1 /**************************************************************************
3 Copyright (c) 2007-2009, Chelsio Inc.
6 Redistribution and use in source and binary forms, with or without
7 modification, are permitted provided that the following conditions are met:
9 1. Redistributions of source code must retain the above copyright notice,
10 this list of conditions and the following disclaimer.
12 2. Neither the name of the Chelsio Corporation nor the names of its
13 contributors may be used to endorse or promote products derived from
14 this software without specific prior written permission.
16 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
17 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
20 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
21 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
22 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
23 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
24 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
25 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
26 POSSIBILITY OF SUCH DAMAGE.
28 ***************************************************************************/
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/kernel.h>
39 #include <sys/module.h>
40 #include <sys/pciio.h>
42 #include <machine/bus.h>
43 #include <machine/resource.h>
44 #include <sys/bus_dma.h>
47 #include <sys/ioccom.h>
49 #include <sys/linker.h>
50 #include <sys/firmware.h>
51 #include <sys/socket.h>
52 #include <sys/sockio.h>
54 #include <sys/sysctl.h>
55 #include <sys/syslog.h>
56 #include <sys/queue.h>
57 #include <sys/taskqueue.h>
61 #include <net/ethernet.h>
63 #include <net/if_arp.h>
64 #include <net/if_dl.h>
65 #include <net/if_media.h>
66 #include <net/if_types.h>
67 #include <net/if_vlan_var.h>
69 #include <netinet/in_systm.h>
70 #include <netinet/in.h>
71 #include <netinet/if_ether.h>
72 #include <netinet/ip.h>
73 #include <netinet/ip.h>
74 #include <netinet/tcp.h>
75 #include <netinet/udp.h>
77 #include <dev/pci/pcireg.h>
78 #include <dev/pci/pcivar.h>
79 #include <dev/pci/pci_private.h>
81 #include <cxgb_include.h>
87 static int cxgb_setup_interrupts(adapter_t *);
88 static void cxgb_teardown_interrupts(adapter_t *);
89 static void cxgb_init(void *);
90 static int cxgb_init_locked(struct port_info *);
91 static int cxgb_uninit_locked(struct port_info *);
92 static int cxgb_uninit_synchronized(struct port_info *);
93 static int cxgb_ioctl(struct ifnet *, unsigned long, caddr_t);
94 static int cxgb_media_change(struct ifnet *);
95 static int cxgb_ifm_type(int);
96 static void cxgb_build_medialist(struct port_info *);
97 static void cxgb_media_status(struct ifnet *, struct ifmediareq *);
98 static int setup_sge_qsets(adapter_t *);
99 static void cxgb_async_intr(void *);
100 static void cxgb_tick_handler(void *, int);
101 static void cxgb_tick(void *);
102 static void link_check_callout(void *);
103 static void check_link_status(void *, int);
104 static void setup_rss(adapter_t *sc);
105 static int alloc_filters(struct adapter *);
106 static int setup_hw_filters(struct adapter *);
107 static int set_filter(struct adapter *, int, const struct filter_info *);
108 static inline void mk_set_tcb_field(struct cpl_set_tcb_field *, unsigned int,
109 unsigned int, u64, u64);
110 static inline void set_tcb_field_ulp(struct cpl_set_tcb_field *, unsigned int,
111 unsigned int, u64, u64);
113 static int cpl_not_handled(struct sge_qset *, struct rsp_desc *, struct mbuf *);
116 /* Attachment glue for the PCI controller end of the device. Each port of
117 * the device is attached separately, as defined later.
119 static int cxgb_controller_probe(device_t);
120 static int cxgb_controller_attach(device_t);
121 static int cxgb_controller_detach(device_t);
122 static void cxgb_free(struct adapter *);
123 static __inline void reg_block_dump(struct adapter *ap, uint8_t *buf, unsigned int start,
125 static void cxgb_get_regs(adapter_t *sc, struct ch_ifconf_regs *regs, uint8_t *buf);
126 static int cxgb_get_regs_len(void);
127 static void touch_bars(device_t dev);
128 static void cxgb_update_mac_settings(struct port_info *p);
130 static int toe_capability(struct port_info *, int);
133 static device_method_t cxgb_controller_methods[] = {
134 DEVMETHOD(device_probe, cxgb_controller_probe),
135 DEVMETHOD(device_attach, cxgb_controller_attach),
136 DEVMETHOD(device_detach, cxgb_controller_detach),
141 static driver_t cxgb_controller_driver = {
143 cxgb_controller_methods,
144 sizeof(struct adapter)
147 static int cxgbc_mod_event(module_t, int, void *);
148 static devclass_t cxgb_controller_devclass;
149 DRIVER_MODULE(cxgbc, pci, cxgb_controller_driver, cxgb_controller_devclass,
151 MODULE_VERSION(cxgbc, 1);
152 MODULE_DEPEND(cxgbc, firmware, 1, 1, 1);
155 * Attachment glue for the ports. Attachment is done directly to the
158 static int cxgb_port_probe(device_t);
159 static int cxgb_port_attach(device_t);
160 static int cxgb_port_detach(device_t);
162 static device_method_t cxgb_port_methods[] = {
163 DEVMETHOD(device_probe, cxgb_port_probe),
164 DEVMETHOD(device_attach, cxgb_port_attach),
165 DEVMETHOD(device_detach, cxgb_port_detach),
169 static driver_t cxgb_port_driver = {
175 static d_ioctl_t cxgb_extension_ioctl;
176 static d_open_t cxgb_extension_open;
177 static d_close_t cxgb_extension_close;
179 static struct cdevsw cxgb_cdevsw = {
180 .d_version = D_VERSION,
182 .d_open = cxgb_extension_open,
183 .d_close = cxgb_extension_close,
184 .d_ioctl = cxgb_extension_ioctl,
188 static devclass_t cxgb_port_devclass;
189 DRIVER_MODULE(cxgb, cxgbc, cxgb_port_driver, cxgb_port_devclass, 0, 0);
190 MODULE_VERSION(cxgb, 1);
192 static struct mtx t3_list_lock;
193 static SLIST_HEAD(, adapter) t3_list;
195 static struct mtx t3_uld_list_lock;
196 static SLIST_HEAD(, uld_info) t3_uld_list;
200 * The driver uses the best interrupt scheme available on a platform in the
201 * order MSI-X, MSI, legacy pin interrupts. This parameter determines which
202 * of these schemes the driver may consider as follows:
204 * msi = 2: choose from among all three options
205 * msi = 1 : only consider MSI and pin interrupts
206 * msi = 0: force pin interrupts
208 static int msi_allowed = 2;
210 TUNABLE_INT("hw.cxgb.msi_allowed", &msi_allowed);
211 SYSCTL_NODE(_hw, OID_AUTO, cxgb, CTLFLAG_RD, 0, "CXGB driver parameters");
212 SYSCTL_INT(_hw_cxgb, OID_AUTO, msi_allowed, CTLFLAG_RDTUN, &msi_allowed, 0,
213 "MSI-X, MSI, INTx selector");
216 * The driver uses an auto-queue algorithm by default.
217 * To disable it and force a single queue-set per port, use multiq = 0
219 static int multiq = 1;
220 TUNABLE_INT("hw.cxgb.multiq", &multiq);
221 SYSCTL_INT(_hw_cxgb, OID_AUTO, multiq, CTLFLAG_RDTUN, &multiq, 0,
222 "use min(ncpus/ports, 8) queue-sets per port");
225 * By default the driver will not update the firmware unless
226 * it was compiled against a newer version
229 static int force_fw_update = 0;
230 TUNABLE_INT("hw.cxgb.force_fw_update", &force_fw_update);
231 SYSCTL_INT(_hw_cxgb, OID_AUTO, force_fw_update, CTLFLAG_RDTUN, &force_fw_update, 0,
232 "update firmware even if up to date");
234 int cxgb_use_16k_clusters = -1;
235 TUNABLE_INT("hw.cxgb.use_16k_clusters", &cxgb_use_16k_clusters);
236 SYSCTL_INT(_hw_cxgb, OID_AUTO, use_16k_clusters, CTLFLAG_RDTUN,
237 &cxgb_use_16k_clusters, 0, "use 16kB clusters for the jumbo queue ");
239 static int nfilters = -1;
240 TUNABLE_INT("hw.cxgb.nfilters", &nfilters);
241 SYSCTL_INT(_hw_cxgb, OID_AUTO, nfilters, CTLFLAG_RDTUN,
242 &nfilters, 0, "max number of entries in the filter table");
245 MAX_TXQ_ENTRIES = 16384,
246 MAX_CTRL_TXQ_ENTRIES = 1024,
247 MAX_RSPQ_ENTRIES = 16384,
248 MAX_RX_BUFFERS = 16384,
249 MAX_RX_JUMBO_BUFFERS = 16384,
251 MIN_CTRL_TXQ_ENTRIES = 4,
252 MIN_RSPQ_ENTRIES = 32,
254 MIN_FL_JUMBO_ENTRIES = 32
269 u32 report_filter_id:1;
277 enum { FILTER_NO_VLAN_PRI = 7 };
279 #define EEPROM_MAGIC 0x38E2F10C
281 #define PORT_MASK ((1 << MAX_NPORTS) - 1)
283 /* Table for probing the cards. The desc field isn't actually used */
289 } cxgb_identifiers[] = {
290 {PCI_VENDOR_ID_CHELSIO, 0x0020, 0, "PE9000"},
291 {PCI_VENDOR_ID_CHELSIO, 0x0021, 1, "T302E"},
292 {PCI_VENDOR_ID_CHELSIO, 0x0022, 2, "T310E"},
293 {PCI_VENDOR_ID_CHELSIO, 0x0023, 3, "T320X"},
294 {PCI_VENDOR_ID_CHELSIO, 0x0024, 1, "T302X"},
295 {PCI_VENDOR_ID_CHELSIO, 0x0025, 3, "T320E"},
296 {PCI_VENDOR_ID_CHELSIO, 0x0026, 2, "T310X"},
297 {PCI_VENDOR_ID_CHELSIO, 0x0030, 2, "T3B10"},
298 {PCI_VENDOR_ID_CHELSIO, 0x0031, 3, "T3B20"},
299 {PCI_VENDOR_ID_CHELSIO, 0x0032, 1, "T3B02"},
300 {PCI_VENDOR_ID_CHELSIO, 0x0033, 4, "T3B04"},
301 {PCI_VENDOR_ID_CHELSIO, 0x0035, 6, "T3C10"},
302 {PCI_VENDOR_ID_CHELSIO, 0x0036, 3, "S320E-CR"},
303 {PCI_VENDOR_ID_CHELSIO, 0x0037, 7, "N320E-G2"},
307 static int set_eeprom(struct port_info *pi, const uint8_t *data, int len, int offset);
311 t3rev2char(struct adapter *adapter)
315 switch(adapter->params.rev) {
330 static struct cxgb_ident *
331 cxgb_get_ident(device_t dev)
333 struct cxgb_ident *id;
335 for (id = cxgb_identifiers; id->desc != NULL; id++) {
336 if ((id->vendor == pci_get_vendor(dev)) &&
337 (id->device == pci_get_device(dev))) {
344 static const struct adapter_info *
345 cxgb_get_adapter_info(device_t dev)
347 struct cxgb_ident *id;
348 const struct adapter_info *ai;
350 id = cxgb_get_ident(dev);
354 ai = t3_get_adapter_info(id->index);
360 cxgb_controller_probe(device_t dev)
362 const struct adapter_info *ai;
363 char *ports, buf[80];
366 ai = cxgb_get_adapter_info(dev);
370 nports = ai->nports0 + ai->nports1;
376 snprintf(buf, sizeof(buf), "%s, %d %s", ai->desc, nports, ports);
377 device_set_desc_copy(dev, buf);
378 return (BUS_PROBE_DEFAULT);
381 #define FW_FNAME "cxgb_t3fw"
382 #define TPEEPROM_NAME "cxgb_t3%c_tp_eeprom"
383 #define TPSRAM_NAME "cxgb_t3%c_protocol_sram"
386 upgrade_fw(adapter_t *sc)
388 const struct firmware *fw;
392 if ((fw = firmware_get(FW_FNAME)) == NULL) {
393 device_printf(sc->dev, "Could not find firmware image %s\n", FW_FNAME);
396 device_printf(sc->dev, "installing firmware on card\n");
397 status = t3_load_fw(sc, (const uint8_t *)fw->data, fw->datasize);
400 device_printf(sc->dev, "failed to install firmware: %d\n",
403 t3_get_fw_version(sc, &vers);
404 snprintf(&sc->fw_version[0], sizeof(sc->fw_version), "%d.%d.%d",
405 G_FW_VERSION_MAJOR(vers), G_FW_VERSION_MINOR(vers),
406 G_FW_VERSION_MICRO(vers));
409 firmware_put(fw, FIRMWARE_UNLOAD);
415 * The cxgb_controller_attach function is responsible for the initial
416 * bringup of the device. Its responsibilities include:
418 * 1. Determine if the device supports MSI or MSI-X.
