent DMA tag for mbufs. */
error = bus_dma_tag_create(bus_get_dma_tag(sc->dev), 1, 0,
BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0, NULL, NULL,
&sc->sc_mbuf_dtag);
if (error != 0) {
device_printf(sc->dev,
"could not allocate parent dma tag for mbuf\n");
return (error);
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* Create DMA tag for mini RX mbufs to use RX ring 0. */
error = bus_dma_tag_create(sc->sc_mbuf_dtag, 1, 0,
BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL, MHLEN, 1,
MHLEN, 0, NULL, NULL, &sc->sc_rx_mini_tag);
if (error) {
device_printf(sc->dev, "could not create mini RX dma tag\n");
return (error);
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* Create DMA tag for standard RX mbufs to use RX ring 1. */
error = bus_dma_tag_create(sc->sc_mbuf_dtag, 1, 0,
BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL, MCLBYTES, 1,
MCLBYTES, 0, NULL, NULL, &sc->sc_rx_tag);
if (error) {
device_printf(sc->dev, "could not create RX dma tag\n");
return (error);
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* Create DMA tag for TX mbufs. */
error = bus_dma_tag_create(sc->sc_mbuf_dtag, 1, 0,
BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
MCLBYTES * ET_NSEG_MAX, ET_NSEG_MAX, MCLBYTES, 0, NULL, NULL,
&sc->sc_tx_tag);
if (error) {
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
device_printf(sc->dev, "could not create TX dma tag\n");
2009-11-19 21:53:21 +00:00
return (error);
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* Initialize RX ring 0. */
rbd = &sc->sc_rx_data[0];
rbd->rbd_bufsize = ET_RXDMA_CTRL_RING0_128;
rbd->rbd_newbuf = et_newbuf_hdr;
rbd->rbd_discard = et_rxbuf_discard;
rbd->rbd_softc = sc;
rbd->rbd_ring = &sc->sc_rx_ring[0];
/* Create DMA maps for mini RX buffers, ring 0. */
for (i = 0; i < ET_RX_NDESC; i++) {
error = bus_dmamap_create(sc->sc_rx_mini_tag, 0,
&rbd->rbd_buf[i].rb_dmap);
if (error) {
device_printf(sc->dev,
"could not create DMA map for mini RX mbufs\n");
return (error);
}
}
/* Create a spare DMA map for mini RX buffers, ring 0. */
error = bus_dmamap_create(sc->sc_rx_mini_tag, 0,
&sc->sc_rx_mini_sparemap);
if (error) {
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
device_printf(sc->dev,
"could not create spare DMA map for mini RX mbuf\n");
2009-11-19 21:53:21 +00:00
return (error);
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* Initialize RX ring 1. */
rbd = &sc->sc_rx_data[1];
rbd->rbd_bufsize = ET_RXDMA_CTRL_RING1_2048;
rbd->rbd_newbuf = et_newbuf_cluster;
rbd->rbd_discard = et_rxbuf_discard;
rbd->rbd_softc = sc;
rbd->rbd_ring = &sc->sc_rx_ring[1];
/* Create DMA maps for standard RX buffers, ring 1. */
for (i = 0; i < ET_RX_NDESC; i++) {
error = bus_dmamap_create(sc->sc_rx_tag, 0,
&rbd->rbd_buf[i].rb_dmap);
if (error) {
device_printf(sc->dev,
"could not create DMA map for mini RX mbufs\n");
return (error);
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* Create a spare DMA map for standard RX buffers, ring 1. */
error = bus_dmamap_create(sc->sc_rx_tag, 0, &sc->sc_rx_sparemap);
if (error) {
device_printf(sc->dev,
"could not create spare DMA map for RX mbuf\n");
return (error);
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* Create DMA maps for TX buffers. */
tbd = &sc->sc_tx_data;
for (i = 0; i < ET_TX_NDESC; i++) {
error = bus_dmamap_create(sc->sc_tx_tag, 0,
&tbd->tbd_buf[i].tb_dmap);
if (error) {
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
device_printf(sc->dev,
"could not create DMA map for TX mbufs\n");
2009-11-19 21:53:21 +00:00
return (error);
}
}
2009-11-19 21:53:21 +00:00
return (0);
}
static void
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
et_dma_free(struct et_softc *sc)
{
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
struct et_txdesc_ring *tx_ring;
struct et_rxdesc_ring *rx_ring;
struct et_txstatus_data *txsd;
struct et_rxstat_ring *rxst_ring;
struct et_rxstatus_data *rxsd;
struct et_rxbuf_data *rbd;
struct et_txbuf_data *tbd;
int i;
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* Destroy DMA maps for mini RX buffers, ring 0. */
rbd = &sc->sc_rx_data[0];
for (i = 0; i < ET_RX_NDESC; i++) {
if (rbd->rbd_buf[i].rb_dmap) {
bus_dmamap_destroy(sc->sc_rx_mini_tag,
rbd->rbd_buf[i].rb_dmap);
rbd->rbd_buf[i].rb_dmap = NULL;
}
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
if (sc->sc_rx_mini_sparemap) {
bus_dmamap_destroy(sc->sc_rx_mini_tag, sc->sc_rx_mini_sparemap);
sc->sc_rx_mini_sparemap = NULL;
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
if (sc->sc_rx_mini_tag) {
bus_dma_tag_destroy(sc->sc_rx_mini_tag);
sc->sc_rx_mini_tag = NULL;
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* Destroy DMA maps for standard RX buffers, ring 1. */
rbd = &sc->sc_rx_data[1];
for (i = 0; i < ET_RX_NDESC; i++) {
if (rbd->rbd_buf[i].rb_dmap) {
bus_dmamap_destroy(sc->sc_rx_tag,
rbd->rbd_buf[i].rb_dmap);
rbd->rbd_buf[i].rb_dmap = NULL;
}
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
if (sc->sc_rx_sparemap) {
bus_dmamap_destroy(sc->sc_rx_tag, sc->sc_rx_sparemap);
sc->sc_rx_sparemap = NULL;
}
if (sc->sc_rx_tag) {
bus_dma_tag_destroy(sc->sc_rx_tag);
sc->sc_rx_tag = NULL;
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* Destroy DMA maps for TX buffers. */
tbd = &sc->sc_tx_data;
for (i = 0; i < ET_TX_NDESC; i++) {
if (tbd->tbd_buf[i].tb_dmap) {
bus_dmamap_destroy(sc->sc_tx_tag,
tbd->tbd_buf[i].tb_dmap);
tbd->tbd_buf[i].tb_dmap = NULL;
}
}
if (sc->sc_tx_tag) {
bus_dma_tag_destroy(sc->sc_tx_tag);
sc->sc_tx_tag = NULL;
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* Destroy mini RX ring, ring 0. */
rx_ring = &sc->sc_rx_ring[0];
et_dma_ring_free(sc, &rx_ring->rr_dtag, (void *)&rx_ring->rr_desc,
rx_ring->rr_dmap, &rx_ring->rr_paddr);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* Destroy standard RX ring, ring 1. */
rx_ring = &sc->sc_rx_ring[1];
et_dma_ring_free(sc, &rx_ring->rr_dtag, (void *)&rx_ring->rr_desc,
rx_ring->rr_dmap, &rx_ring->rr_paddr);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* Destroy RX stat ring. */
rxst_ring = &sc->sc_rxstat_ring;
et_dma_ring_free(sc, &rxst_ring->rsr_dtag, (void *)&rxst_ring->rsr_stat,
rxst_ring->rsr_dmap, &rxst_ring->rsr_paddr);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* Destroy RX status block. */
rxsd = &sc->sc_rx_status;
et_dma_ring_free(sc, &rxst_ring->rsr_dtag, (void *)&rxst_ring->rsr_stat,
rxst_ring->rsr_dmap, &rxst_ring->rsr_paddr);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* Destroy TX ring. */
tx_ring = &sc->sc_tx_ring;
et_dma_ring_free(sc, &tx_ring->tr_dtag, (void *)&tx_ring->tr_desc,
tx_ring->tr_dmap, &tx_ring->tr_paddr);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* Destroy TX status block. */
txsd = &sc->sc_tx_status;
et_dma_ring_free(sc, &txsd->txsd_dtag, (void *)&txsd->txsd_status,
txsd->txsd_dmap, &txsd->txsd_paddr);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* Destroy the parent tag. */
if (sc->sc_dtag) {
bus_dma_tag_destroy(sc->sc_dtag);
sc->sc_dtag = NULL;
}
}
static void
et_chip_attach(struct et_softc *sc)
{
uint32_t val;
/*
* Perform minimal initialization
*/
/* Disable loopback */
CSR_WRITE_4(sc, ET_LOOPBACK, 0);
/* Reset MAC */
CSR_WRITE_4(sc, ET_MAC_CFG1,
ET_MAC_CFG1_RST_TXFUNC | ET_MAC_CFG1_RST_RXFUNC |
ET_MAC_CFG1_RST_TXMC | ET_MAC_CFG1_RST_RXMC |
ET_MAC_CFG1_SIM_RST | ET_MAC_CFG1_SOFT_RST);
/*
* Setup half duplex mode
*/
val = (10 << ET_MAC_HDX_ALT_BEB_TRUNC_SHIFT) |
(15 << ET_MAC_HDX_REXMIT_MAX_SHIFT) |
(55 << ET_MAC_HDX_COLLWIN_SHIFT) |
ET_MAC_HDX_EXC_DEFER;
CSR_WRITE_4(sc, ET_MAC_HDX, val);
/* Clear MAC control */
CSR_WRITE_4(sc, ET_MAC_CTRL, 0);
/* Reset MII */
CSR_WRITE_4(sc, ET_MII_CFG, ET_MII_CFG_CLKRST);
/* Bring MAC out of reset state */
CSR_WRITE_4(sc, ET_MAC_CFG1, 0);
/* Enable memory controllers */
CSR_WRITE_4(sc, ET_MMC_CTRL, ET_MMC_CTRL_ENABLE);
}
static void
et_intr(void *xsc)
{
2012-01-10 20:52:02 +00:00
struct et_softc *sc;
struct ifnet *ifp;
uint32_t status;
2012-01-10 20:52:02 +00:00
sc = xsc;
ET_LOCK(sc);
ifp = sc->ifp;
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
goto done;
status = CSR_READ_4(sc, ET_INTR_STATUS);
if ((status & ET_INTRS) == 0)
goto done;
/* Disable further interrupts. */
CSR_WRITE_4(sc, ET_INTR_MASK, 0xffffffff);
if (status & (ET_INTR_RXDMA_ERROR | ET_INTR_TXDMA_ERROR)) {
device_printf(sc->dev, "DMA error(0x%08x) -- resetting\n",
status);
ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
et_init_locked(sc);
ET_UNLOCK(sc);
return;
}
if (status & ET_INTR_RXDMA)
et_rxeof(sc);
if (status & (ET_INTR_TXDMA | ET_INTR_TIMER))
et_txeof(sc);
if (status & ET_INTR_TIMER)
CSR_WRITE_4(sc, ET_TIMER, sc->sc_timer);
if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
CSR_WRITE_4(sc, ET_INTR_MASK, ~ET_INTRS);
if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
et_start_locked(ifp);
}
done:
ET_UNLOCK(sc);
}
static void
et_init_locked(struct et_softc *sc)
{
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
struct ifnet *ifp;
int error;
ET_LOCK_ASSERT(sc);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
ifp = sc->ifp;
if (ifp->if_drv_flags & IFF_DRV_RUNNING)
return;
et_stop(sc);
Rework link state tracking and TX/RX MAC configuration. o Do not report link status if driver is not running. o TX/RX MAC configuration should be done with resolved speed, duplex and flow control after establishing a link so it can't be done in driver initialization routine. Move the configuration to miibus_statchg callback which will be called whenever any link state change is detected. At this moment, flow-control is not enabled yet mainly because I was not able to set correct flow control parameters to generate TX pause frames. o Now TX/RX MAC is enabled only when a valid link is detected. Rearragnge hardware initialization routine a bit to leave enabling MAC to miibus_statchg callback. In order to that, TX/RX DMA engine is enabled in et_init_locked(). o Introduce ET_FLAG_LINK flag to track current link state. o Introduce ET_FLAG_FASTETHER flag to mark whether controller is fast ethernet. This flag is checked in miibus_statchg callback to know whether PHY established a valid link. o In et_stop(), TX/RX MAC is explicitly disabled instead of relying on et_reset(). And move et_reset() from et_stop() to controller initialization. Controler reset is not required here and it would also clear critial registers(i.e station address, RX filter configuration, WOL etc) that are required to make WOL work. o Switching to current media is done in et_init_locked() after setting IFF_DRV_RUNNING flag. This should ensure reliable auto-negotiation/manual link establishment. o In et_start_locked(), check whether driver got a valid link before trying to send frames. o Remove checking a link in et_tick() as this is done by miibus_statchg callback.