419 * 2. Allocate bus resources so that we can access the Base Address Register
420 * 3. Create and initialize mutexes for the controller and its control
421 * logic such as SGE and MDIO.
422 * 4. Call hardware specific setup routine for the adapter as a whole.
423 * 5. Allocate the BAR for doing MSI-X.
424 * 6. Setup the line interrupt iff MSI-X is not supported.
425 * 7. Create the driver's taskq.
426 * 8. Start one task queue service thread.
427 * 9. Check if the firmware and SRAM are up-to-date. They will be
428 * auto-updated later (before FULL_INIT_DONE), if required.
429 * 10. Create a child device for each MAC (port)
430 * 11. Initialize T3 private state.
431 * 12. Trigger the LED
432 * 13. Setup offload iff supported.
433 * 14. Reset/restart the tick callout.
436 * NOTE: Any modification or deviation from this list MUST be reflected in
437 * the above comment. Failure to do so will result in problems on various
438 * error conditions including link flapping.
441 cxgb_controller_attach(device_t dev)
444 const struct adapter_info *ai;
452 sc = device_get_softc(dev);
455 ai = cxgb_get_adapter_info(dev);
457 snprintf(sc->lockbuf, ADAPTER_LOCK_NAME_LEN, "cxgb controller lock %d",
458 device_get_unit(dev));
459 ADAPTER_LOCK_INIT(sc, sc->lockbuf);
461 snprintf(sc->reglockbuf, ADAPTER_LOCK_NAME_LEN, "SGE reg lock %d",
462 device_get_unit(dev));
463 snprintf(sc->mdiolockbuf, ADAPTER_LOCK_NAME_LEN, "cxgb mdio lock %d",
464 device_get_unit(dev));
465 snprintf(sc->elmerlockbuf, ADAPTER_LOCK_NAME_LEN, "cxgb elmer lock %d",
466 device_get_unit(dev));
468 MTX_INIT(&sc->sge.reg_lock, sc->reglockbuf, NULL, MTX_SPIN);
469 MTX_INIT(&sc->mdio_lock, sc->mdiolockbuf, NULL, MTX_DEF);
470 MTX_INIT(&sc->elmer_lock, sc->elmerlockbuf, NULL, MTX_DEF);
472 mtx_lock(&t3_list_lock);
473 SLIST_INSERT_HEAD(&t3_list, sc, link);
474 mtx_unlock(&t3_list_lock);
476 /* find the PCIe link width and set max read request to 4KB*/
477 if (pci_find_cap(dev, PCIY_EXPRESS, ®) == 0) {
480 lnk = pci_read_config(dev, reg + PCIER_LINK_STA, 2);
481 sc->link_width = (lnk & PCIEM_LINK_STA_WIDTH) >> 4;
482 if (sc->link_width < 8 &&
483 (ai->caps & SUPPORTED_10000baseT_Full)) {
484 device_printf(sc->dev,
485 "PCIe x%d Link, expect reduced performance\n",
489 pci_set_max_read_req(dev, 4096);
493 pci_enable_busmaster(dev);
495 * Allocate the registers and make them available to the driver.
496 * The registers that we care about for NIC mode are in BAR 0
498 sc->regs_rid = PCIR_BAR(0);
499 if ((sc->regs_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
500 &sc->regs_rid, RF_ACTIVE)) == NULL) {
501 device_printf(dev, "Cannot allocate BAR region 0\n");
506 sc->bt = rman_get_bustag(sc->regs_res);
507 sc->bh = rman_get_bushandle(sc->regs_res);
508 sc->mmio_len = rman_get_size(sc->regs_res);
510 for (i = 0; i < MAX_NPORTS; i++)
511 sc->port[i].adapter = sc;
513 if (t3_prep_adapter(sc, ai, 1) < 0) {
514 printf("prep adapter failed\n");
519 sc->udbs_rid = PCIR_BAR(2);
521 if (is_offload(sc) &&
522 ((sc->udbs_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
523 &sc->udbs_rid, RF_ACTIVE)) == NULL)) {
524 device_printf(dev, "Cannot allocate BAR region 1\n");
529 /* Allocate the BAR for doing MSI-X. If it succeeds, try to allocate
530 * enough messages for the queue sets. If that fails, try falling
531 * back to MSI. If that fails, then try falling back to the legacy
532 * interrupt pin model.
534 sc->msix_regs_rid = 0x20;
535 if ((msi_allowed >= 2) &&
536 (sc->msix_regs_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
537 &sc->msix_regs_rid, RF_ACTIVE)) != NULL) {
540 port_qsets = min(SGE_QSETS/sc->params.nports, mp_ncpus);
541 msi_needed = sc->msi_count = sc->params.nports * port_qsets + 1;
543 if (pci_msix_count(dev) == 0 ||
544 (error = pci_alloc_msix(dev, &sc->msi_count)) != 0 ||
545 sc->msi_count != msi_needed) {
546 device_printf(dev, "alloc msix failed - "
547 "msi_count=%d, msi_needed=%d, err=%d; "
548 "will try MSI\n", sc->msi_count,
552 pci_release_msi(dev);
553 bus_release_resource(dev, SYS_RES_MEMORY,
554 sc->msix_regs_rid, sc->msix_regs_res);
555 sc->msix_regs_res = NULL;
557 sc->flags |= USING_MSIX;
558 sc->cxgb_intr = cxgb_async_intr;
560 "using MSI-X interrupts (%u vectors)\n",
565 if ((msi_allowed >= 1) && (sc->msi_count == 0)) {
567 if ((error = pci_alloc_msi(dev, &sc->msi_count)) != 0) {
568 device_printf(dev, "alloc msi failed - "
569 "err=%d; will try INTx\n", error);
572 pci_release_msi(dev);
574 sc->flags |= USING_MSI;
575 sc->cxgb_intr = t3_intr_msi;
576 device_printf(dev, "using MSI interrupts\n");
579 if (sc->msi_count == 0) {
580 device_printf(dev, "using line interrupts\n");
581 sc->cxgb_intr = t3b_intr;
584 /* Create a private taskqueue thread for handling driver events */
585 sc->tq = taskqueue_create("cxgb_taskq", M_NOWAIT,
586 taskqueue_thread_enqueue, &sc->tq);
587 if (sc->tq == NULL) {
588 device_printf(dev, "failed to allocate controller task queue\n");
592 taskqueue_start_threads(&sc->tq, 1, PI_NET, "%s taskq",
593 device_get_nameunit(dev));
594 TASK_INIT(&sc->tick_task, 0, cxgb_tick_handler, sc);
597 /* Create a periodic callout for checking adapter status */
598 callout_init(&sc->cxgb_tick_ch, TRUE);
600 if (t3_check_fw_version(sc) < 0 || force_fw_update) {
602 * Warn user that a firmware update will be attempted in init.
604 device_printf(dev, "firmware needs to be updated to version %d.%d.%d\n",
605 FW_VERSION_MAJOR, FW_VERSION_MINOR, FW_VERSION_MICRO);
606 sc->flags &= ~FW_UPTODATE;
608 sc->flags |= FW_UPTODATE;
611 if (t3_check_tpsram_version(sc) < 0) {
613 * Warn user that a firmware update will be attempted in init.
615 device_printf(dev, "SRAM needs to be updated to version %c-%d.%d.%d\n",
616 t3rev2char(sc), TP_VERSION_MAJOR, TP_VERSION_MINOR, TP_VERSION_MICRO);
617 sc->flags &= ~TPS_UPTODATE;
619 sc->flags |= TPS_UPTODATE;
623 * Create a child device for each MAC. The ethernet attachment
624 * will be done in these children.
626 for (i = 0; i < (sc)->params.nports; i++) {
627 struct port_info *pi;
629 if ((child = device_add_child(dev, "cxgb", -1)) == NULL) {
630 device_printf(dev, "failed to add child port\n");
636 pi->nqsets = port_qsets;
637 pi->first_qset = i*port_qsets;
639 pi->tx_chan = i >= ai->nports0;
640 pi->txpkt_intf = pi->tx_chan ? 2 * (i - ai->nports0) + 1 : 2 * i;
641 sc->rxpkt_map[pi->txpkt_intf] = i;
642 sc->port[i].tx_chan = i >= ai->nports0;
643 sc->portdev[i] = child;
644 device_set_softc(child, pi);
646 if ((error = bus_generic_attach(dev)) != 0)
649 /* initialize sge private state */
650 t3_sge_init_adapter(sc);
654 error = t3_get_fw_version(sc, &vers);
658 snprintf(&sc->fw_version[0], sizeof(sc->fw_version), "%d.%d.%d",
659 G_FW_VERSION_MAJOR(vers), G_FW_VERSION_MINOR(vers),
660 G_FW_VERSION_MICRO(vers));
662 snprintf(buf, sizeof(buf), "%s %sNIC\t E/C: %s S/N: %s",
663 ai->desc, is_offload(sc) ? "R" : "",
664 sc->params.vpd.ec, sc->params.vpd.sn);
665 device_set_desc_copy(dev, buf);
667 snprintf(&sc->port_types[0], sizeof(sc->port_types), "%x%x%x%x",
668 sc->params.vpd.port_type[0], sc->params.vpd.port_type[1],
669 sc->params.vpd.port_type[2], sc->params.vpd.port_type[3]);
671 device_printf(sc->dev, "Firmware Version %s\n", &sc->fw_version[0]);
672 callout_reset(&sc->cxgb_tick_ch, hz, cxgb_tick, sc);
673 t3_add_attach_sysctls(sc);
676 for (i = 0; i < NUM_CPL_HANDLERS; i++)
677 sc->cpl_handler[i] = cpl_not_handled;
681 error = cxgb_setup_interrupts(sc);
690 * The cxgb_controller_detach routine is called with the device is
691 * unloaded from the system.
695 cxgb_controller_detach(device_t dev)
699 sc = device_get_softc(dev);
707 * The cxgb_free() is called by the cxgb_controller_detach() routine
708 * to tear down the structures that were built up in
709 * cxgb_controller_attach(), and should be the final piece of work
710 * done when fully unloading the driver.
713 * 1. Shutting down the threads started by the cxgb_controller_attach()
715 * 2. Stopping the lower level device and all callouts (cxgb_down_locked()).
716 * 3. Detaching all of the port devices created during the
717 * cxgb_controller_attach() routine.
718 * 4. Removing the device children created via cxgb_controller_attach().
719 * 5. Releasing PCI resources associated with the device.
720 * 6. Turning off the offload support, iff it was turned on.
721 * 7. Destroying the mutexes created in cxgb_controller_attach().
725 cxgb_free(struct adapter *sc)
730 sc->flags |= CXGB_SHUTDOWN;
734 * Make sure all child devices are gone.
736 bus_generic_detach(sc->dev);
737 for (i = 0; i < (sc)->params.nports; i++) {
738 if (sc->portdev[i] &&
739 device_delete_child(sc->dev, sc->portdev[i]) != 0)
740 device_printf(sc->dev, "failed to delete child port\n");
741 nqsets += sc->port[i].nqsets;
745 * At this point, it is as if cxgb_port_detach has run on all ports, and
746 * cxgb_down has run on the adapter. All interrupts have been silenced,
747 * all open devices have been closed.
749 KASSERT(sc->open_device_map == 0, ("%s: device(s) still open (%x)",
750 __func__, sc->open_device_map));
751 for (i = 0; i < sc->params.nports; i++) {
752 KASSERT(sc->port[i].ifp == NULL, ("%s: port %i undead!",
757 * Finish off the adapter's callouts.
759 callout_drain(&sc->cxgb_tick_ch);
760 callout_drain(&sc->sge_timer_ch);
763 * Release resources grabbed under FULL_INIT_DONE by cxgb_up. The
764 * sysctls are cleaned up by the kernel linker.
766 if (sc->flags & FULL_INIT_DONE) {
767 t3_free_sge_resources(sc, nqsets);
768 sc->flags &= ~FULL_INIT_DONE;
772 * Release all interrupt resources.
774 cxgb_teardown_interrupts(sc);
775 if (sc->flags & (USING_MSI | USING_MSIX)) {
776 device_printf(sc->dev, "releasing msi message(s)\n");
777 pci_release_msi(sc->dev);
779 device_printf(sc->dev, "no msi message to release\n");
782 if (sc->msix_regs_res != NULL) {
783 bus_release_resource(sc->dev, SYS_RES_MEMORY, sc->msix_regs_rid,
788 * Free the adapter's taskqueue.
790 if (sc->tq != NULL) {
791 taskqueue_free(sc->tq);
795 free(sc->filters, M_DEVBUF);
798 if (sc->udbs_res != NULL)
799 bus_release_resource(sc->dev, SYS_RES_MEMORY, sc->udbs_rid,
802 if (sc->regs_res != NULL)
803 bus_release_resource(sc->dev, SYS_RES_MEMORY, sc->regs_rid,
806 MTX_DESTROY(&sc->mdio_lock);
807 MTX_DESTROY(&sc->sge.reg_lock);
808 MTX_DESTROY(&sc->elmer_lock);
809 mtx_lock(&t3_list_lock);
810 SLIST_REMOVE(&t3_list, sc, adapter, link);
811 mtx_unlock(&t3_list_lock);
812 ADAPTER_LOCK_DEINIT(sc);
816 * setup_sge_qsets - configure SGE Tx/Rx/response queues
817 * @sc: the controller softc
819 * Determines how many sets of SGE queues to use and initializes them.
820 * We support multiple queue sets per port if we have MSI-X, otherwise
821 * just one queue set per port.
824 setup_sge_qsets(adapter_t *sc)
826 int i, j, err, irq_idx = 0, qset_idx = 0;
827 u_int ntxq = SGE_TXQ_PER_SET;
829 if ((err = t3_sge_alloc(sc)) != 0) {
830 device_printf(sc->dev, "t3_sge_alloc returned %d\n", err);
834 if (sc->params.rev > 0 && !(sc->flags & USING_MSI))
837 for (i = 0; i < (sc)->params.nports; i++) {
838 struct port_info *pi = &sc->port[i];
840 for (j = 0; j < pi->nqsets; j++, qset_idx++) {
841 err = t3_sge_alloc_qset(sc, qset_idx, (sc)->params.nports,
842 (sc->flags & USING_MSIX) ? qset_idx + 1 : irq_idx,
843 &sc->params.sge.qset[qset_idx], ntxq, pi);
845 t3_free_sge_resources(sc, qset_idx);
846 device_printf(sc->dev,
847 "t3_sge_alloc_qset failed with %d\n", err);
857 cxgb_teardown_interrupts(adapter_t *sc)
861 for (i = 0; i < SGE_QSETS; i++) {
862 if (sc->msix_intr_tag[i] == NULL) {
864 /* Should have been setup fully or not at all */
865 KASSERT(sc->msix_irq_res[i] == NULL &&
866 sc->msix_irq_rid[i] == 0,
867 ("%s: half-done interrupt (%d).", __func__, i));
872 bus_teardown_intr(sc->dev, sc->msix_irq_res[i],
873 sc->msix_intr_tag[i]);
874 bus_release_resource(sc->dev, SYS_RES_IRQ, sc->msix_irq_rid[i],
875 sc->msix_irq_res[i]);
877 sc->msix_irq_res[i] = sc->msix_intr_tag[i] = NULL;
878 sc->msix_irq_rid[i] = 0;
882 KASSERT(sc->irq_res != NULL,
883 ("%s: half-done interrupt.", __func__));
885 bus_teardown_intr(sc->dev, sc->irq_res, sc->intr_tag);
886 bus_release_resource(sc->dev, SYS_RES_IRQ, sc->irq_rid,
889 sc->irq_res = sc->intr_tag = NULL;
895 cxgb_setup_interrupts(adapter_t *sc)
897 struct resource *res;
899 int i, rid, err, intr_flag = sc->flags & (USING_MSI | USING_MSIX);
901 sc->irq_rid = intr_flag ? 1 : 0;
902 sc->irq_res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &sc->irq_rid,
903 RF_SHAREABLE | RF_ACTIVE);
904 if (sc->irq_res == NULL) {
905 device_printf(sc->dev, "Cannot allocate interrupt (%x, %u)\n",
906 intr_flag, sc->irq_rid);
910 err = bus_setup_intr(sc->dev, sc->irq_res,
911 INTR_MPSAFE | INTR_TYPE_NET, NULL,
912 sc->cxgb_intr, sc, &sc->intr_tag);
915 device_printf(sc->dev,
916 "Cannot set up interrupt (%x, %u, %d)\n",
917 intr_flag, sc->irq_rid, err);
918 bus_release_resource(sc->dev, SYS_RES_IRQ, sc->irq_rid,
920 sc->irq_res = sc->intr_tag = NULL;
925 /* That's all for INTx or MSI */
926 if (!(intr_flag & USING_MSIX) || err)
929 bus_describe_intr(sc->dev, sc->irq_res, sc->intr_tag, "err");
930 for (i = 0; i < sc->msi_count - 1; i++) {
932 res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &rid,
933 RF_SHAREABLE | RF_ACTIVE);
935 device_printf(sc->dev, "Cannot allocate interrupt "
936 "for message %d\n", rid);
941 err = bus_setup_intr(sc->dev, res, INTR_MPSAFE | INTR_TYPE_NET,
942 NULL, t3_intr_msix, &sc->sge.qs[i], &tag);
944 device_printf(sc->dev, "Cannot set up interrupt "
945 "for message %d (%d)\n", rid, err);
946 bus_release_resource(sc->dev, SYS_RES_IRQ, rid, res);
950 sc->msix_irq_rid[i] = rid;
951 sc->msix_irq_res[i] = res;
952 sc->msix_intr_tag[i] = tag;
953 bus_describe_intr(sc->dev, res, tag, "qs%d", i);
957 cxgb_teardown_interrupts(sc);
964 cxgb_port_probe(device_t dev)
970 p = device_get_softc(dev);
972 snprintf(buf, sizeof(buf), "Port %d %s", p->port_id, desc);
973 device_set_desc_copy(dev, buf);
979 cxgb_makedev(struct port_info *pi)
982 pi->port_cdev = make_dev(&cxgb_cdevsw, pi->ifp->if_dunit,
983 UID_ROOT, GID_WHEEL, 0600, "%s", if_name(pi->ifp));
985 if (pi->port_cdev == NULL)
988 pi->port_cdev->si_drv1 = (void *)pi;
993 #define CXGB_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
994 IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
995 IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6)
996 #define CXGB_CAP_ENABLE CXGB_CAP
999 cxgb_port_attach(device_t dev)
1001 struct port_info *p;
1006 p = device_get_softc(dev);
1008 snprintf(p->lockbuf, PORT_NAME_LEN, "cxgb port lock %d:%d",
1009 device_get_unit(device_get_parent(dev)), p->port_id);
1010 PORT_LOCK_INIT(p, p->lockbuf);
1012 callout_init(&p->link_check_ch, CALLOUT_MPSAFE);
1013 TASK_INIT(&p->link_check_task, 0, check_link_status, p);
1015 /* Allocate an ifnet object and set it up */
1016 ifp = p->ifp = if_alloc(IFT_ETHER);
1018 device_printf(dev, "Cannot allocate ifnet\n");
1022 if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1023 ifp->if_init = cxgb_init;
1025 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1026 ifp->if_ioctl = cxgb_ioctl;
1027 ifp->if_transmit = cxgb_transmit;
1028 ifp->if_qflush = cxgb_qflush;
1030 ifp->if_capabilities = CXGB_CAP;
1033 ifp->if_capabilities |= IFCAP_TOE4;
1035 ifp->if_capenable = CXGB_CAP_ENABLE;
1036 ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
1037 CSUM_UDP_IPV6 | CSUM_TCP_IPV6;
1040 * Disable TSO on 4-port - it isn't supported by the firmware.