2011-12-07 21:29:51 +00:00
et_reset(sc);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
et_init_tx_ring(sc);
error = et_init_rx_ring(sc);
if (error)
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
return;
error = et_chip_init(sc);
if (error)
Rework link state tracking and TX/RX MAC configuration. o Do not report link status if driver is not running. o TX/RX MAC configuration should be done with resolved speed, duplex and flow control after establishing a link so it can't be done in driver initialization routine. Move the configuration to miibus_statchg callback which will be called whenever any link state change is detected. At this moment, flow-control is not enabled yet mainly because I was not able to set correct flow control parameters to generate TX pause frames. o Now TX/RX MAC is enabled only when a valid link is detected. Rearragnge hardware initialization routine a bit to leave enabling MAC to miibus_statchg callback. In order to that, TX/RX DMA engine is enabled in et_init_locked(). o Introduce ET_FLAG_LINK flag to track current link state. o Introduce ET_FLAG_FASTETHER flag to mark whether controller is fast ethernet. This flag is checked in miibus_statchg callback to know whether PHY established a valid link. o In et_stop(), TX/RX MAC is explicitly disabled instead of relying on et_reset(). And move et_reset() from et_stop() to controller initialization. Controler reset is not required here and it would also clear critial registers(i.e station address, RX filter configuration, WOL etc) that are required to make WOL work. o Switching to current media is done in et_init_locked() after setting IFF_DRV_RUNNING flag. This should ensure reliable auto-negotiation/manual link establishment. o In et_start_locked(), check whether driver got a valid link before trying to send frames. o Remove checking a link in et_tick() as this is done by miibus_statchg callback.
2011-12-07 21:29:51 +00:00
goto fail;
Rework link state tracking and TX/RX MAC configuration. o Do not report link status if driver is not running. o TX/RX MAC configuration should be done with resolved speed, duplex and flow control after establishing a link so it can't be done in driver initialization routine. Move the configuration to miibus_statchg callback which will be called whenever any link state change is detected. At this moment, flow-control is not enabled yet mainly because I was not able to set correct flow control parameters to generate TX pause frames. o Now TX/RX MAC is enabled only when a valid link is detected. Rearragnge hardware initialization routine a bit to leave enabling MAC to miibus_statchg callback. In order to that, TX/RX DMA engine is enabled in et_init_locked(). o Introduce ET_FLAG_LINK flag to track current link state. o Introduce ET_FLAG_FASTETHER flag to mark whether controller is fast ethernet. This flag is checked in miibus_statchg callback to know whether PHY established a valid link. o In et_stop(), TX/RX MAC is explicitly disabled instead of relying on et_reset(). And move et_reset() from et_stop() to controller initialization. Controler reset is not required here and it would also clear critial registers(i.e station address, RX filter configuration, WOL etc) that are required to make WOL work. o Switching to current media is done in et_init_locked() after setting IFF_DRV_RUNNING flag. This should ensure reliable auto-negotiation/manual link establishment. o In et_start_locked(), check whether driver got a valid link before trying to send frames. o Remove checking a link in et_tick() as this is done by miibus_statchg callback.
2011-12-07 21:29:51 +00:00
/*
* Start TX/RX DMA engine
*/
error = et_start_rxdma(sc);
if (error)
Rework link state tracking and TX/RX MAC configuration. o Do not report link status if driver is not running. o TX/RX MAC configuration should be done with resolved speed, duplex and flow control after establishing a link so it can't be done in driver initialization routine. Move the configuration to miibus_statchg callback which will be called whenever any link state change is detected. At this moment, flow-control is not enabled yet mainly because I was not able to set correct flow control parameters to generate TX pause frames. o Now TX/RX MAC is enabled only when a valid link is detected. Rearragnge hardware initialization routine a bit to leave enabling MAC to miibus_statchg callback. In order to that, TX/RX DMA engine is enabled in et_init_locked(). o Introduce ET_FLAG_LINK flag to track current link state. o Introduce ET_FLAG_FASTETHER flag to mark whether controller is fast ethernet. This flag is checked in miibus_statchg callback to know whether PHY established a valid link. o In et_stop(), TX/RX MAC is explicitly disabled instead of relying on et_reset(). And move et_reset() from et_stop() to controller initialization. Controler reset is not required here and it would also clear critial registers(i.e station address, RX filter configuration, WOL etc) that are required to make WOL work. o Switching to current media is done in et_init_locked() after setting IFF_DRV_RUNNING flag. This should ensure reliable auto-negotiation/manual link establishment. o In et_start_locked(), check whether driver got a valid link before trying to send frames. o Remove checking a link in et_tick() as this is done by miibus_statchg callback.
2011-12-07 21:29:51 +00:00
return;
error = et_start_txdma(sc);
if (error)
return;
/* Enable interrupts. */
CSR_WRITE_4(sc, ET_INTR_MASK, ~ET_INTRS);
CSR_WRITE_4(sc, ET_TIMER, sc->sc_timer);
ifp->if_drv_flags |= IFF_DRV_RUNNING;
ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
Rework link state tracking and TX/RX MAC configuration. o Do not report link status if driver is not running. o TX/RX MAC configuration should be done with resolved speed, duplex and flow control after establishing a link so it can't be done in driver initialization routine. Move the configuration to miibus_statchg callback which will be called whenever any link state change is detected. At this moment, flow-control is not enabled yet mainly because I was not able to set correct flow control parameters to generate TX pause frames. o Now TX/RX MAC is enabled only when a valid link is detected. Rearragnge hardware initialization routine a bit to leave enabling MAC to miibus_statchg callback. In order to that, TX/RX DMA engine is enabled in et_init_locked(). o Introduce ET_FLAG_LINK flag to track current link state. o Introduce ET_FLAG_FASTETHER flag to mark whether controller is fast ethernet. This flag is checked in miibus_statchg callback to know whether PHY established a valid link. o In et_stop(), TX/RX MAC is explicitly disabled instead of relying on et_reset(). And move et_reset() from et_stop() to controller initialization. Controler reset is not required here and it would also clear critial registers(i.e station address, RX filter configuration, WOL etc) that are required to make WOL work. o Switching to current media is done in et_init_locked() after setting IFF_DRV_RUNNING flag. This should ensure reliable auto-negotiation/manual link establishment. o In et_start_locked(), check whether driver got a valid link before trying to send frames. o Remove checking a link in et_tick() as this is done by miibus_statchg callback.
2011-12-07 21:29:51 +00:00
sc->sc_flags &= ~ET_FLAG_LINK;
et_ifmedia_upd_locked(ifp);
callout_reset(&sc->sc_tick, hz, et_tick, sc);
fail:
if (error)
et_stop(sc);
}
static void
et_init(void *xsc)
{
struct et_softc *sc = xsc;
ET_LOCK(sc);
et_init_locked(sc);
ET_UNLOCK(sc);
}
static int
et_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
{
2012-01-10 20:52:02 +00:00
struct et_softc *sc;
struct mii_data *mii;
struct ifreq *ifr;
int error, mask, max_framelen;
sc = ifp->if_softc;
ifr = (struct ifreq *)data;
error = 0;
/* XXX LOCKSUSED */
switch (cmd) {
case SIOCSIFFLAGS:
ET_LOCK(sc);
if (ifp->if_flags & IFF_UP) {
if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
if ((ifp->if_flags ^ sc->sc_if_flags) &
(IFF_ALLMULTI | IFF_PROMISC | IFF_BROADCAST))
et_setmulti(sc);
} else {
et_init_locked(sc);
}
} else {
if (ifp->if_drv_flags & IFF_DRV_RUNNING)
et_stop(sc);
}
sc->sc_if_flags = ifp->if_flags;
ET_UNLOCK(sc);
break;
case SIOCSIFMEDIA:
case SIOCGIFMEDIA:
2012-01-10 20:52:02 +00:00
mii = device_get_softc(sc->sc_miibus);
error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, cmd);
break;
case SIOCADDMULTI:
case SIOCDELMULTI:
if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
ET_LOCK(sc);
et_setmulti(sc);
ET_UNLOCK(sc);
}
break;
case SIOCSIFMTU:
ET_LOCK(sc);
#if 0
if (sc->sc_flags & ET_FLAG_JUMBO)
max_framelen = ET_JUMBO_FRAMELEN;
else
#endif
max_framelen = MCLBYTES - 1;
if (ET_FRAMELEN(ifr->ifr_mtu) > max_framelen) {
error = EOPNOTSUPP;
ET_UNLOCK(sc);
break;
}
if (ifp->if_mtu != ifr->ifr_mtu) {
ifp->if_mtu = ifr->ifr_mtu;
if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
et_init_locked(sc);
}
}
ET_UNLOCK(sc);
break;
case SIOCSIFCAP:
ET_LOCK(sc);
mask = ifr->ifr_reqcap ^ ifp->if_capenable;
if ((mask & IFCAP_TXCSUM) != 0 &&
(IFCAP_TXCSUM & ifp->if_capabilities) != 0) {
ifp->if_capenable ^= IFCAP_TXCSUM;
if ((IFCAP_TXCSUM & ifp->if_capenable) != 0)
ifp->if_hwassist |= ET_CSUM_FEATURES;
else
ifp->if_hwassist &= ~ET_CSUM_FEATURES;
}
ET_UNLOCK(sc);
break;
default:
error = ether_ioctl(ifp, cmd, data);
break;
}
2009-11-19 21:53:21 +00:00
return (error);
}
static void
et_start_locked(struct ifnet *ifp)
{
struct et_softc *sc;
struct mbuf *m_head = NULL;
struct et_txdesc_ring *tx_ring;
struct et_txbuf_data *tbd;
uint32_t tx_ready_pos;
int enq;
sc = ifp->if_softc;
ET_LOCK_ASSERT(sc);
Rework link state tracking and TX/RX MAC configuration. o Do not report link status if driver is not running. o TX/RX MAC configuration should be done with resolved speed, duplex and flow control after establishing a link so it can't be done in driver initialization routine. Move the configuration to miibus_statchg callback which will be called whenever any link state change is detected. At this moment, flow-control is not enabled yet mainly because I was not able to set correct flow control parameters to generate TX pause frames. o Now TX/RX MAC is enabled only when a valid link is detected. Rearragnge hardware initialization routine a bit to leave enabling MAC to miibus_statchg callback. In order to that, TX/RX DMA engine is enabled in et_init_locked(). o Introduce ET_FLAG_LINK flag to track current link state. o Introduce ET_FLAG_FASTETHER flag to mark whether controller is fast ethernet. This flag is checked in miibus_statchg callback to know whether PHY established a valid link. o In et_stop(), TX/RX MAC is explicitly disabled instead of relying on et_reset(). And move et_reset() from et_stop() to controller initialization. Controler reset is not required here and it would also clear critial registers(i.e station address, RX filter configuration, WOL etc) that are required to make WOL work. o Switching to current media is done in et_init_locked() after setting IFF_DRV_RUNNING flag. This should ensure reliable auto-negotiation/manual link establishment. o In et_start_locked(), check whether driver got a valid link before trying to send frames. o Remove checking a link in et_tick() as this is done by miibus_statchg callback.
2011-12-07 21:29:51 +00:00
if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
IFF_DRV_RUNNING ||
(sc->sc_flags & (ET_FLAG_LINK | ET_FLAG_TXRX_ENABLED)) !=
(ET_FLAG_LINK | ET_FLAG_TXRX_ENABLED))
return;
/*
* Driver does not request TX completion interrupt for every
* queued frames to prevent generating excessive interrupts.
* This means driver may wait for TX completion interrupt even
* though some frames were successfully transmitted. Reclaiming
* transmitted frames will ensure driver see all available
* descriptors.
*/
tbd = &sc->sc_tx_data;
if (tbd->tbd_used > (ET_TX_NDESC * 2) / 3)
et_txeof(sc);
for (enq = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd); ) {
if (tbd->tbd_used + ET_NSEG_SPARE >= ET_TX_NDESC) {
ifp->if_drv_flags |= IFF_DRV_OACTIVE;
break;
}
IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
if (m_head == NULL)
break;
if (et_encap(sc, &m_head)) {
if (m_head == NULL) {
if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
break;
}
IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
if (tbd->tbd_used > 0)
ifp->if_drv_flags |= IFF_DRV_OACTIVE;
break;
}
enq++;
ETHER_BPF_MTAP(ifp, m_head);
}
if (enq > 0) {
tx_ring = &sc->sc_tx_ring;
bus_dmamap_sync(tx_ring->tr_dtag, tx_ring->tr_dmap,
BUS_DMASYNC_PREWRITE);
tx_ready_pos = tx_ring->tr_ready_index &
ET_TX_READY_POS_INDEX_MASK;
if (tx_ring->tr_ready_wrap)
tx_ready_pos |= ET_TX_READY_POS_WRAP;
CSR_WRITE_4(sc, ET_TX_READY_POS, tx_ready_pos);
sc->watchdog_timer = 5;
}
}
static void
et_start(struct ifnet *ifp)
{
2012-01-10 20:52:02 +00:00
struct et_softc *sc;
2012-01-10 20:52:02 +00:00
sc = ifp->if_softc;
ET_LOCK(sc);
et_start_locked(ifp);
ET_UNLOCK(sc);
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
static int
et_watchdog(struct et_softc *sc)
{
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
uint32_t status;
ET_LOCK_ASSERT(sc);
if (sc->watchdog_timer == 0 || --sc->watchdog_timer)
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
return (0);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
bus_dmamap_sync(sc->sc_tx_status.txsd_dtag, sc->sc_tx_status.txsd_dmap,
BUS_DMASYNC_POSTREAD);
status = le32toh(*(sc->sc_tx_status.txsd_status));
if_printf(sc->ifp, "watchdog timed out (0x%08x) -- resetting\n",
status);
if_inc_counter(sc->ifp, IFCOUNTER_OERRORS, 1);
sc->ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
et_init_locked(sc);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
return (EJUSTRETURN);
}
static int
et_stop_rxdma(struct et_softc *sc)
{
2012-01-10 20:52:02 +00:00
CSR_WRITE_4(sc, ET_RXDMA_CTRL,
ET_RXDMA_CTRL_HALT | ET_RXDMA_CTRL_RING1_ENABLE);
DELAY(5);
if ((CSR_READ_4(sc, ET_RXDMA_CTRL) & ET_RXDMA_CTRL_HALTED) == 0) {
if_printf(sc->ifp, "can't stop RX DMA engine\n");
2009-11-19 21:53:21 +00:00
return (ETIMEDOUT);
}
2009-11-19 21:53:21 +00:00
return (0);
}
static int
et_stop_txdma(struct et_softc *sc)
{
2012-01-10 20:52:02 +00:00
CSR_WRITE_4(sc, ET_TXDMA_CTRL,
ET_TXDMA_CTRL_HALT | ET_TXDMA_CTRL_SINGLE_EPKT);
2009-11-19 21:53:21 +00:00
return (0);
}
static void
et_free_tx_ring(struct et_softc *sc)
{
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
struct et_txdesc_ring *tx_ring;
struct et_txbuf_data *tbd;
struct et_txbuf *tb;
int i;
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
tbd = &sc->sc_tx_data;
tx_ring = &sc->sc_tx_ring;
for (i = 0; i < ET_TX_NDESC; ++i) {
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
tb = &tbd->tbd_buf[i];
if (tb->tb_mbuf != NULL) {
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
bus_dmamap_sync(sc->sc_tx_tag, tb->tb_dmap,
BUS_DMASYNC_POSTWRITE);
bus_dmamap_unload(sc->sc_mbuf_dtag, tb->tb_dmap);
m_freem(tb->tb_mbuf);
tb->tb_mbuf = NULL;
}
}
}
static void
et_free_rx_ring(struct et_softc *sc)
{
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
struct et_rxbuf_data *rbd;
struct et_rxdesc_ring *rx_ring;
struct et_rxbuf *rb;
int i;
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* Ring 0 */
rx_ring = &sc->sc_rx_ring[0];
rbd = &sc->sc_rx_data[0];
for (i = 0; i < ET_RX_NDESC; ++i) {
rb = &rbd->rbd_buf[i];
if (rb->rb_mbuf != NULL) {
bus_dmamap_sync(sc->sc_rx_mini_tag, rx_ring->rr_dmap,
BUS_DMASYNC_POSTREAD);
bus_dmamap_unload(sc->sc_rx_mini_tag, rb->rb_dmap);
m_freem(rb->rb_mbuf);
rb->rb_mbuf = NULL;
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* Ring 1 */
rx_ring = &sc->sc_rx_ring[1];
rbd = &sc->sc_rx_data[1];
for (i = 0; i < ET_RX_NDESC; ++i) {
rb = &rbd->rbd_buf[i];
if (rb->rb_mbuf != NULL) {
bus_dmamap_sync(sc->sc_rx_tag, rx_ring->rr_dmap,
BUS_DMASYNC_POSTREAD);
bus_dmamap_unload(sc->sc_rx_tag, rb->rb_dmap);
m_freem(rb->rb_mbuf);
rb->rb_mbuf = NULL;
}
}
}
static void
et_setmulti(struct et_softc *sc)
{
struct ifnet *ifp;
uint32_t hash[4] = { 0, 0, 0, 0 };
uint32_t rxmac_ctrl, pktfilt;
struct ifmultiaddr *ifma;
int i, count;
ET_LOCK_ASSERT(sc);
ifp = sc->ifp;
pktfilt = CSR_READ_4(sc, ET_PKTFILT);
rxmac_ctrl = CSR_READ_4(sc, ET_RXMAC_CTRL);
pktfilt &= ~(ET_PKTFILT_BCAST | ET_PKTFILT_MCAST | ET_PKTFILT_UCAST);
if (ifp->if_flags & (IFF_PROMISC | IFF_ALLMULTI)) {
rxmac_ctrl |= ET_RXMAC_CTRL_NO_PKTFILT;
goto back;
}
count = 0;
if_maddr_rlock(ifp);
ifnet: Replace if_addr_lock rwlock with epoch + mutex Run on LLNW canaries and tested by pho@ gallatin: Using a 14-core, 28-HTT single socket E5-2697 v3 with a 40GbE MLX5 based ConnectX 4-LX NIC, I see an almost 12% improvement in received packet rate, and a larger improvement in bytes delivered all the way to userspace. When the host receiving 64 streams of netperf -H $DUT -t UDP_STREAM -- -m 1, I see, using nstat -I mce0 1 before the patch: InMpps OMpps InGbs OGbs err TCP Est %CPU syscalls csw irq GBfree 4.98 0.00 4.42 0.00 4235592 33 83.80 4720653 2149771 1235 247.32 4.73 0.00 4.20 0.00 4025260 33 82.99 4724900 2139833 1204 247.32 4.72 0.00 4.20 0.00 4035252 33 82.14 4719162 2132023 1264 247.32 4.71 0.00 4.21 0.00 4073206 33 83.68 4744973 2123317 1347 247.32 4.72 0.00 4.21 0.00 4061118 33 80.82 4713615 2188091 1490 247.32 4.72 0.00 4.21 0.00 4051675 33 85.29 4727399 2109011 1205 247.32 4.73 0.00 4.21 0.00 4039056 33 84.65 4724735 2102603 1053 247.32 After the patch InMpps OMpps InGbs OGbs err TCP Est %CPU syscalls csw irq GBfree 5.43 0.00 4.20 0.00 3313143 33 84.96 5434214 1900162 2656 245.51 5.43 0.00 4.20 0.00 3308527 33 85.24 5439695 1809382 2521 245.51 5.42 0.00 4.19 0.00 3316778 33 87.54 5416028 1805835 2256 245.51 5.42 0.00 4.19 0.00 3317673 33 90.44 5426044 1763056 2332 245.51 5.42 0.00 4.19 0.00 3314839 33 88.11 5435732 1792218 2499 245.52 5.44 0.00 4.19 0.00 3293228 33 91.84 5426301 1668597 2121 245.52 Similarly, netperf reports 230Mb/s before the patch, and 270Mb/s after the patch Reviewed by: gallatin Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D15366
2018-05-18 20:13:34 +00:00
CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
uint32_t *hp, h;
if (ifma->ifma_addr->sa_family != AF_LINK)
continue;
h = ether_crc32_be(LLADDR((struct sockaddr_dl *)
ifma->ifma_addr), ETHER_ADDR_LEN);
h = (h & 0x3f800000) >> 23;
hp = &hash[0];
if (h >= 32 && h < 64) {
h -= 32;
hp = &hash[1];
} else if (h >= 64 && h < 96) {
h -= 64;
hp = &hash[2];
} else if (h >= 96) {
h -= 96;
hp = &hash[3];
}
*hp |= (1 << h);
++count;
}
if_maddr_runlock(ifp);
for (i = 0; i < 4; ++i)
CSR_WRITE_4(sc, ET_MULTI_HASH + (i * 4), hash[i]);
if (count > 0)
pktfilt |= ET_PKTFILT_MCAST;
rxmac_ctrl &= ~ET_RXMAC_CTRL_NO_PKTFILT;
back:
CSR_WRITE_4(sc, ET_PKTFILT, pktfilt);
CSR_WRITE_4(sc, ET_RXMAC_CTRL, rxmac_ctrl);
}
static int
et_chip_init(struct et_softc *sc)
{
2012-01-10 20:52:02 +00:00
struct ifnet *ifp;
uint32_t rxq_end;
int error, frame_len, rxmem_size;
2012-01-10 20:52:02 +00:00
ifp = sc->ifp;
/*
* Split 16Kbytes internal memory between TX and RX
* according to frame length.