1042 if (sc->params.nports > 2) {
1043 ifp->if_capabilities &= ~(IFCAP_TSO | IFCAP_VLAN_HWTSO);
1044 ifp->if_capenable &= ~(IFCAP_TSO | IFCAP_VLAN_HWTSO);
1045 ifp->if_hwassist &= ~CSUM_TSO;
1048 ether_ifattach(ifp, p->hw_addr);
1050 #ifdef DEFAULT_JUMBO
1051 if (sc->params.nports <= 2)
1052 ifp->if_mtu = ETHERMTU_JUMBO;
1054 if ((err = cxgb_makedev(p)) != 0) {
1055 printf("makedev failed %d\n", err);
1059 /* Create a list of media supported by this port */
1060 ifmedia_init(&p->media, IFM_IMASK, cxgb_media_change,
1062 cxgb_build_medialist(p);
1064 t3_sge_init_port(p);
1070 * cxgb_port_detach() is called via the device_detach methods when
1071 * cxgb_free() calls the bus_generic_detach. It is responsible for
1072 * removing the device from the view of the kernel, i.e. from all
1073 * interfaces lists etc. This routine is only called when the driver is
1074 * being unloaded, not when the link goes down.
1077 cxgb_port_detach(device_t dev)
1079 struct port_info *p;
1083 p = device_get_softc(dev);
1086 /* Tell cxgb_ioctl and if_init that the port is going away */
1091 mtx_sleep(&sc->flags, &sc->lock, 0, "cxgbdtch", 0);
1095 if (p->port_cdev != NULL)
1096 destroy_dev(p->port_cdev);
1098 cxgb_uninit_synchronized(p);
1099 ether_ifdetach(p->ifp);
1101 for (i = p->first_qset; i < p->first_qset + p->nqsets; i++) {
1102 struct sge_qset *qs = &sc->sge.qs[i];
1103 struct sge_txq *txq = &qs->txq[TXQ_ETH];
1105 callout_drain(&txq->txq_watchdog);
1106 callout_drain(&txq->txq_timer);
1109 PORT_LOCK_DEINIT(p);
1115 wakeup_one(&sc->flags);
1121 t3_fatal_err(struct adapter *sc)
1125 if (sc->flags & FULL_INIT_DONE) {
1127 t3_write_reg(sc, A_XGM_TX_CTRL, 0);
1128 t3_write_reg(sc, A_XGM_RX_CTRL, 0);
1129 t3_write_reg(sc, XGM_REG(A_XGM_TX_CTRL, 1), 0);
1130 t3_write_reg(sc, XGM_REG(A_XGM_RX_CTRL, 1), 0);
1131 t3_intr_disable(sc);
1133 device_printf(sc->dev,"encountered fatal error, operation suspended\n");
1134 if (!t3_cim_ctl_blk_read(sc, 0xa0, 4, fw_status))
1135 device_printf(sc->dev, "FW_ status: 0x%x, 0x%x, 0x%x, 0x%x\n",
1136 fw_status[0], fw_status[1], fw_status[2], fw_status[3]);
1140 t3_os_find_pci_capability(adapter_t *sc, int cap)
1143 struct pci_devinfo *dinfo;
1149 dinfo = device_get_ivars(dev);
1152 status = pci_read_config(dev, PCIR_STATUS, 2);
1153 if (!(status & PCIM_STATUS_CAPPRESENT))
1156 switch (cfg->hdrtype & PCIM_HDRTYPE) {
1162 ptr = PCIR_CAP_PTR_2;
1168 ptr = pci_read_config(dev, ptr, 1);
1171 if (pci_read_config(dev, ptr + PCICAP_ID, 1) == cap)
1173 ptr = pci_read_config(dev, ptr + PCICAP_NEXTPTR, 1);
1180 t3_os_pci_save_state(struct adapter *sc)
1183 struct pci_devinfo *dinfo;
1186 dinfo = device_get_ivars(dev);
1188 pci_cfg_save(dev, dinfo, 0);
1193 t3_os_pci_restore_state(struct adapter *sc)
1196 struct pci_devinfo *dinfo;
1199 dinfo = device_get_ivars(dev);
1201 pci_cfg_restore(dev, dinfo);
1206 * t3_os_link_changed - handle link status changes
1207 * @sc: the adapter associated with the link change
1208 * @port_id: the port index whose link status has changed
1209 * @link_status: the new status of the link
1210 * @speed: the new speed setting
1211 * @duplex: the new duplex setting
1212 * @fc: the new flow-control setting
1214 * This is the OS-dependent handler for link status changes. The OS
1215 * neutral handler takes care of most of the processing for these events,
1216 * then calls this handler for any OS-specific processing.
1219 t3_os_link_changed(adapter_t *adapter, int port_id, int link_status, int speed,
1220 int duplex, int fc, int mac_was_reset)
1222 struct port_info *pi = &adapter->port[port_id];
1223 struct ifnet *ifp = pi->ifp;
1225 /* no race with detach, so ifp should always be good */
1226 KASSERT(ifp, ("%s: if detached.", __func__));
1228 /* Reapply mac settings if they were lost due to a reset */
1229 if (mac_was_reset) {
1231 cxgb_update_mac_settings(pi);
1236 ifp->if_baudrate = IF_Mbps(speed);
1237 if_link_state_change(ifp, LINK_STATE_UP);
1239 if_link_state_change(ifp, LINK_STATE_DOWN);
1243 * t3_os_phymod_changed - handle PHY module changes
1244 * @phy: the PHY reporting the module change
1245 * @mod_type: new module type
1247 * This is the OS-dependent handler for PHY module changes. It is
1248 * invoked when a PHY module is removed or inserted for any OS-specific
1251 void t3_os_phymod_changed(struct adapter *adap, int port_id)
1253 static const char *mod_str[] = {
1254 NULL, "SR", "LR", "LRM", "TWINAX", "TWINAX-L", "unknown"
1256 struct port_info *pi = &adap->port[port_id];
1257 int mod = pi->phy.modtype;
1259 if (mod != pi->media.ifm_cur->ifm_data)
1260 cxgb_build_medialist(pi);
1262 if (mod == phy_modtype_none)
1263 if_printf(pi->ifp, "PHY module unplugged\n");
1265 KASSERT(mod < ARRAY_SIZE(mod_str),
1266 ("invalid PHY module type %d", mod));
1267 if_printf(pi->ifp, "%s PHY module inserted\n", mod_str[mod]);
1272 t3_os_set_hw_addr(adapter_t *adapter, int port_idx, u8 hw_addr[])
1276 * The ifnet might not be allocated before this gets called,
1277 * as this is called early on in attach by t3_prep_adapter
1278 * save the address off in the port structure
1281 printf("set_hw_addr on idx %d addr %6D\n", port_idx, hw_addr, ":");
1282 bcopy(hw_addr, adapter->port[port_idx].hw_addr, ETHER_ADDR_LEN);
1286 * Programs the XGMAC based on the settings in the ifnet. These settings
1287 * include MTU, MAC address, mcast addresses, etc.
1290 cxgb_update_mac_settings(struct port_info *p)
1292 struct ifnet *ifp = p->ifp;
1293 struct t3_rx_mode rm;
1294 struct cmac *mac = &p->mac;
1297 PORT_LOCK_ASSERT_OWNED(p);
1299 bcopy(IF_LLADDR(ifp), p->hw_addr, ETHER_ADDR_LEN);
1302 if (ifp->if_capenable & IFCAP_VLAN_MTU)
1303 mtu += ETHER_VLAN_ENCAP_LEN;
1305 hwtagging = (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0;
1307 t3_mac_set_mtu(mac, mtu);
1308 t3_set_vlan_accel(p->adapter, 1 << p->tx_chan, hwtagging);
1309 t3_mac_set_address(mac, 0, p->hw_addr);
1310 t3_init_rx_mode(&rm, p);
1311 t3_mac_set_rx_mode(mac, &rm);
1316 await_mgmt_replies(struct adapter *adap, unsigned long init_cnt,
1321 while (adap->sge.qs[0].rspq.offload_pkts < init_cnt + n) {
1330 init_tp_parity(struct adapter *adap)
1334 struct cpl_set_tcb_field *greq;
1335 unsigned long cnt = adap->sge.qs[0].rspq.offload_pkts;
1337 t3_tp_set_offload_mode(adap, 1);
1339 for (i = 0; i < 16; i++) {
1340 struct cpl_smt_write_req *req;
1342 m = m_gethdr(M_WAITOK, MT_DATA);
1343 req = mtod(m, struct cpl_smt_write_req *);
1344 m->m_len = m->m_pkthdr.len = sizeof(*req);
1345 memset(req, 0, sizeof(*req));
1346 req->wr.wrh_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1347 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SMT_WRITE_REQ, i));
1349 t3_mgmt_tx(adap, m);
1352 for (i = 0; i < 2048; i++) {
1353 struct cpl_l2t_write_req *req;
1355 m = m_gethdr(M_WAITOK, MT_DATA);
1356 req = mtod(m, struct cpl_l2t_write_req *);
1357 m->m_len = m->m_pkthdr.len = sizeof(*req);
1358 memset(req, 0, sizeof(*req));
1359 req->wr.wrh_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1360 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_L2T_WRITE_REQ, i));
1361 req->params = htonl(V_L2T_W_IDX(i));
1362 t3_mgmt_tx(adap, m);
1365 for (i = 0; i < 2048; i++) {
1366 struct cpl_rte_write_req *req;
1368 m = m_gethdr(M_WAITOK, MT_DATA);
1369 req = mtod(m, struct cpl_rte_write_req *);
1370 m->m_len = m->m_pkthdr.len = sizeof(*req);
1371 memset(req, 0, sizeof(*req));
1372 req->wr.wrh_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1373 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RTE_WRITE_REQ, i));
1374 req->l2t_idx = htonl(V_L2T_W_IDX(i));
1375 t3_mgmt_tx(adap, m);
1378 m = m_gethdr(M_WAITOK, MT_DATA);
1379 greq = mtod(m, struct cpl_set_tcb_field *);
1380 m->m_len = m->m_pkthdr.len = sizeof(*greq);
1381 memset(greq, 0, sizeof(*greq));
1382 greq->wr.wrh_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
1383 OPCODE_TID(greq) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, 0));
1384 greq->mask = htobe64(1);
1385 t3_mgmt_tx(adap, m);
1387 i = await_mgmt_replies(adap, cnt, 16 + 2048 + 2048 + 1);
1388 t3_tp_set_offload_mode(adap, 0);
1393 * setup_rss - configure Receive Side Steering (per-queue connection demux)
1394 * @adap: the adapter
1396 * Sets up RSS to distribute packets to multiple receive queues. We
1397 * configure the RSS CPU lookup table to distribute to the number of HW
1398 * receive queues, and the response queue lookup table to narrow that
1399 * down to the response queues actually configured for each port.