*/
frame_len = ET_FRAMELEN(ifp->if_mtu);
if (frame_len < 2048) {
rxmem_size = ET_MEM_RXSIZE_DEFAULT;
} else if (frame_len <= ET_RXMAC_CUT_THRU_FRMLEN) {
rxmem_size = ET_MEM_SIZE / 2;
} else {
rxmem_size = ET_MEM_SIZE -
roundup(frame_len + ET_MEM_TXSIZE_EX, ET_MEM_UNIT);
}
rxq_end = ET_QUEUE_ADDR(rxmem_size);
CSR_WRITE_4(sc, ET_RXQUEUE_START, ET_QUEUE_ADDR_START);
CSR_WRITE_4(sc, ET_RXQUEUE_END, rxq_end);
CSR_WRITE_4(sc, ET_TXQUEUE_START, rxq_end + 1);
CSR_WRITE_4(sc, ET_TXQUEUE_END, ET_QUEUE_ADDR_END);
/* No loopback */
CSR_WRITE_4(sc, ET_LOOPBACK, 0);
/* Clear MSI configure */
2009-11-19 21:45:06 +00:00
if ((sc->sc_flags & ET_FLAG_MSI) == 0)
CSR_WRITE_4(sc, ET_MSI_CFG, 0);
/* Disable timer */
CSR_WRITE_4(sc, ET_TIMER, 0);
/* Initialize MAC */
et_init_mac(sc);
/* Enable memory controllers */
CSR_WRITE_4(sc, ET_MMC_CTRL, ET_MMC_CTRL_ENABLE);
/* Initialize RX MAC */
et_init_rxmac(sc);
/* Initialize TX MAC */
et_init_txmac(sc);
/* Initialize RX DMA engine */
error = et_init_rxdma(sc);
if (error)
2009-11-19 21:53:21 +00:00
return (error);
/* Initialize TX DMA engine */
error = et_init_txdma(sc);
if (error)
2009-11-19 21:53:21 +00:00
return (error);
2009-11-19 21:53:21 +00:00
return (0);
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
static void
et_init_tx_ring(struct et_softc *sc)
{
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
struct et_txdesc_ring *tx_ring;
struct et_txbuf_data *tbd;
struct et_txstatus_data *txsd;
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
tx_ring = &sc->sc_tx_ring;
bzero(tx_ring->tr_desc, ET_TX_RING_SIZE);
bus_dmamap_sync(tx_ring->tr_dtag, tx_ring->tr_dmap,
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
BUS_DMASYNC_PREWRITE);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
tbd = &sc->sc_tx_data;
tbd->tbd_start_index = 0;
tbd->tbd_start_wrap = 0;
tbd->tbd_used = 0;
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
txsd = &sc->sc_tx_status;
bzero(txsd->txsd_status, sizeof(uint32_t));
bus_dmamap_sync(txsd->txsd_dtag, txsd->txsd_dmap,
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
}
static int
et_init_rx_ring(struct et_softc *sc)
{
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
struct et_rxstatus_data *rxsd;
struct et_rxstat_ring *rxst_ring;
struct et_rxbuf_data *rbd;
int i, error, n;
for (n = 0; n < ET_RX_NRING; ++n) {
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
rbd = &sc->sc_rx_data[n];
for (i = 0; i < ET_RX_NDESC; ++i) {
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
error = rbd->rbd_newbuf(rbd, i);
if (error) {
if_printf(sc->ifp, "%d ring %d buf, "
"newbuf failed: %d\n", n, i, error);
2009-11-19 21:53:21 +00:00
return (error);
}
}
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
rxsd = &sc->sc_rx_status;
bzero(rxsd->rxsd_status, sizeof(struct et_rxstatus));
bus_dmamap_sync(rxsd->rxsd_dtag, rxsd->rxsd_dmap,
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
rxst_ring = &sc->sc_rxstat_ring;
bzero(rxst_ring->rsr_stat, ET_RXSTAT_RING_SIZE);
bus_dmamap_sync(rxst_ring->rsr_dtag, rxst_ring->rsr_dmap,
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2009-11-19 21:53:21 +00:00
return (0);
}
static int
et_init_rxdma(struct et_softc *sc)
{
2012-01-10 20:52:02 +00:00
struct et_rxstatus_data *rxsd;
struct et_rxstat_ring *rxst_ring;
struct et_rxdesc_ring *rx_ring;
int error;
error = et_stop_rxdma(sc);
if (error) {
if_printf(sc->ifp, "can't init RX DMA engine\n");
2009-11-19 21:53:21 +00:00
return (error);
}
/*
* Install RX status
*/
2012-01-10 20:52:02 +00:00
rxsd = &sc->sc_rx_status;
CSR_WRITE_4(sc, ET_RX_STATUS_HI, ET_ADDR_HI(rxsd->rxsd_paddr));
CSR_WRITE_4(sc, ET_RX_STATUS_LO, ET_ADDR_LO(rxsd->rxsd_paddr));
/*
* Install RX stat ring
*/
2012-01-10 20:52:02 +00:00
rxst_ring = &sc->sc_rxstat_ring;
CSR_WRITE_4(sc, ET_RXSTAT_HI, ET_ADDR_HI(rxst_ring->rsr_paddr));
CSR_WRITE_4(sc, ET_RXSTAT_LO, ET_ADDR_LO(rxst_ring->rsr_paddr));
CSR_WRITE_4(sc, ET_RXSTAT_CNT, ET_RX_NSTAT - 1);
CSR_WRITE_4(sc, ET_RXSTAT_POS, 0);
CSR_WRITE_4(sc, ET_RXSTAT_MINCNT, ((ET_RX_NSTAT * 15) / 100) - 1);
/* Match ET_RXSTAT_POS */
rxst_ring->rsr_index = 0;
rxst_ring->rsr_wrap = 0;
/*
* Install the 2nd RX descriptor ring
*/
rx_ring = &sc->sc_rx_ring[1];
CSR_WRITE_4(sc, ET_RX_RING1_HI, ET_ADDR_HI(rx_ring->rr_paddr));
CSR_WRITE_4(sc, ET_RX_RING1_LO, ET_ADDR_LO(rx_ring->rr_paddr));
CSR_WRITE_4(sc, ET_RX_RING1_CNT, ET_RX_NDESC - 1);
CSR_WRITE_4(sc, ET_RX_RING1_POS, ET_RX_RING1_POS_WRAP);
CSR_WRITE_4(sc, ET_RX_RING1_MINCNT, ((ET_RX_NDESC * 15) / 100) - 1);
/* Match ET_RX_RING1_POS */
rx_ring->rr_index = 0;
rx_ring->rr_wrap = 1;
/*
* Install the 1st RX descriptor ring
*/
rx_ring = &sc->sc_rx_ring[0];
CSR_WRITE_4(sc, ET_RX_RING0_HI, ET_ADDR_HI(rx_ring->rr_paddr));
CSR_WRITE_4(sc, ET_RX_RING0_LO, ET_ADDR_LO(rx_ring->rr_paddr));
CSR_WRITE_4(sc, ET_RX_RING0_CNT, ET_RX_NDESC - 1);
CSR_WRITE_4(sc, ET_RX_RING0_POS, ET_RX_RING0_POS_WRAP);
CSR_WRITE_4(sc, ET_RX_RING0_MINCNT, ((ET_RX_NDESC * 15) / 100) - 1);
/* Match ET_RX_RING0_POS */
rx_ring->rr_index = 0;
rx_ring->rr_wrap = 1;
/*
* RX intr moderation
*/
CSR_WRITE_4(sc, ET_RX_INTR_NPKTS, sc->sc_rx_intr_npkts);
CSR_WRITE_4(sc, ET_RX_INTR_DELAY, sc->sc_rx_intr_delay);
2009-11-19 21:53:21 +00:00
return (0);
}
static int
et_init_txdma(struct et_softc *sc)
{
2012-01-10 20:52:02 +00:00
struct et_txdesc_ring *tx_ring;
struct et_txstatus_data *txsd;
int error;
error = et_stop_txdma(sc);
if (error) {
if_printf(sc->ifp, "can't init TX DMA engine\n");
2009-11-19 21:53:21 +00:00
return (error);
}
/*
* Install TX descriptor ring
*/
2012-01-10 20:52:02 +00:00
tx_ring = &sc->sc_tx_ring;
CSR_WRITE_4(sc, ET_TX_RING_HI, ET_ADDR_HI(tx_ring->tr_paddr));
CSR_WRITE_4(sc, ET_TX_RING_LO, ET_ADDR_LO(tx_ring->tr_paddr));
CSR_WRITE_4(sc, ET_TX_RING_CNT, ET_TX_NDESC - 1);
/*
* Install TX status
*/
2012-01-10 20:52:02 +00:00
txsd = &sc->sc_tx_status;
CSR_WRITE_4(sc, ET_TX_STATUS_HI, ET_ADDR_HI(txsd->txsd_paddr));
CSR_WRITE_4(sc, ET_TX_STATUS_LO, ET_ADDR_LO(txsd->txsd_paddr));
CSR_WRITE_4(sc, ET_TX_READY_POS, 0);
/* Match ET_TX_READY_POS */
tx_ring->tr_ready_index = 0;
tx_ring->tr_ready_wrap = 0;
2009-11-19 21:53:21 +00:00
return (0);
}
static void
et_init_mac(struct et_softc *sc)
{
2012-01-10 20:52:02 +00:00
struct ifnet *ifp;
const uint8_t *eaddr;
uint32_t val;
/* Reset MAC */
CSR_WRITE_4(sc, ET_MAC_CFG1,
ET_MAC_CFG1_RST_TXFUNC | ET_MAC_CFG1_RST_RXFUNC |
ET_MAC_CFG1_RST_TXMC | ET_MAC_CFG1_RST_RXMC |
ET_MAC_CFG1_SIM_RST | ET_MAC_CFG1_SOFT_RST);
/*
* Setup inter packet gap
*/
val = (56 << ET_IPG_NONB2B_1_SHIFT) |
(88 << ET_IPG_NONB2B_2_SHIFT) |
(80 << ET_IPG_MINIFG_SHIFT) |
(96 << ET_IPG_B2B_SHIFT);
CSR_WRITE_4(sc, ET_IPG, val);
/*
* Setup half duplex mode
*/
val = (10 << ET_MAC_HDX_ALT_BEB_TRUNC_SHIFT) |
(15 << ET_MAC_HDX_REXMIT_MAX_SHIFT) |
(55 << ET_MAC_HDX_COLLWIN_SHIFT) |
ET_MAC_HDX_EXC_DEFER;
CSR_WRITE_4(sc, ET_MAC_HDX, val);
/* Clear MAC control */
CSR_WRITE_4(sc, ET_MAC_CTRL, 0);
/* Reset MII */
CSR_WRITE_4(sc, ET_MII_CFG, ET_MII_CFG_CLKRST);
/*
* Set MAC address
*/
2012-01-10 20:52:02 +00:00
ifp = sc->ifp;
eaddr = IF_LLADDR(ifp);
val = eaddr[2] | (eaddr[3] << 8) | (eaddr[4] << 16) | (eaddr[5] << 24);
CSR_WRITE_4(sc, ET_MAC_ADDR1, val);
val = (eaddr[0] << 16) | (eaddr[1] << 24);
CSR_WRITE_4(sc, ET_MAC_ADDR2, val);
/* Set max frame length */
CSR_WRITE_4(sc, ET_MAX_FRMLEN, ET_FRAMELEN(ifp->if_mtu));
/* Bring MAC out of reset state */
CSR_WRITE_4(sc, ET_MAC_CFG1, 0);
}
static void
et_init_rxmac(struct et_softc *sc)
{
2012-01-10 20:52:02 +00:00
struct ifnet *ifp;
const uint8_t *eaddr;
uint32_t val;
int i;
/* Disable RX MAC and WOL */
CSR_WRITE_4(sc, ET_RXMAC_CTRL, ET_RXMAC_CTRL_WOL_DISABLE);
/*
* Clear all WOL related registers
*/
for (i = 0; i < 3; ++i)
CSR_WRITE_4(sc, ET_WOL_CRC + (i * 4), 0);
for (i = 0; i < 20; ++i)
CSR_WRITE_4(sc, ET_WOL_MASK + (i * 4), 0);
/*
* Set WOL source address. XXX is this necessary?
*/
2012-01-10 20:52:02 +00:00
ifp = sc->ifp;
eaddr = IF_LLADDR(ifp);
val = (eaddr[2] << 24) | (eaddr[3] << 16) | (eaddr[4] << 8) | eaddr[5];
CSR_WRITE_4(sc, ET_WOL_SA_LO, val);
val = (eaddr[0] << 8) | eaddr[1];
CSR_WRITE_4(sc, ET_WOL_SA_HI, val);
/* Clear packet filters */
CSR_WRITE_4(sc, ET_PKTFILT, 0);
/* No ucast filtering */
CSR_WRITE_4(sc, ET_UCAST_FILTADDR1, 0);
CSR_WRITE_4(sc, ET_UCAST_FILTADDR2, 0);
CSR_WRITE_4(sc, ET_UCAST_FILTADDR3, 0);
if (ET_FRAMELEN(ifp->if_mtu) > ET_RXMAC_CUT_THRU_FRMLEN) {
/*
* In order to transmit jumbo packets greater than
* ET_RXMAC_CUT_THRU_FRMLEN bytes, the FIFO between
* RX MAC and RX DMA needs to be reduced in size to
* (ET_MEM_SIZE - ET_MEM_TXSIZE_EX - framelen). In
* order to implement this, we must use "cut through"
* mode in the RX MAC, which chops packets down into
* segments. In this case we selected 256 bytes,
* since this is the size of the PCI-Express TLP's
* that the ET1310 uses.
*/
val = (ET_RXMAC_SEGSZ(256) & ET_RXMAC_MC_SEGSZ_MAX_MASK) |
ET_RXMAC_MC_SEGSZ_ENABLE;
} else {
val = 0;
}
CSR_WRITE_4(sc, ET_RXMAC_MC_SEGSZ, val);
CSR_WRITE_4(sc, ET_RXMAC_MC_WATERMARK, 0);
/* Initialize RX MAC management register */
CSR_WRITE_4(sc, ET_RXMAC_MGT, 0);
CSR_WRITE_4(sc, ET_RXMAC_SPACE_AVL, 0);
CSR_WRITE_4(sc, ET_RXMAC_MGT,
ET_RXMAC_MGT_PASS_ECRC |
ET_RXMAC_MGT_PASS_ELEN |
ET_RXMAC_MGT_PASS_ETRUNC |
ET_RXMAC_MGT_CHECK_PKT);
/*
* Configure runt filtering (may not work on certain chip generation)
*/
val = (ETHER_MIN_LEN << ET_PKTFILT_MINLEN_SHIFT) &
ET_PKTFILT_MINLEN_MASK;
val |= ET_PKTFILT_FRAG;
CSR_WRITE_4(sc, ET_PKTFILT, val);
/* Enable RX MAC but leave WOL disabled */
CSR_WRITE_4(sc, ET_RXMAC_CTRL,
ET_RXMAC_CTRL_WOL_DISABLE | ET_RXMAC_CTRL_ENABLE);
/*
* Setup multicast hash and allmulti/promisc mode
*/
et_setmulti(sc);
}
static void
et_init_txmac(struct et_softc *sc)
{
2012-01-10 20:52:02 +00:00
/* Disable TX MAC and FC(?) */
CSR_WRITE_4(sc, ET_TXMAC_CTRL, ET_TXMAC_CTRL_FC_DISABLE);
/*
* Initialize pause time.
* This register should be set before XON/XOFF frame is
* sent by driver.
*/
CSR_WRITE_4(sc, ET_TXMAC_FLOWCTRL, 0 << ET_TXMAC_FLOWCTRL_CFPT_SHIFT);
/* Enable TX MAC but leave FC(?) diabled */
CSR_WRITE_4(sc, ET_TXMAC_CTRL,
ET_TXMAC_CTRL_ENABLE | ET_TXMAC_CTRL_FC_DISABLE);
}
static int
et_start_rxdma(struct et_softc *sc)
{
2012-01-10 20:52:02 +00:00
uint32_t val;
2012-01-10 20:52:02 +00:00
val = (sc->sc_rx_data[0].rbd_bufsize & ET_RXDMA_CTRL_RING0_SIZE_MASK) |
ET_RXDMA_CTRL_RING0_ENABLE;
val |= (sc->sc_rx_data[1].rbd_bufsize & ET_RXDMA_CTRL_RING1_SIZE_MASK) |
2012-01-10 20:52:02 +00:00
ET_RXDMA_CTRL_RING1_ENABLE;
CSR_WRITE_4(sc, ET_RXDMA_CTRL, val);
DELAY(5);
if (CSR_READ_4(sc, ET_RXDMA_CTRL) & ET_RXDMA_CTRL_HALTED) {
if_printf(sc->ifp, "can't start RX DMA engine\n");
2009-11-19 21:53:21 +00:00
return (ETIMEDOUT);
}
2009-11-19 21:53:21 +00:00
return (0);
}
static int
et_start_txdma(struct et_softc *sc)
{
2012-01-10 20:52:02 +00:00
CSR_WRITE_4(sc, ET_TXDMA_CTRL, ET_TXDMA_CTRL_SINGLE_EPKT);
2009-11-19 21:53:21 +00:00
return (0);
}
static void
et_rxeof(struct et_softc *sc)
{
struct et_rxstatus_data *rxsd;
struct et_rxstat_ring *rxst_ring;
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
struct et_rxbuf_data *rbd;
struct et_rxdesc_ring *rx_ring;
struct et_rxstat *st;
struct ifnet *ifp;
struct mbuf *m;
uint32_t rxstat_pos, rxring_pos;
uint32_t rxst_info1, rxst_info2, rxs_stat_ring;
int buflen, buf_idx, npost[2], ring_idx;
int rxst_index, rxst_wrap;
ET_LOCK_ASSERT(sc);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
ifp = sc->ifp;
rxsd = &sc->sc_rx_status;
rxst_ring = &sc->sc_rxstat_ring;
if ((sc->sc_flags & ET_FLAG_TXRX_ENABLED) == 0)
return;
bus_dmamap_sync(rxsd->rxsd_dtag, rxsd->rxsd_dmap,
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
BUS_DMASYNC_POSTREAD);
bus_dmamap_sync(rxst_ring->rsr_dtag, rxst_ring->rsr_dmap,
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
BUS_DMASYNC_POSTREAD);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
npost[0] = npost[1] = 0;
rxs_stat_ring = le32toh(rxsd->rxsd_status->rxs_stat_ring);
rxst_wrap = (rxs_stat_ring & ET_RXS_STATRING_WRAP) ? 1 : 0;
rxst_index = (rxs_stat_ring & ET_RXS_STATRING_INDEX_MASK) >>
ET_RXS_STATRING_INDEX_SHIFT;
while (rxst_index != rxst_ring->rsr_index ||
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
rxst_wrap != rxst_ring->rsr_wrap) {
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
break;
MPASS(rxst_ring->rsr_index < ET_RX_NSTAT);
st = &rxst_ring->rsr_stat[rxst_ring->rsr_index];
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
rxst_info1 = le32toh(st->rxst_info1);
rxst_info2 = le32toh(st->rxst_info2);
buflen = (rxst_info2 & ET_RXST_INFO2_LEN_MASK) >>
ET_RXST_INFO2_LEN_SHIFT;
buf_idx = (rxst_info2 & ET_RXST_INFO2_BUFIDX_MASK) >>
ET_RXST_INFO2_BUFIDX_SHIFT;
ring_idx = (rxst_info2 & ET_RXST_INFO2_RINGIDX_MASK) >>
ET_RXST_INFO2_RINGIDX_SHIFT;
if (++rxst_ring->rsr_index == ET_RX_NSTAT) {
rxst_ring->rsr_index = 0;
rxst_ring->rsr_wrap ^= 1;
}
rxstat_pos = rxst_ring->rsr_index & ET_RXSTAT_POS_INDEX_MASK;
if (rxst_ring->rsr_wrap)
rxstat_pos |= ET_RXSTAT_POS_WRAP;
CSR_WRITE_4(sc, ET_RXSTAT_POS, rxstat_pos);
if (ring_idx >= ET_RX_NRING) {
if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
if_printf(ifp, "invalid ring index %d\n", ring_idx);
continue;
}
if (buf_idx >= ET_RX_NDESC) {
if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
if_printf(ifp, "invalid buf index %d\n", buf_idx);
continue;
}
rbd = &sc->sc_rx_data[ring_idx];
m = rbd->rbd_buf[buf_idx].rb_mbuf;
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
if ((rxst_info1 & ET_RXST_INFO1_OK) == 0){
/* Discard errored frame. */
rbd->rbd_discard(rbd, buf_idx);
} else if (rbd->rbd_newbuf(rbd, buf_idx) != 0) {
/* No available mbufs, discard it. */
if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
rbd->rbd_discard(rbd, buf_idx);
} else {
buflen -= ETHER_CRC_LEN;
if (buflen < ETHER_HDR_LEN) {
m_freem(m);
if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
} else {
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
m->m_pkthdr.len = m->m_len = buflen;
m->m_pkthdr.rcvif = ifp;
ET_UNLOCK(sc);
ifp->if_input(ifp, m);
ET_LOCK(sc);
}
}
rx_ring = &sc->sc_rx_ring[ring_idx];
if (buf_idx != rx_ring->rr_index) {
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
if_printf(ifp,
"WARNING!! ring %d, buf_idx %d, rr_idx %d\n",
ring_idx, buf_idx, rx_ring->rr_index);
}
MPASS(rx_ring->rr_index < ET_RX_NDESC);
if (++rx_ring->rr_index == ET_RX_NDESC) {
rx_ring->rr_index = 0;
rx_ring->rr_wrap ^= 1;
}
rxring_pos = rx_ring->rr_index & ET_RX_RING_POS_INDEX_MASK;
if (rx_ring->rr_wrap)
rxring_pos |= ET_RX_RING_POS_WRAP;
CSR_WRITE_4(sc, rx_ring->rr_posreg, rxring_pos);
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