1400 * We always configure the RSS mapping for two ports since the mapping
1401 * table has plenty of entries.
1404 setup_rss(adapter_t *adap)
1408 uint8_t cpus[SGE_QSETS + 1];
1409 uint16_t rspq_map[RSS_TABLE_SIZE];
1411 for (i = 0; i < SGE_QSETS; ++i)
1413 cpus[SGE_QSETS] = 0xff;
1416 for_each_port(adap, i) {
1417 const struct port_info *pi = adap2pinfo(adap, i);
1419 nq[pi->tx_chan] += pi->nqsets;
1421 for (i = 0; i < RSS_TABLE_SIZE / 2; ++i) {
1422 rspq_map[i] = nq[0] ? i % nq[0] : 0;
1423 rspq_map[i + RSS_TABLE_SIZE / 2] = nq[1] ? i % nq[1] + nq[0] : 0;
1426 /* Calculate the reverse RSS map table */
1427 for (i = 0; i < SGE_QSETS; ++i)
1428 adap->rrss_map[i] = 0xff;
1429 for (i = 0; i < RSS_TABLE_SIZE; ++i)
1430 if (adap->rrss_map[rspq_map[i]] == 0xff)
1431 adap->rrss_map[rspq_map[i]] = i;
1433 t3_config_rss(adap, F_RQFEEDBACKENABLE | F_TNLLKPEN | F_TNLMAPEN |
1434 F_TNLPRTEN | F_TNL2TUPEN | F_TNL4TUPEN | F_OFDMAPEN |
1435 F_RRCPLMAPEN | V_RRCPLCPUSIZE(6) | F_HASHTOEPLITZ,
1440 send_pktsched_cmd(struct adapter *adap, int sched, int qidx, int lo,
1444 struct mngt_pktsched_wr *req;
1446 m = m_gethdr(M_NOWAIT, MT_DATA);
1448 req = mtod(m, struct mngt_pktsched_wr *);
1449 req->wr.wrh_hi = htonl(V_WR_OP(FW_WROPCODE_MNGT));
1450 req->mngt_opcode = FW_MNGTOPCODE_PKTSCHED_SET;
1455 req->binding = port;
1456 m->m_len = m->m_pkthdr.len = sizeof(*req);
1457 t3_mgmt_tx(adap, m);
1462 bind_qsets(adapter_t *sc)
1466 for (i = 0; i < (sc)->params.nports; ++i) {
1467 const struct port_info *pi = adap2pinfo(sc, i);
1469 for (j = 0; j < pi->nqsets; ++j) {
1470 send_pktsched_cmd(sc, 1, pi->first_qset + j, -1,
1478 update_tpeeprom(struct adapter *adap)
1480 const struct firmware *tpeeprom;
1483 unsigned int major, minor;
1487 t3_seeprom_read(adap, TP_SRAM_OFFSET, &version);
1489 major = G_TP_VERSION_MAJOR(version);
1490 minor = G_TP_VERSION_MINOR(version);
1491 if (major == TP_VERSION_MAJOR && minor == TP_VERSION_MINOR)
1494 rev = t3rev2char(adap);
1495 snprintf(name, sizeof(name), TPEEPROM_NAME, rev);
1497 tpeeprom = firmware_get(name);
1498 if (tpeeprom == NULL) {
1499 device_printf(adap->dev,
1500 "could not load TP EEPROM: unable to load %s\n",
1505 len = tpeeprom->datasize - 4;
1507 ret = t3_check_tpsram(adap, tpeeprom->data, tpeeprom->datasize);
1509 goto release_tpeeprom;
1511 if (len != TP_SRAM_LEN) {
1512 device_printf(adap->dev,
1513 "%s length is wrong len=%d expected=%d\n", name,
1518 ret = set_eeprom(&adap->port[0], tpeeprom->data, tpeeprom->datasize,
1522 device_printf(adap->dev,
1523 "Protocol SRAM image updated in EEPROM to %d.%d.%d\n",
1524 TP_VERSION_MAJOR, TP_VERSION_MINOR, TP_VERSION_MICRO);
1526 device_printf(adap->dev,
1527 "Protocol SRAM image update in EEPROM failed\n");
1530 firmware_put(tpeeprom, FIRMWARE_UNLOAD);
1536 update_tpsram(struct adapter *adap)
1538 const struct firmware *tpsram;
1542 rev = t3rev2char(adap);
1543 snprintf(name, sizeof(name), TPSRAM_NAME, rev);
1545 update_tpeeprom(adap);
1547 tpsram = firmware_get(name);
1548 if (tpsram == NULL){
1549 device_printf(adap->dev, "could not load TP SRAM\n");
1552 device_printf(adap->dev, "updating TP SRAM\n");
1554 ret = t3_check_tpsram(adap, tpsram->data, tpsram->datasize);
1556 goto release_tpsram;
1558 ret = t3_set_proto_sram(adap, tpsram->data);
1560 device_printf(adap->dev, "loading protocol SRAM failed\n");
1563 firmware_put(tpsram, FIRMWARE_UNLOAD);
1569 * cxgb_up - enable the adapter
1570 * @adap: adapter being enabled
1572 * Called when the first port is enabled, this function performs the
1573 * actions necessary to make an adapter operational, such as completing
1574 * the initialization of HW modules, and enabling interrupts.
1577 cxgb_up(struct adapter *sc)
1580 unsigned int mxf = t3_mc5_size(&sc->mc5) - MC5_MIN_TIDS;
1582 KASSERT(sc->open_device_map == 0, ("%s: device(s) already open (%x)",
1583 __func__, sc->open_device_map));
1585 if ((sc->flags & FULL_INIT_DONE) == 0) {
1587 ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
1589 if ((sc->flags & FW_UPTODATE) == 0)
1590 if ((err = upgrade_fw(sc)))
1593 if ((sc->flags & TPS_UPTODATE) == 0)
1594 if ((err = update_tpsram(sc)))
1597 if (is_offload(sc) && nfilters != 0) {
1598 sc->params.mc5.nservers = 0;
1601 sc->params.mc5.nfilters = mxf;
1603 sc->params.mc5.nfilters = min(nfilters, mxf);
1606 err = t3_init_hw(sc, 0);
1610 t3_set_reg_field(sc, A_TP_PARA_REG5, 0, F_RXDDPOFFINIT);
1611 t3_write_reg(sc, A_ULPRX_TDDP_PSZ, V_HPZ0(PAGE_SHIFT - 12));
1613 err = setup_sge_qsets(sc);
1620 t3_add_configured_sysctls(sc);
1621 sc->flags |= FULL_INIT_DONE;
1628 if (sc->params.rev >= T3_REV_C && !(sc->flags & TP_PARITY_INIT) &&
1629 is_offload(sc) && init_tp_parity(sc) == 0)
1630 sc->flags |= TP_PARITY_INIT;
1632 if (sc->flags & TP_PARITY_INIT) {
1633 t3_write_reg(sc, A_TP_INT_CAUSE, F_CMCACHEPERR | F_ARPLUTPERR);
1634 t3_write_reg(sc, A_TP_INT_ENABLE, 0x7fbfffff);
1637 if (!(sc->flags & QUEUES_BOUND)) {
1639 setup_hw_filters(sc);
1640 sc->flags |= QUEUES_BOUND;
1643 t3_sge_reset_adapter(sc);
1649 * Called when the last open device is closed. Does NOT undo all of cxgb_up's
1650 * work. Specifically, the resources grabbed under FULL_INIT_DONE are released
1651 * during controller_detach, not here.
1654 cxgb_down(struct adapter *sc)
1657 t3_intr_disable(sc);
1661 * if_init for cxgb ports.
1664 cxgb_init(void *arg)
1666 struct port_info *p = arg;
1667 struct adapter *sc = p->adapter;
1670 cxgb_init_locked(p); /* releases adapter lock */
1671 ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
1675 cxgb_init_locked(struct port_info *p)
1677 struct adapter *sc = p->adapter;
1678 struct ifnet *ifp = p->ifp;
1679 struct cmac *mac = &p->mac;
1680 int i, rc = 0, may_sleep = 0, gave_up_lock = 0;
1682 ADAPTER_LOCK_ASSERT_OWNED(sc);
1684 while (!IS_DOOMED(p) && IS_BUSY(sc)) {
1686 if (mtx_sleep(&sc->flags, &sc->lock, PCATCH, "cxgbinit", 0)) {
1695 KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
1698 * The code that runs during one-time adapter initialization can sleep
1699 * so it's important not to hold any locks across it.