bus_dmamap_sync(rxsd->rxsd_dtag, rxsd->rxsd_dmap,
BUS_DMASYNC_PREREAD);
bus_dmamap_sync(rxst_ring->rsr_dtag, rxst_ring->rsr_dmap,
BUS_DMASYNC_PREREAD);
}
static int
et_encap(struct et_softc *sc, struct mbuf **m0)
{
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
struct et_txdesc_ring *tx_ring;
struct et_txbuf_data *tbd;
struct et_txdesc *td;
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
struct mbuf *m;
bus_dma_segment_t segs[ET_NSEG_MAX];
bus_dmamap_t map;
uint32_t csum_flags, last_td_ctrl2;
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
int error, i, idx, first_idx, last_idx, nsegs;
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
tx_ring = &sc->sc_tx_ring;
MPASS(tx_ring->tr_ready_index < ET_TX_NDESC);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
tbd = &sc->sc_tx_data;
first_idx = tx_ring->tr_ready_index;
map = tbd->tbd_buf[first_idx].tb_dmap;
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
error = bus_dmamap_load_mbuf_sg(sc->sc_tx_tag, map, *m0, segs, &nsegs,
0);
if (error == EFBIG) {
m = m_collapse(*m0, M_NOWAIT, ET_NSEG_MAX);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
if (m == NULL) {
m_freem(*m0);
*m0 = NULL;
return (ENOMEM);
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
*m0 = m;
error = bus_dmamap_load_mbuf_sg(sc->sc_tx_tag, map, *m0, segs,
&nsegs, 0);
if (error != 0) {
m_freem(*m0);
*m0 = NULL;
return (error);
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
} else if (error != 0)
return (error);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* Check for descriptor overruns. */
if (tbd->tbd_used + nsegs > ET_TX_NDESC - 1) {
bus_dmamap_unload(sc->sc_tx_tag, map);
return (ENOBUFS);
}
bus_dmamap_sync(sc->sc_tx_tag, map, BUS_DMASYNC_PREWRITE);
last_td_ctrl2 = ET_TDCTRL2_LAST_FRAG;
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
sc->sc_tx += nsegs;
if (sc->sc_tx / sc->sc_tx_intr_nsegs != sc->sc_tx_intr) {
sc->sc_tx_intr = sc->sc_tx / sc->sc_tx_intr_nsegs;
last_td_ctrl2 |= ET_TDCTRL2_INTR;
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
m = *m0;
csum_flags = 0;
if ((m->m_pkthdr.csum_flags & ET_CSUM_FEATURES) != 0) {
if ((m->m_pkthdr.csum_flags & CSUM_IP) != 0)
csum_flags |= ET_TDCTRL2_CSUM_IP;
if ((m->m_pkthdr.csum_flags & CSUM_UDP) != 0)
csum_flags |= ET_TDCTRL2_CSUM_UDP;
else if ((m->m_pkthdr.csum_flags & CSUM_TCP) != 0)
csum_flags |= ET_TDCTRL2_CSUM_TCP;
}
last_idx = -1;
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
for (i = 0; i < nsegs; ++i) {
idx = (first_idx + i) % ET_TX_NDESC;
td = &tx_ring->tr_desc[idx];
td->td_addr_hi = htole32(ET_ADDR_HI(segs[i].ds_addr));
td->td_addr_lo = htole32(ET_ADDR_LO(segs[i].ds_addr));
td->td_ctrl1 = htole32(segs[i].ds_len & ET_TDCTRL1_LEN_MASK);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
if (i == nsegs - 1) {
/* Last frag */
td->td_ctrl2 = htole32(last_td_ctrl2 | csum_flags);
last_idx = idx;
} else
td->td_ctrl2 = htole32(csum_flags);
MPASS(tx_ring->tr_ready_index < ET_TX_NDESC);
if (++tx_ring->tr_ready_index == ET_TX_NDESC) {
tx_ring->tr_ready_index = 0;
tx_ring->tr_ready_wrap ^= 1;
}
}
td = &tx_ring->tr_desc[first_idx];
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
/* First frag */
td->td_ctrl2 |= htole32(ET_TDCTRL2_FIRST_FRAG);
MPASS(last_idx >= 0);
tbd->tbd_buf[first_idx].tb_dmap = tbd->tbd_buf[last_idx].tb_dmap;
tbd->tbd_buf[last_idx].tb_dmap = map;
tbd->tbd_buf[last_idx].tb_mbuf = m;
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
tbd->tbd_used += nsegs;
MPASS(tbd->tbd_used <= ET_TX_NDESC);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
return (0);
}
static void
et_txeof(struct et_softc *sc)
{
struct et_txdesc_ring *tx_ring;
struct et_txbuf_data *tbd;
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
struct et_txbuf *tb;
struct ifnet *ifp;
uint32_t tx_done;
int end, wrap;
ET_LOCK_ASSERT(sc);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
ifp = sc->ifp;
tx_ring = &sc->sc_tx_ring;
tbd = &sc->sc_tx_data;
if ((sc->sc_flags & ET_FLAG_TXRX_ENABLED) == 0)
return;
if (tbd->tbd_used == 0)
return;
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
bus_dmamap_sync(tx_ring->tr_dtag, tx_ring->tr_dmap,
BUS_DMASYNC_POSTWRITE);
tx_done = CSR_READ_4(sc, ET_TX_DONE_POS);
end = tx_done & ET_TX_DONE_POS_INDEX_MASK;
wrap = (tx_done & ET_TX_DONE_POS_WRAP) ? 1 : 0;
while (tbd->tbd_start_index != end || tbd->tbd_start_wrap != wrap) {
MPASS(tbd->tbd_start_index < ET_TX_NDESC);
tb = &tbd->tbd_buf[tbd->tbd_start_index];
if (tb->tb_mbuf != NULL) {
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
bus_dmamap_sync(sc->sc_tx_tag, tb->tb_dmap,
BUS_DMASYNC_POSTWRITE);
bus_dmamap_unload(sc->sc_tx_tag, tb->tb_dmap);
m_freem(tb->tb_mbuf);
tb->tb_mbuf = NULL;
}
if (++tbd->tbd_start_index == ET_TX_NDESC) {
tbd->tbd_start_index = 0;
tbd->tbd_start_wrap ^= 1;
}
MPASS(tbd->tbd_used > 0);
tbd->tbd_used--;
}
if (tbd->tbd_used == 0)
sc->watchdog_timer = 0;
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
if (tbd->tbd_used + ET_NSEG_SPARE < ET_TX_NDESC)
ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
}
Rework link state tracking and TX/RX MAC configuration. o Do not report link status if driver is not running. o TX/RX MAC configuration should be done with resolved speed, duplex and flow control after establishing a link so it can't be done in driver initialization routine. Move the configuration to miibus_statchg callback which will be called whenever any link state change is detected. At this moment, flow-control is not enabled yet mainly because I was not able to set correct flow control parameters to generate TX pause frames. o Now TX/RX MAC is enabled only when a valid link is detected. Rearragnge hardware initialization routine a bit to leave enabling MAC to miibus_statchg callback. In order to that, TX/RX DMA engine is enabled in et_init_locked(). o Introduce ET_FLAG_LINK flag to track current link state. o Introduce ET_FLAG_FASTETHER flag to mark whether controller is fast ethernet. This flag is checked in miibus_statchg callback to know whether PHY established a valid link. o In et_stop(), TX/RX MAC is explicitly disabled instead of relying on et_reset(). And move et_reset() from et_stop() to controller initialization. Controler reset is not required here and it would also clear critial registers(i.e station address, RX filter configuration, WOL etc) that are required to make WOL work. o Switching to current media is done in et_init_locked() after setting IFF_DRV_RUNNING flag. This should ensure reliable auto-negotiation/manual link establishment. o In et_start_locked(), check whether driver got a valid link before trying to send frames. o Remove checking a link in et_tick() as this is done by miibus_statchg callback.
2011-12-07 21:29:51 +00:00
static void
et_tick(void *xsc)
{
2012-01-10 20:52:02 +00:00
struct et_softc *sc;
struct ifnet *ifp;
struct mii_data *mii;
2012-01-10 20:52:02 +00:00
sc = xsc;
ET_LOCK_ASSERT(sc);
ifp = sc->ifp;
mii = device_get_softc(sc->sc_miibus);
mii_tick(mii);
et_stats_update(sc);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
if (et_watchdog(sc) == EJUSTRETURN)
return;
callout_reset(&sc->sc_tick, hz, et_tick, sc);
}
static int
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
et_newbuf_cluster(struct et_rxbuf_data *rbd, int buf_idx)
{
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
struct et_softc *sc;
struct et_rxdesc *desc;
struct et_rxbuf *rb;
struct mbuf *m;
bus_dma_segment_t segs[1];
bus_dmamap_t dmap;
int nsegs;
MPASS(buf_idx < ET_RX_NDESC);
m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
if (m == NULL)
return (ENOBUFS);
m->m_len = m->m_pkthdr.len = MCLBYTES;
m_adj(m, ETHER_ALIGN);
sc = rbd->rbd_softc;
rb = &rbd->rbd_buf[buf_idx];
if (bus_dmamap_load_mbuf_sg(sc->sc_rx_tag, sc->sc_rx_sparemap, m,
segs, &nsegs, 0) != 0) {
m_freem(m);
return (ENOBUFS);
}
KASSERT(nsegs == 1, ("%s: %d segments returned!", __func__, nsegs));
if (rb->rb_mbuf != NULL) {
bus_dmamap_sync(sc->sc_rx_tag, rb->rb_dmap,
BUS_DMASYNC_POSTREAD);
bus_dmamap_unload(sc->sc_rx_tag, rb->rb_dmap);
}
dmap = rb->rb_dmap;
rb->rb_dmap = sc->sc_rx_sparemap;
sc->sc_rx_sparemap = dmap;
bus_dmamap_sync(sc->sc_rx_tag, rb->rb_dmap, BUS_DMASYNC_PREREAD);
rb->rb_mbuf = m;
desc = &rbd->rbd_ring->rr_desc[buf_idx];
desc->rd_addr_hi = htole32(ET_ADDR_HI(segs[0].ds_addr));
desc->rd_addr_lo = htole32(ET_ADDR_LO(segs[0].ds_addr));
desc->rd_ctrl = htole32(buf_idx & ET_RDCTRL_BUFIDX_MASK);
bus_dmamap_sync(rbd->rbd_ring->rr_dtag, rbd->rbd_ring->rr_dmap,
BUS_DMASYNC_PREWRITE);
return (0);
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
static void
et_rxbuf_discard(struct et_rxbuf_data *rbd, int buf_idx)
{
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
struct et_rxdesc *desc;
desc = &rbd->rbd_ring->rr_desc[buf_idx];
desc->rd_ctrl = htole32(buf_idx & ET_RDCTRL_BUFIDX_MASK);
bus_dmamap_sync(rbd->rbd_ring->rr_dtag, rbd->rbd_ring->rr_dmap,
BUS_DMASYNC_PREWRITE);
}
static int
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
et_newbuf_hdr(struct et_rxbuf_data *rbd, int buf_idx)
{
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
struct et_softc *sc;
struct et_rxdesc *desc;
struct et_rxbuf *rb;
struct mbuf *m;
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
bus_dma_segment_t segs[1];
bus_dmamap_t dmap;
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
int nsegs;
MPASS(buf_idx < ET_RX_NDESC);
MGETHDR(m, M_NOWAIT, MT_DATA);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
if (m == NULL)
return (ENOBUFS);
m->m_len = m->m_pkthdr.len = MHLEN;
m_adj(m, ETHER_ALIGN);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
sc = rbd->rbd_softc;
rb = &rbd->rbd_buf[buf_idx];
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
if (bus_dmamap_load_mbuf_sg(sc->sc_rx_mini_tag, sc->sc_rx_mini_sparemap,
m, segs, &nsegs, 0) != 0) {
m_freem(m);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
return (ENOBUFS);
}
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
KASSERT(nsegs == 1, ("%s: %d segments returned!", __func__, nsegs));
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
if (rb->rb_mbuf != NULL) {
bus_dmamap_sync(sc->sc_rx_mini_tag, rb->rb_dmap,
BUS_DMASYNC_POSTREAD);
bus_dmamap_unload(sc->sc_rx_mini_tag, rb->rb_dmap);
}
dmap = rb->rb_dmap;
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
rb->rb_dmap = sc->sc_rx_mini_sparemap;
sc->sc_rx_mini_sparemap = dmap;
bus_dmamap_sync(sc->sc_rx_mini_tag, rb->rb_dmap, BUS_DMASYNC_PREREAD);
Overhaul bus_dma(9) usage in et(4) and clean up TX/RX path. This change should make et(4) work on any architectures. o Remove m_getl inline function and replace it with stanard mbuf interfaces. Previous code tried to minimize code duplication but this came from incorrect use of common DMA tag. Driver may be still use a common RX allocation handler with additional structure changes but I don't see much point to do that it would make it hard to understand the code. o Remove DragonflyBSD specific constant EVL_ENCAPLEN, use ETHER_VLAN_ENCAP_LEN instead. o Add bunch of new RX status definition. It seems controller supports RX checksum offloading but I was not able to make the feature work yet. Currently driver checks whether recevied frame is good one or not. o Avoid a typedef ending in '_t' as style(9) says. o Controller has no restriction on DMA address space, so there is no reason to limit the DMA address to 32bit. Descriptor rings, status blocks and TX/RX buffers now use full 64bit DMA addressing. o Allocate DMA memory shared between host and controller as coherent. o Create 3 separate DMA tags to be used as TX, mini RX ring and stanard RX ring. Previously it created a single DMA tag and it was used to all three rings. o et(4) does not support jumbo frame at this moment and I still don't quite understand how jumbo frame works on this controller so use two RX rings to handle small sized frame and normal sized frame respectively. The mini RX ring will be used to receive frames that are less than or equal to 127 bytes. The second RX ring is used to receive frames that are not handled by the first RX ring. If jumbo frame support is implemented, driver may have to choose better RX scheme by letting the second RX ring handle jumbo frames. This scheme will mimic Broadcom's efficient jumbo frame handling feature. However RAM buffer size(16KB) of the controller is too small to hold 2 jumbo frames, if 9KB jumbo frame is used, I'm not sure how good performance would it have. o In et_rxeof(), make sure to check whether controller received good frame or not. Passing corrupted frame to upper layer is bad idea. o If driver receives a bad frame or driver fails to allocate RX buffer due to resource shortage condition, reuse previously loaded DMA map for RX buffer instead of unloading/loading RX buffer again. o et_init_tx_ring() never fails so change return type to void. o In watchdog handler, show TX DMA write back status of errored frame which could be used as a clue to debug watchdog timeout. o Add missing bus_dmamap_sync() in various places such that et(4) should work with bounce buffers(e.g. PAE). o TX side bus_dmamap_load_mbuf_sg(9) support. o RX side bus_dmamap_load_mbuf_sg(9) support. o Controller has no DMA alignment limit in RX buffer so use m_adj(9) in RX buffer allocation to make IP header align on 2 bytes boundary. Otherwise it would trigger unaligned access error in upper layer on strict alignment architectures. One of down side of controller is it provides limited set of RX buffer length like most Intel controllers. This is not problem at this moment because driver does not support jumbo frame yet but it may require alignment fixup code to support jumbo frame on strict alignment architectures. o In et_txeof(), don't zero TX descriptors for transmitted frames. TX descriptors don't need write access after transmission. Driver sets IFF_DRV_OACTIVE when the number of available TX descriptors are less than or equal to ET_NSEG_SPARE. Make sure to clear IFF_DRV_OACTIVE only when the number of available TX descriptor is greater than ET_NSEG_SPARE.
2011-12-07 18:17:09 +00:00
rb->rb_mbuf = m;
desc = &rbd->rbd_ring->rr_desc[buf_idx];
desc->rd_addr_hi = htole32(ET_ADDR_HI(segs[0].ds_addr));
desc->rd_addr_lo = htole32(ET_ADDR_LO(segs[0].ds_addr));
desc->rd_ctrl = htole32(buf_idx & ET_RDCTRL_BUFIDX_MASK);
bus_dmamap_sync(rbd->rbd_ring->rr_dtag, rbd->rbd_ring->rr_dmap,
BUS_DMASYNC_PREWRITE);
return (0);
}
#define ET_SYSCTL_STAT_ADD32(c, h, n, p, d) \
SYSCTL_ADD_UINT(c, h, OID_AUTO, n, CTLFLAG_RD, p, 0, d)
#define ET_SYSCTL_STAT_ADD64(c, h, n, p, d) \
SYSCTL_ADD_UQUAD(c, h, OID_AUTO, n, CTLFLAG_RD, p, d)
/*
* Create sysctl tree
*/
static void
et_add_sysctls(struct et_softc * sc)
{
struct sysctl_ctx_list *ctx;
struct sysctl_oid_list *children, *parent;
struct sysctl_oid *tree;
struct et_hw_stats *stats;
ctx = device_get_sysctl_ctx(sc->dev);
children = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->dev));
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rx_intr_npkts",
CTLTYPE_INT | CTLFLAG_RW, sc, 0, et_sysctl_rx_intr_npkts, "I",
"RX IM, # packets per RX interrupt");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rx_intr_delay",
CTLTYPE_INT | CTLFLAG_RW, sc, 0, et_sysctl_rx_intr_delay, "I",
"RX IM, RX interrupt delay (x10 usec)");
SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_intr_nsegs",
CTLFLAG_RW, &sc->sc_tx_intr_nsegs, 0,
"TX IM, # segments per TX interrupt");
SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "timer",
CTLFLAG_RW, &sc->sc_timer, 0, "TX timer");
tree = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats", CTLFLAG_RD,
NULL, "ET statistics");
parent = SYSCTL_CHILDREN(tree);
/* TX/RX statistics. */
stats = &sc->sc_stats;
ET_SYSCTL_STAT_ADD64(ctx, parent, "frames_64", &stats->pkts_64,
"0 to 64 bytes frames");
ET_SYSCTL_STAT_ADD64(ctx, parent, "frames_65_127", &stats->pkts_65,
"65 to 127 bytes frames");
ET_SYSCTL_STAT_ADD64(ctx, parent, "frames_128_255", &stats->pkts_128,
"128 to 255 bytes frames");
ET_SYSCTL_STAT_ADD64(ctx, parent, "frames_256_511", &stats->pkts_256,
"256 to 511 bytes frames");