1701 may_sleep = sc->flags & FULL_INIT_DONE ? 0 : 1;
1709 if (sc->open_device_map == 0 && ((rc = cxgb_up(sc)) != 0))
1713 if (isset(&sc->open_device_map, p->port_id) &&
1714 (ifp->if_drv_flags & IFF_DRV_RUNNING)) {
1718 t3_port_intr_enable(sc, p->port_id);
1719 if (!mac->multiport)
1721 cxgb_update_mac_settings(p);
1722 t3_link_start(&p->phy, mac, &p->link_config);
1723 t3_mac_enable(mac, MAC_DIRECTION_RX | MAC_DIRECTION_TX);
1724 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1725 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1728 for (i = p->first_qset; i < p->first_qset + p->nqsets; i++) {
1729 struct sge_qset *qs = &sc->sge.qs[i];
1730 struct sge_txq *txq = &qs->txq[TXQ_ETH];
1732 callout_reset_on(&txq->txq_watchdog, hz, cxgb_tx_watchdog, qs,
1733 txq->txq_watchdog.c_cpu);
1737 setbit(&sc->open_device_map, p->port_id);
1738 callout_reset(&p->link_check_ch,
1739 p->phy.caps & SUPPORTED_LINK_IRQ ? hz * 3 : hz / 4,
1740 link_check_callout, p);
1745 KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
1749 wakeup_one(&sc->flags);
1755 cxgb_uninit_locked(struct port_info *p)
1757 struct adapter *sc = p->adapter;
1760 ADAPTER_LOCK_ASSERT_OWNED(sc);
1762 while (!IS_DOOMED(p) && IS_BUSY(sc)) {
1763 if (mtx_sleep(&sc->flags, &sc->lock, PCATCH, "cxgbunin", 0)) {
1772 KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
1776 rc = cxgb_uninit_synchronized(p);
1779 KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
1781 wakeup_one(&sc->flags);
1788 * Called on "ifconfig down", and from port_detach
1791 cxgb_uninit_synchronized(struct port_info *pi)
1793 struct adapter *sc = pi->adapter;
1794 struct ifnet *ifp = pi->ifp;
1797 * taskqueue_drain may cause a deadlock if the adapter lock is held.
1799 ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
1802 * Clear this port's bit from the open device map, and then drain all
1803 * the tasks that can access/manipulate this port's port_info or ifp.
1804 * We disable this port's interrupts here and so the slow/ext
1805 * interrupt tasks won't be enqueued. The tick task will continue to
1806 * be enqueued every second but the runs after this drain will not see
1807 * this port in the open device map.
1809 * A well behaved task must take open_device_map into account and ignore
1810 * ports that are not open.
1812 clrbit(&sc->open_device_map, pi->port_id);
1813 t3_port_intr_disable(sc, pi->port_id);
1814 taskqueue_drain(sc->tq, &sc->slow_intr_task);
1815 taskqueue_drain(sc->tq, &sc->tick_task);
1817 callout_drain(&pi->link_check_ch);
1818 taskqueue_drain(sc->tq, &pi->link_check_task);
1821 ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
1823 /* disable pause frames */
1824 t3_set_reg_field(sc, A_XGM_TX_CFG + pi->mac.offset, F_TXPAUSEEN, 0);
1826 /* Reset RX FIFO HWM */
1827 t3_set_reg_field(sc, A_XGM_RXFIFO_CFG + pi->mac.offset,
1828 V_RXFIFOPAUSEHWM(M_RXFIFOPAUSEHWM), 0);
1832 /* Wait for TXFIFO empty */
1833 t3_wait_op_done(sc, A_XGM_TXFIFO_CFG + pi->mac.offset,
1834 F_TXFIFO_EMPTY, 1, 20, 5);
1837 t3_mac_disable(&pi->mac, MAC_DIRECTION_RX);
1839 pi->phy.ops->power_down(&pi->phy, 1);
1843 pi->link_config.link_ok = 0;
1844 t3_os_link_changed(sc, pi->port_id, 0, 0, 0, 0, 0);
1846 if (sc->open_device_map == 0)
1847 cxgb_down(pi->adapter);
1853 * Mark lro enabled or disabled in all qsets for this port
1856 cxgb_set_lro(struct port_info *p, int enabled)
1859 struct adapter *adp = p->adapter;
1862 for (i = 0; i < p->nqsets; i++) {
1863 q = &adp->sge.qs[p->first_qset + i];
1864 q->lro.enabled = (enabled != 0);
1870 cxgb_ioctl(struct ifnet *ifp, unsigned long command, caddr_t data)
1872 struct port_info *p = ifp->if_softc;
1873 struct adapter *sc = p->adapter;
1874 struct ifreq *ifr = (struct ifreq *)data;
1875 int flags, error = 0, mtu;
1881 error = IS_DOOMED(p) ? ENXIO : (IS_BUSY(sc) ? EBUSY : 0);
1889 if ((mtu < ETHERMIN) || (mtu > ETHERMTU_JUMBO)) {
1894 cxgb_update_mac_settings(p);
1905 if (ifp->if_flags & IFF_UP) {
1906 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1907 flags = p->if_flags;
1908 if (((ifp->if_flags ^ flags) & IFF_PROMISC) ||
1909 ((ifp->if_flags ^ flags) & IFF_ALLMULTI)) {
1915 cxgb_update_mac_settings(p);
1920 error = cxgb_init_locked(p);
1921 p->if_flags = ifp->if_flags;
1922 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1923 error = cxgb_uninit_locked(p);
1927 ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
1932 error = IS_DOOMED(p) ? ENXIO : (IS_BUSY(sc) ? EBUSY : 0);
1936 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1938 cxgb_update_mac_settings(p);
1946 error = IS_DOOMED(p) ? ENXIO : (IS_BUSY(sc) ? EBUSY : 0);
1950 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1951 if (mask & IFCAP_TXCSUM) {
1952 ifp->if_capenable ^= IFCAP_TXCSUM;
1953 ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
1955 if (IFCAP_TSO4 & ifp->if_capenable &&
1956 !(IFCAP_TXCSUM & ifp->if_capenable)) {
1957 ifp->if_capenable &= ~IFCAP_TSO4;
1959 "tso4 disabled due to -txcsum.\n");
1962 if (mask & IFCAP_TXCSUM_IPV6) {
1963 ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1964 ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
1966 if (IFCAP_TSO6 & ifp->if_capenable &&
1967 !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1968 ifp->if_capenable &= ~IFCAP_TSO6;
1970 "tso6 disabled due to -txcsum6.\n");
1973 if (mask & IFCAP_RXCSUM)
1974 ifp->if_capenable ^= IFCAP_RXCSUM;
1975 if (mask & IFCAP_RXCSUM_IPV6)
1976 ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1979 * Note that we leave CSUM_TSO alone (it is always set). The
1980 * kernel takes both IFCAP_TSOx and CSUM_TSO into account before
1981 * sending a TSO request our way, so it's sufficient to toggle
1984 if (mask & IFCAP_TSO4) {
1985 if (!(IFCAP_TSO4 & ifp->if_capenable) &&
1986 !(IFCAP_TXCSUM & ifp->if_capenable)) {
1987 if_printf(ifp, "enable txcsum first.\n");
1991 ifp->if_capenable ^= IFCAP_TSO4;
1993 if (mask & IFCAP_TSO6) {
1994 if (!(IFCAP_TSO6 & ifp->if_capenable) &&
1995 !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1996 if_printf(ifp, "enable txcsum6 first.\n");
2000 ifp->if_capenable ^= IFCAP_TSO6;
2002 if (mask & IFCAP_LRO) {
2003 ifp->if_capenable ^= IFCAP_LRO;
2005 /* Safe to do this even if cxgb_up not called yet */
2006 cxgb_set_lro(p, ifp->if_capenable & IFCAP_LRO);
2009 if (mask & IFCAP_TOE4) {
2010 int enable = (ifp->if_capenable ^ mask) & IFCAP_TOE4;
2012 error = toe_capability(p, enable);
2014 ifp->if_capenable ^= mask;
2017 if (mask & IFCAP_VLAN_HWTAGGING) {
2018 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
2019 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
2021 cxgb_update_mac_settings(p);
2025 if (mask & IFCAP_VLAN_MTU) {
2026 ifp->if_capenable ^= IFCAP_VLAN_MTU;
2027 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
2029 cxgb_update_mac_settings(p);
2033 if (mask & IFCAP_VLAN_HWTSO)
2034 ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
2035 if (mask & IFCAP_VLAN_HWCSUM)
2036 ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
2038 #ifdef VLAN_CAPABILITIES
2039 VLAN_CAPABILITIES(ifp);
2045 error = ifmedia_ioctl(ifp, ifr, &p->media, command);
2048 error = ether_ioctl(ifp, command, data);
2055 cxgb_media_change(struct ifnet *ifp)
2057 return (EOPNOTSUPP);
2061 * Translates phy->modtype to the correct Ethernet media subtype.
2064 cxgb_ifm_type(int mod)
2067 case phy_modtype_sr:
2068 return (IFM_10G_SR);
2069 case phy_modtype_lr:
2070 return (IFM_10G_LR);
2071 case phy_modtype_lrm:
2072 return (IFM_10G_LRM);
2073 case phy_modtype_twinax:
2074 return (IFM_10G_TWINAX);
2075 case phy_modtype_twinax_long:
2076 return (IFM_10G_TWINAX_LONG);
2077 case phy_modtype_none:
2079 case phy_modtype_unknown:
2080 return (IFM_UNKNOWN);
2083 KASSERT(0, ("%s: modtype %d unknown", __func__, mod));
2084 return (IFM_UNKNOWN);
2088 * Rebuilds the ifmedia list for this port, and sets the current media.
2091 cxgb_build_medialist(struct port_info *p)
2093 struct cphy *phy = &p->phy;
2094 struct ifmedia *media = &p->media;
2095 int mod = phy->modtype;
2096 int m = IFM_ETHER | IFM_FDX;
2100 ifmedia_removeall(media);
2101 if (phy->caps & SUPPORTED_TP && phy->caps & SUPPORTED_Autoneg) {
2104 if (phy->caps & SUPPORTED_10000baseT_Full)
2105 ifmedia_add(media, m | IFM_10G_T, mod, NULL);
2107 if (phy->caps & SUPPORTED_1000baseT_Full)
2108 ifmedia_add(media, m | IFM_1000_T, mod, NULL);
2110 if (phy->caps & SUPPORTED_100baseT_Full)
2111 ifmedia_add(media, m | IFM_100_TX, mod, NULL);
2113 if (phy->caps & SUPPORTED_10baseT_Full)
2114 ifmedia_add(media, m | IFM_10_T, mod, NULL);
2116 ifmedia_add(media, IFM_ETHER | IFM_AUTO, mod, NULL);
2117 ifmedia_set(media, IFM_ETHER | IFM_AUTO);
2119 } else if (phy->caps & SUPPORTED_TP) {
2122 KASSERT(phy->caps & SUPPORTED_10000baseT_Full,
2123 ("%s: unexpected cap 0x%x", __func__, phy->caps));
2125 ifmedia_add(media, m | IFM_10G_CX4, mod, NULL);
2126 ifmedia_set(media, m | IFM_10G_CX4);
2128 } else if (phy->caps & SUPPORTED_FIBRE &&
2129 phy->caps & SUPPORTED_10000baseT_Full) {
2130 /* 10G optical (but includes SFP+ twinax) */
2132 m |= cxgb_ifm_type(mod);
2133 if (IFM_SUBTYPE(m) == IFM_NONE)
2136 ifmedia_add(media, m, mod, NULL);
2137 ifmedia_set(media, m);
2139 } else if (phy->caps & SUPPORTED_FIBRE &&
2140 phy->caps & SUPPORTED_1000baseT_Full) {
2143 /* XXX: Lie and claim to be SX, could actually be any 1G-X */
2144 ifmedia_add(media, m | IFM_1000_SX, mod, NULL);
2145 ifmedia_set(media, m | IFM_1000_SX);
2148 KASSERT(0, ("%s: don't know how to handle 0x%x.", __func__,
2156 cxgb_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
2158 struct port_info *p = ifp->if_softc;
2159 struct ifmedia_entry *cur = p->media.ifm_cur;
2160 int speed = p->link_config.speed;
2162 if (cur->ifm_data != p->phy.modtype) {
2163 cxgb_build_medialist(p);
2164 cur = p->media.ifm_cur;
2167 ifmr->ifm_status = IFM_AVALID;
2168 if (!p->link_config.link_ok)
2171 ifmr->ifm_status |= IFM_ACTIVE;
2174 * active and current will differ iff current media is autoselect. That
2175 * can happen only for copper RJ45.
2177 if (IFM_SUBTYPE(cur->ifm_media) != IFM_AUTO)
2179 KASSERT(p->phy.caps & SUPPORTED_TP && p->phy.caps & SUPPORTED_Autoneg,
2180 ("%s: unexpected PHY caps 0x%x", __func__, p->phy.caps));
2182 ifmr->ifm_active = IFM_ETHER | IFM_FDX;
2183 if (speed == SPEED_10000)
2184 ifmr->ifm_active |= IFM_10G_T;
2185 else if (speed == SPEED_1000)
2186 ifmr->ifm_active |= IFM_1000_T;
2187 else if (speed == SPEED_100)
2188 ifmr->ifm_active |= IFM_100_TX;
2189 else if (speed == SPEED_10)
2190 ifmr->ifm_active |= IFM_10_T;
2192 KASSERT(0, ("%s: link up but speed unknown (%u)", __func__,
2197 cxgb_async_intr(void *data)
2199 adapter_t *sc = data;
2201 t3_write_reg(sc, A_PL_INT_ENABLE0, 0);
2202 (void) t3_read_reg(sc, A_PL_INT_ENABLE0);
2203 taskqueue_enqueue(sc->tq, &sc->slow_intr_task);
2207 link_check_callout(void *arg)
2209 struct port_info *pi = arg;
2210 struct adapter *sc = pi->adapter;
2212 if (!isset(&sc->open_device_map, pi->port_id))
2215 taskqueue_enqueue(sc->tq, &pi->link_check_task);
2219 check_link_status(void *arg, int pending)
2221 struct port_info *pi = arg;
2222 struct adapter *sc = pi->adapter;
2224 if (!isset(&sc->open_device_map, pi->port_id))
2227 t3_link_changed(sc, pi->port_id);
2229 if (pi->link_fault || !(pi->phy.caps & SUPPORTED_LINK_IRQ) ||
2230 pi->link_config.link_ok == 0)
2231 callout_reset(&pi->link_check_ch, hz, link_check_callout, pi);
2235 t3_os_link_intr(struct port_info *pi)
2238 * Schedule a link check in the near future. If the link is flapping
2239 * rapidly we'll keep resetting the callout and delaying the check until
2240 * things stabilize a bit.
2242 callout_reset(&pi->link_check_ch, hz / 4, link_check_callout, pi);
2246 check_t3b2_mac(struct adapter *sc)
2250 if (sc->flags & CXGB_SHUTDOWN)
2253 for_each_port(sc, i) {
2254 struct port_info *p = &sc->port[i];
2257 struct ifnet *ifp = p->ifp;
2260 if (!isset(&sc->open_device_map, p->port_id) || p->link_fault ||
2261 !p->link_config.link_ok)
2264 KASSERT(ifp->if_drv_flags & IFF_DRV_RUNNING,
2265 ("%s: state mismatch (drv_flags %x, device_map %x)",
2266 __func__, ifp->if_drv_flags, sc->open_device_map));
2269 status = t3b2_mac_watchdog_task(&p->mac);
2271 p->mac.stats.num_toggled++;
2272 else if (status == 2) {
2273 struct cmac *mac = &p->mac;
2275 cxgb_update_mac_settings(p);
2276 t3_link_start(&p->phy, mac, &p->link_config);
2277 t3_mac_enable(mac, MAC_DIRECTION_RX | MAC_DIRECTION_TX);
2278 t3_port_intr_enable(sc, p->port_id);
2279 p->mac.stats.num_resets++;
2286 cxgb_tick(void *arg)
2288 adapter_t *sc = (adapter_t *)arg;
2290 if (sc->flags & CXGB_SHUTDOWN)
2293 taskqueue_enqueue(sc->tq, &sc->tick_task);
2294 callout_reset(&sc->cxgb_tick_ch, hz, cxgb_tick, sc);
2298 cxgb_tick_handler(void *arg, int count)
2300 adapter_t *sc = (adapter_t *)arg;
2301 const struct adapter_params *p = &sc->params;
2303 uint32_t cause, reset;
2305 if (sc->flags & CXGB_SHUTDOWN || !(sc->flags & FULL_INIT_DONE))
2308 if (p->rev == T3_REV_B2 && p->nports < 4 && sc->open_device_map)
2311 cause = t3_read_reg(sc, A_SG_INT_CAUSE) & (F_RSPQSTARVE | F_FLEMPTY);
2313 struct sge_qset *qs = &sc->sge.qs[0];
2316 v = t3_read_reg(sc, A_SG_RSPQ_FL_STATUS) & ~0xff00;
2319 for (i = 0; i < SGE_QSETS; i++) {
2321 qs[i].rspq.starved++;
2325 mask <<= SGE_QSETS; /* skip RSPQXDISABLED */
2327 for (i = 0; i < SGE_QSETS * 2; i++) {
2329 qs[i / 2].fl[i % 2].empty++;
2335 t3_write_reg(sc, A_SG_RSPQ_FL_STATUS, v);
2336 t3_write_reg(sc, A_SG_INT_CAUSE, cause);
2339 for (i = 0; i < sc->params.nports; i++) {
2340 struct port_info *pi = &sc->port[i];
2341 struct ifnet *ifp = pi->ifp;
2342 struct cmac *mac = &pi->mac;
2343 struct mac_stats *mstats = &mac->stats;
2346 if (!isset(&sc->open_device_map, pi->port_id))
2350 t3_mac_update_stats(mac);
2353 ifp->if_opackets = mstats->tx_frames;
2354 ifp->if_ipackets = mstats->rx_frames;
2355 ifp->if_obytes = mstats->tx_octets;
2356 ifp->if_ibytes = mstats->rx_octets;
2357 ifp->if_omcasts = mstats->tx_mcast_frames;
2358 ifp->if_imcasts = mstats->rx_mcast_frames;
2359 ifp->if_collisions = mstats->tx_total_collisions;
2360 ifp->if_iqdrops = mstats->rx_cong_drops;
2363 for (j = pi->first_qset; j < pi->first_qset + pi->nqsets; j++)
2364 drops += sc->sge.qs[j].txq[TXQ_ETH].txq_mr->br_drops;
2365 ifp->if_snd.ifq_drops = drops;
2368 mstats->tx_excess_collisions +
2369 mstats->tx_underrun +
2370 mstats->tx_len_errs +
2371 mstats->tx_mac_internal_errs +
2372 mstats->tx_excess_deferral +
2373 mstats->tx_fcs_errs;
2376 mstats->rx_data_errs +
2377 mstats->rx_sequence_errs +
2379 mstats->rx_too_long +
2380 mstats->rx_mac_internal_errs +
2382 mstats->rx_fcs_errs;
2387 /* Count rx fifo overflows, once per second */
2388 cause = t3_read_reg(sc, A_XGM_INT_CAUSE + mac->offset);
2390 if (cause & F_RXFIFO_OVERFLOW) {
2391 mac->stats.rx_fifo_ovfl++;
2392 reset |= F_RXFIFO_OVERFLOW;
2394 t3_write_reg(sc, A_XGM_INT_CAUSE + mac->offset, reset);
2399 touch_bars(device_t dev)
2404 #if !defined(__LP64__) && 0
2407 pci_read_config_dword(pdev, PCI_BASE_ADDRESS_1, &v);
2408 pci_write_config_dword(pdev, PCI_BASE_ADDRESS_1, v);
2409 pci_read_config_dword(pdev, PCI_BASE_ADDRESS_3, &v);
2410 pci_write_config_dword(pdev, PCI_BASE_ADDRESS_3, v);
2411 pci_read_config_dword(pdev, PCI_BASE_ADDRESS_5, &v);
2412 pci_write_config_dword(pdev, PCI_BASE_ADDRESS_5, v);
2417 set_eeprom(struct port_info *pi, const uint8_t *data, int len, int offset)
2421 u32 aligned_offset, aligned_len, *p;
2422 struct adapter *adapter = pi->adapter;
2425 aligned_offset = offset & ~3;
2426 aligned_len = (len + (offset & 3) + 3) & ~3;
2428 if (aligned_offset != offset || aligned_len != len) {
2429 buf = malloc(aligned_len, M_DEVBUF, M_WAITOK|M_ZERO);
2432 err = t3_seeprom_read(adapter, aligned_offset, (u32 *)buf);
2433 if (!err && aligned_len > 4)
2434 err = t3_seeprom_read(adapter,
2435 aligned_offset + aligned_len - 4,
2436 (u32 *)&buf[aligned_len - 4]);
2439 memcpy(buf + (offset & 3), data, len);
2441 buf = (uint8_t *)(uintptr_t)data;
2443 err = t3_seeprom_wp(adapter, 0);
2447 for (p = (u32 *)buf; !err && aligned_len; aligned_len -= 4, p++) {
2448 err = t3_seeprom_write(adapter, aligned_offset, *p);
2449 aligned_offset += 4;
2453 err = t3_seeprom_wp(adapter, 1);
2456 free(buf, M_DEVBUF);
2462 in_range(int val, int lo, int hi)
2464 return val < 0 || (val <= hi && val >= lo);
2468 cxgb_extension_open(struct cdev *dev, int flags, int fmp, struct thread *td)
2474 cxgb_extension_close(struct cdev *dev, int flags, int fmt, struct thread *td)
2480 cxgb_extension_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data,
2481 int fflag, struct thread *td)
2484 struct port_info *pi = dev->si_drv1;
2485 adapter_t *sc = pi->adapter;
2487 #ifdef PRIV_SUPPORTED
2488 if (priv_check(td, PRIV_DRIVER)) {
2490 printf("user does not have access to privileged ioctls\n");
2496 printf("user does not have access to privileged ioctls\n");
2502 case CHELSIO_GET_MIIREG: {
2504 struct cphy *phy = &pi->phy;
2505 struct ch_mii_data *mid = (struct ch_mii_data *)data;
2507 if (!phy->mdio_read)
2508 return (EOPNOTSUPP);
2510 mmd = mid->phy_id >> 8;
2513 else if (mmd > MDIO_DEV_VEND2)
2516 error = phy->mdio_read(sc, mid->phy_id & 0x1f, mmd,