ET_SYSCTL_STAT_ADD64(ctx, parent, "frames_512_1023", &stats->pkts_512,
"512 to 1023 bytes frames");
ET_SYSCTL_STAT_ADD64(ctx, parent, "frames_1024_1518", &stats->pkts_1024,
"1024 to 1518 bytes frames");
ET_SYSCTL_STAT_ADD64(ctx, parent, "frames_1519_1522", &stats->pkts_1519,
"1519 to 1522 bytes frames");
/* RX statistics. */
tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "rx", CTLFLAG_RD,
NULL, "RX MAC statistics");
children = SYSCTL_CHILDREN(tree);
ET_SYSCTL_STAT_ADD64(ctx, children, "bytes",
&stats->rx_bytes, "Good bytes");
ET_SYSCTL_STAT_ADD64(ctx, children, "frames",
&stats->rx_frames, "Good frames");
ET_SYSCTL_STAT_ADD32(ctx, children, "crc_errs",
&stats->rx_crcerrs, "CRC errors");
ET_SYSCTL_STAT_ADD64(ctx, children, "mcast_frames",
&stats->rx_mcast, "Multicast frames");
ET_SYSCTL_STAT_ADD64(ctx, children, "bcast_frames",
&stats->rx_bcast, "Broadcast frames");
ET_SYSCTL_STAT_ADD32(ctx, children, "control",
&stats->rx_control, "Control frames");
ET_SYSCTL_STAT_ADD32(ctx, children, "pause",
&stats->rx_pause, "Pause frames");
ET_SYSCTL_STAT_ADD32(ctx, children, "unknown_control",
&stats->rx_unknown_control, "Unknown control frames");
ET_SYSCTL_STAT_ADD32(ctx, children, "align_errs",
&stats->rx_alignerrs, "Alignment errors");
ET_SYSCTL_STAT_ADD32(ctx, children, "len_errs",
&stats->rx_lenerrs, "Frames with length mismatched");
ET_SYSCTL_STAT_ADD32(ctx, children, "code_errs",
&stats->rx_codeerrs, "Frames with code error");
ET_SYSCTL_STAT_ADD32(ctx, children, "cs_errs",
&stats->rx_cserrs, "Frames with carrier sense error");
ET_SYSCTL_STAT_ADD32(ctx, children, "runts",
&stats->rx_runts, "Too short frames");
ET_SYSCTL_STAT_ADD64(ctx, children, "oversize",
&stats->rx_oversize, "Oversized frames");
ET_SYSCTL_STAT_ADD32(ctx, children, "fragments",
&stats->rx_fragments, "Fragmented frames");
ET_SYSCTL_STAT_ADD32(ctx, children, "jabbers",
&stats->rx_jabbers, "Frames with jabber error");
ET_SYSCTL_STAT_ADD32(ctx, children, "drop",
&stats->rx_drop, "Dropped frames");
/* TX statistics. */
tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "tx", CTLFLAG_RD,
NULL, "TX MAC statistics");
children = SYSCTL_CHILDREN(tree);
ET_SYSCTL_STAT_ADD64(ctx, children, "bytes",
&stats->tx_bytes, "Good bytes");
ET_SYSCTL_STAT_ADD64(ctx, children, "frames",
&stats->tx_frames, "Good frames");
ET_SYSCTL_STAT_ADD64(ctx, children, "mcast_frames",
&stats->tx_mcast, "Multicast frames");
ET_SYSCTL_STAT_ADD64(ctx, children, "bcast_frames",
&stats->tx_bcast, "Broadcast frames");
ET_SYSCTL_STAT_ADD32(ctx, children, "pause",
&stats->tx_pause, "Pause frames");
ET_SYSCTL_STAT_ADD32(ctx, children, "deferred",
&stats->tx_deferred, "Deferred frames");
ET_SYSCTL_STAT_ADD32(ctx, children, "excess_deferred",
&stats->tx_excess_deferred, "Excessively deferred frames");
ET_SYSCTL_STAT_ADD32(ctx, children, "single_colls",
&stats->tx_single_colls, "Single collisions");
ET_SYSCTL_STAT_ADD32(ctx, children, "multi_colls",
&stats->tx_multi_colls, "Multiple collisions");
ET_SYSCTL_STAT_ADD32(ctx, children, "late_colls",
&stats->tx_late_colls, "Late collisions");
ET_SYSCTL_STAT_ADD32(ctx, children, "excess_colls",
&stats->tx_excess_colls, "Excess collisions");
ET_SYSCTL_STAT_ADD32(ctx, children, "total_colls",
&stats->tx_total_colls, "Total collisions");
ET_SYSCTL_STAT_ADD32(ctx, children, "pause_honored",
&stats->tx_pause_honored, "Honored pause frames");
ET_SYSCTL_STAT_ADD32(ctx, children, "drop",
&stats->tx_drop, "Dropped frames");
ET_SYSCTL_STAT_ADD32(ctx, children, "jabbers",
&stats->tx_jabbers, "Frames with jabber errors");
ET_SYSCTL_STAT_ADD32(ctx, children, "crc_errs",
&stats->tx_crcerrs, "Frames with CRC errors");
ET_SYSCTL_STAT_ADD32(ctx, children, "control",
&stats->tx_control, "Control frames");
ET_SYSCTL_STAT_ADD64(ctx, children, "oversize",
&stats->tx_oversize, "Oversized frames");
ET_SYSCTL_STAT_ADD32(ctx, children, "undersize",
&stats->tx_undersize, "Undersized frames");
ET_SYSCTL_STAT_ADD32(ctx, children, "fragments",
&stats->tx_fragments, "Fragmented frames");
}
#undef ET_SYSCTL_STAT_ADD32
#undef ET_SYSCTL_STAT_ADD64
static int
et_sysctl_rx_intr_npkts(SYSCTL_HANDLER_ARGS)
{
2012-01-10 20:52:02 +00:00
struct et_softc *sc;
struct ifnet *ifp;
int error, v;
2012-01-10 20:52:02 +00:00
sc = arg1;
ifp = sc->ifp;
v = sc->sc_rx_intr_npkts;
error = sysctl_handle_int(oidp, &v, 0, req);
if (error || req->newptr == NULL)
goto back;
if (v <= 0) {
error = EINVAL;
goto back;
}
if (sc->sc_rx_intr_npkts != v) {
if (ifp->if_drv_flags & IFF_DRV_RUNNING)
CSR_WRITE_4(sc, ET_RX_INTR_NPKTS, v);
sc->sc_rx_intr_npkts = v;
}
back:
2009-11-19 21:53:21 +00:00
return (error);
}
static int
et_sysctl_rx_intr_delay(SYSCTL_HANDLER_ARGS)
{
2012-01-10 20:52:02 +00:00
struct et_softc *sc;
struct ifnet *ifp;
int error, v;
2012-01-10 20:52:02 +00:00
sc = arg1;
ifp = sc->ifp;
v = sc->sc_rx_intr_delay;
error = sysctl_handle_int(oidp, &v, 0, req);
if (error || req->newptr == NULL)
goto back;
if (v <= 0) {
error = EINVAL;
goto back;
}
if (sc->sc_rx_intr_delay != v) {
if (ifp->if_drv_flags & IFF_DRV_RUNNING)
CSR_WRITE_4(sc, ET_RX_INTR_DELAY, v);
sc->sc_rx_intr_delay = v;
}
back:
2009-11-19 21:53:21 +00:00
return (error);
}
static void
et_stats_update(struct et_softc *sc)
{
struct et_hw_stats *stats;
stats = &sc->sc_stats;
stats->pkts_64 += CSR_READ_4(sc, ET_STAT_PKTS_64);
stats->pkts_65 += CSR_READ_4(sc, ET_STAT_PKTS_65_127);
stats->pkts_128 += CSR_READ_4(sc, ET_STAT_PKTS_128_255);
stats->pkts_256 += CSR_READ_4(sc, ET_STAT_PKTS_256_511);
stats->pkts_512 += CSR_READ_4(sc, ET_STAT_PKTS_512_1023);
stats->pkts_1024 += CSR_READ_4(sc, ET_STAT_PKTS_1024_1518);
stats->pkts_1519 += CSR_READ_4(sc, ET_STAT_PKTS_1519_1522);
stats->rx_bytes += CSR_READ_4(sc, ET_STAT_RX_BYTES);
stats->rx_frames += CSR_READ_4(sc, ET_STAT_RX_FRAMES);
stats->rx_crcerrs += CSR_READ_4(sc, ET_STAT_RX_CRC_ERR);
stats->rx_mcast += CSR_READ_4(sc, ET_STAT_RX_MCAST);
stats->rx_bcast += CSR_READ_4(sc, ET_STAT_RX_BCAST);
stats->rx_control += CSR_READ_4(sc, ET_STAT_RX_CTL);
stats->rx_pause += CSR_READ_4(sc, ET_STAT_RX_PAUSE);
stats->rx_unknown_control += CSR_READ_4(sc, ET_STAT_RX_UNKNOWN_CTL);
stats->rx_alignerrs += CSR_READ_4(sc, ET_STAT_RX_ALIGN_ERR);
stats->rx_lenerrs += CSR_READ_4(sc, ET_STAT_RX_LEN_ERR);
stats->rx_codeerrs += CSR_READ_4(sc, ET_STAT_RX_CODE_ERR);
stats->rx_cserrs += CSR_READ_4(sc, ET_STAT_RX_CS_ERR);
stats->rx_runts += CSR_READ_4(sc, ET_STAT_RX_RUNT);
stats->rx_oversize += CSR_READ_4(sc, ET_STAT_RX_OVERSIZE);
stats->rx_fragments += CSR_READ_4(sc, ET_STAT_RX_FRAG);
stats->rx_jabbers += CSR_READ_4(sc, ET_STAT_RX_JABBER);
stats->rx_drop += CSR_READ_4(sc, ET_STAT_RX_DROP);
stats->tx_bytes += CSR_READ_4(sc, ET_STAT_TX_BYTES);
stats->tx_frames += CSR_READ_4(sc, ET_STAT_TX_FRAMES);
stats->tx_mcast += CSR_READ_4(sc, ET_STAT_TX_MCAST);
stats->tx_bcast += CSR_READ_4(sc, ET_STAT_TX_BCAST);
stats->tx_pause += CSR_READ_4(sc, ET_STAT_TX_PAUSE);
stats->tx_deferred += CSR_READ_4(sc, ET_STAT_TX_DEFER);
stats->tx_excess_deferred += CSR_READ_4(sc, ET_STAT_TX_EXCESS_DEFER);
stats->tx_single_colls += CSR_READ_4(sc, ET_STAT_TX_SINGLE_COL);
stats->tx_multi_colls += CSR_READ_4(sc, ET_STAT_TX_MULTI_COL);
stats->tx_late_colls += CSR_READ_4(sc, ET_STAT_TX_LATE_COL);
stats->tx_excess_colls += CSR_READ_4(sc, ET_STAT_TX_EXCESS_COL);
stats->tx_total_colls += CSR_READ_4(sc, ET_STAT_TX_TOTAL_COL);
stats->tx_pause_honored += CSR_READ_4(sc, ET_STAT_TX_PAUSE_HONOR);
stats->tx_drop += CSR_READ_4(sc, ET_STAT_TX_DROP);
stats->tx_jabbers += CSR_READ_4(sc, ET_STAT_TX_JABBER);
stats->tx_crcerrs += CSR_READ_4(sc, ET_STAT_TX_CRC_ERR);
stats->tx_control += CSR_READ_4(sc, ET_STAT_TX_CTL);
stats->tx_oversize += CSR_READ_4(sc, ET_STAT_TX_OVERSIZE);
stats->tx_undersize += CSR_READ_4(sc, ET_STAT_TX_UNDERSIZE);
stats->tx_fragments += CSR_READ_4(sc, ET_STAT_TX_FRAG);
}
static uint64_t
et_get_counter(struct ifnet *ifp, ift_counter cnt)
{
struct et_softc *sc;
struct et_hw_stats *stats;
sc = if_getsoftc(ifp);
stats = &sc->sc_stats;
switch (cnt) {
case IFCOUNTER_OPACKETS:
return (stats->tx_frames);
case IFCOUNTER_COLLISIONS:
return (stats->tx_total_colls);
case IFCOUNTER_OERRORS:
return (stats->tx_drop + stats->tx_jabbers +
stats->tx_crcerrs + stats->tx_excess_deferred +
stats->tx_late_colls);
case IFCOUNTER_IPACKETS:
return (stats->rx_frames);
case IFCOUNTER_IERRORS:
return (stats->rx_crcerrs + stats->rx_alignerrs +
stats->rx_lenerrs + stats->rx_codeerrs + stats->rx_cserrs +
stats->rx_runts + stats->rx_jabbers + stats->rx_drop);
default:
return (if_get_counter_default(ifp, cnt));
}
}
static int
et_suspend(device_t dev)
{
struct et_softc *sc;
uint32_t pmcfg;
sc = device_get_softc(dev);
ET_LOCK(sc);
if ((sc->ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
et_stop(sc);
/* Diable all clocks and put PHY into COMA. */
pmcfg = CSR_READ_4(sc, ET_PM);
pmcfg &= ~(EM_PM_GIGEPHY_ENB | ET_PM_SYSCLK_GATE | ET_PM_TXCLK_GATE |
ET_PM_RXCLK_GATE);
pmcfg |= ET_PM_PHY_SW_COMA;
CSR_WRITE_4(sc, ET_PM, pmcfg);
ET_UNLOCK(sc);
return (0);
}
static int
et_resume(device_t dev)
{
struct et_softc *sc;
uint32_t pmcfg;
sc = device_get_softc(dev);
ET_LOCK(sc);
/* Take PHY out of COMA and enable clocks. */
pmcfg = ET_PM_SYSCLK_GATE | ET_PM_TXCLK_GATE | ET_PM_RXCLK_GATE;
if ((sc->sc_flags & ET_FLAG_FASTETHER) == 0)
pmcfg |= EM_PM_GIGEPHY_ENB;
CSR_WRITE_4(sc, ET_PM, pmcfg);
if ((sc->ifp->if_flags & IFF_UP) != 0)
et_init_locked(sc);
ET_UNLOCK(sc);
return (0);
}