2517 mid->reg_num, &val);
2519 error = phy->mdio_read(sc, mid->phy_id & 0x1f, 0,
2520 mid->reg_num & 0x1f, &val);
2525 case CHELSIO_SET_MIIREG: {
2526 struct cphy *phy = &pi->phy;
2527 struct ch_mii_data *mid = (struct ch_mii_data *)data;
2529 if (!phy->mdio_write)
2530 return (EOPNOTSUPP);
2532 mmd = mid->phy_id >> 8;
2535 else if (mmd > MDIO_DEV_VEND2)
2538 error = phy->mdio_write(sc, mid->phy_id & 0x1f,
2539 mmd, mid->reg_num, mid->val_in);
2541 error = phy->mdio_write(sc, mid->phy_id & 0x1f, 0,
2542 mid->reg_num & 0x1f,
2546 case CHELSIO_SETREG: {
2547 struct ch_reg *edata = (struct ch_reg *)data;
2548 if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
2550 t3_write_reg(sc, edata->addr, edata->val);
2553 case CHELSIO_GETREG: {
2554 struct ch_reg *edata = (struct ch_reg *)data;
2555 if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
2557 edata->val = t3_read_reg(sc, edata->addr);
2560 case CHELSIO_GET_SGE_CONTEXT: {
2561 struct ch_cntxt *ecntxt = (struct ch_cntxt *)data;
2562 mtx_lock_spin(&sc->sge.reg_lock);
2563 switch (ecntxt->cntxt_type) {
2564 case CNTXT_TYPE_EGRESS:
2565 error = -t3_sge_read_ecntxt(sc, ecntxt->cntxt_id,
2569 error = -t3_sge_read_fl(sc, ecntxt->cntxt_id,
2572 case CNTXT_TYPE_RSP:
2573 error = -t3_sge_read_rspq(sc, ecntxt->cntxt_id,
2577 error = -t3_sge_read_cq(sc, ecntxt->cntxt_id,
2584 mtx_unlock_spin(&sc->sge.reg_lock);
2587 case CHELSIO_GET_SGE_DESC: {
2588 struct ch_desc *edesc = (struct ch_desc *)data;
2590 if (edesc->queue_num >= SGE_QSETS * 6)
2592 ret = t3_get_desc(&sc->sge.qs[edesc->queue_num / 6],
2593 edesc->queue_num % 6, edesc->idx, edesc->data);
2599 case CHELSIO_GET_QSET_PARAMS: {
2600 struct qset_params *q;
2601 struct ch_qset_params *t = (struct ch_qset_params *)data;
2602 int q1 = pi->first_qset;
2603 int nqsets = pi->nqsets;
2606 if (t->qset_idx >= nqsets)
2609 i = q1 + t->qset_idx;
2610 q = &sc->params.sge.qset[i];
2611 t->rspq_size = q->rspq_size;
2612 t->txq_size[0] = q->txq_size[0];
2613 t->txq_size[1] = q->txq_size[1];
2614 t->txq_size[2] = q->txq_size[2];
2615 t->fl_size[0] = q->fl_size;
2616 t->fl_size[1] = q->jumbo_size;
2617 t->polling = q->polling;
2619 t->intr_lat = q->coalesce_usecs;
2620 t->cong_thres = q->cong_thres;
2623 if ((sc->flags & FULL_INIT_DONE) == 0)
2625 else if (sc->flags & USING_MSIX)
2626 t->vector = rman_get_start(sc->msix_irq_res[i]);
2628 t->vector = rman_get_start(sc->irq_res);
2632 case CHELSIO_GET_QSET_NUM: {
2633 struct ch_reg *edata = (struct ch_reg *)data;
2634 edata->val = pi->nqsets;
2637 case CHELSIO_LOAD_FW: {
2640 struct ch_mem_range *t = (struct ch_mem_range *)data;
2643 * You're allowed to load a firmware only before FULL_INIT_DONE
2645 * FW_UPTODATE is also set so the rest of the initialization
2646 * will not overwrite what was loaded here. This gives you the
2647 * flexibility to load any firmware (and maybe shoot yourself in
2652 if (sc->open_device_map || sc->flags & FULL_INIT_DONE) {
2657 fw_data = malloc(t->len, M_DEVBUF, M_NOWAIT);
2661 error = copyin(t->buf, fw_data, t->len);
2664 error = -t3_load_fw(sc, fw_data, t->len);
2666 if (t3_get_fw_version(sc, &vers) == 0) {
2667 snprintf(&sc->fw_version[0], sizeof(sc->fw_version),
2668 "%d.%d.%d", G_FW_VERSION_MAJOR(vers),
2669 G_FW_VERSION_MINOR(vers), G_FW_VERSION_MICRO(vers));
2673 sc->flags |= FW_UPTODATE;
2675 free(fw_data, M_DEVBUF);
2679 case CHELSIO_LOAD_BOOT: {
2681 struct ch_mem_range *t = (struct ch_mem_range *)data;
2683 boot_data = malloc(t->len, M_DEVBUF, M_NOWAIT);
2687 error = copyin(t->buf, boot_data, t->len);
2689 error = -t3_load_boot(sc, boot_data, t->len);
2691 free(boot_data, M_DEVBUF);
2694 case CHELSIO_GET_PM: {
2695 struct ch_pm *m = (struct ch_pm *)data;
2696 struct tp_params *p = &sc->params.tp;
2698 if (!is_offload(sc))
2699 return (EOPNOTSUPP);
2701 m->tx_pg_sz = p->tx_pg_size;
2702 m->tx_num_pg = p->tx_num_pgs;
2703 m->rx_pg_sz = p->rx_pg_size;
2704 m->rx_num_pg = p->rx_num_pgs;
2705 m->pm_total = p->pmtx_size + p->chan_rx_size * p->nchan;
2709 case CHELSIO_SET_PM: {
2710 struct ch_pm *m = (struct ch_pm *)data;
2711 struct tp_params *p = &sc->params.tp;
2713 if (!is_offload(sc))
2714 return (EOPNOTSUPP);
2715 if (sc->flags & FULL_INIT_DONE)
2718 if (!m->rx_pg_sz || (m->rx_pg_sz & (m->rx_pg_sz - 1)) ||
2719 !m->tx_pg_sz || (m->tx_pg_sz & (m->tx_pg_sz - 1)))
2720 return (EINVAL); /* not power of 2 */
2721 if (!(m->rx_pg_sz & 0x14000))
2722 return (EINVAL); /* not 16KB or 64KB */
2723 if (!(m->tx_pg_sz & 0x1554000))
2725 if (m->tx_num_pg == -1)
2726 m->tx_num_pg = p->tx_num_pgs;
2727 if (m->rx_num_pg == -1)
2728 m->rx_num_pg = p->rx_num_pgs;
2729 if (m->tx_num_pg % 24 || m->rx_num_pg % 24)
2731 if (m->rx_num_pg * m->rx_pg_sz > p->chan_rx_size ||
2732 m->tx_num_pg * m->tx_pg_sz > p->chan_tx_size)
2735 p->rx_pg_size = m->rx_pg_sz;
2736 p->tx_pg_size = m->tx_pg_sz;
2737 p->rx_num_pgs = m->rx_num_pg;
2738 p->tx_num_pgs = m->tx_num_pg;
2741 case CHELSIO_SETMTUTAB: {
2742 struct ch_mtus *m = (struct ch_mtus *)data;
2745 if (!is_offload(sc))
2746 return (EOPNOTSUPP);
2747 if (offload_running(sc))
2749 if (m->nmtus != NMTUS)
2751 if (m->mtus[0] < 81) /* accommodate SACK */
2755 * MTUs must be in ascending order
2757 for (i = 1; i < NMTUS; ++i)
2758 if (m->mtus[i] < m->mtus[i - 1])
2761 memcpy(sc->params.mtus, m->mtus, sizeof(sc->params.mtus));
2764 case CHELSIO_GETMTUTAB: {
2765 struct ch_mtus *m = (struct ch_mtus *)data;
2767 if (!is_offload(sc))
2768 return (EOPNOTSUPP);
2770 memcpy(m->mtus, sc->params.mtus, sizeof(m->mtus));
2774 case CHELSIO_GET_MEM: {
2775 struct ch_mem_range *t = (struct ch_mem_range *)data;
2781 * Use these to avoid modifying len/addr in the return
2784 uint32_t len = t->len, addr = t->addr;
2786 if (!is_offload(sc))
2787 return (EOPNOTSUPP);
2788 if (!(sc->flags & FULL_INIT_DONE))
2789 return (EIO); /* need the memory controllers */
2790 if ((addr & 0x7) || (len & 0x7))
2792 if (t->mem_id == MEM_CM)
2794 else if (t->mem_id == MEM_PMRX)
2796 else if (t->mem_id == MEM_PMTX)
2803 * bits 0..9: chip version
2804 * bits 10..15: chip revision
2806 t->version = 3 | (sc->params.rev << 10);
2809 * Read 256 bytes at a time as len can be large and we don't
2810 * want to use huge intermediate buffers.
2812 useraddr = (uint8_t *)t->buf;
2814 unsigned int chunk = min(len, sizeof(buf));
2816 error = t3_mc7_bd_read(mem, addr / 8, chunk / 8, buf);
2819 if (copyout(buf, useraddr, chunk))
2827 case CHELSIO_READ_TCAM_WORD: {
2828 struct ch_tcam_word *t = (struct ch_tcam_word *)data;
2830 if (!is_offload(sc))
2831 return (EOPNOTSUPP);
2832 if (!(sc->flags & FULL_INIT_DONE))
2833 return (EIO); /* need MC5 */
2834 return -t3_read_mc5_range(&sc->mc5, t->addr, 1, t->buf);
2837 case CHELSIO_SET_TRACE_FILTER: {
2838 struct ch_trace *t = (struct ch_trace *)data;
2839 const struct trace_params *tp;
2841 tp = (const struct trace_params *)&t->sip;
2843 t3_config_trace_filter(sc, tp, 0, t->invert_match,
2846 t3_config_trace_filter(sc, tp, 1, t->invert_match,
2850 case CHELSIO_SET_PKTSCHED: {
2851 struct ch_pktsched_params *p = (struct ch_pktsched_params *)data;
2852 if (sc->open_device_map == 0)
2854 send_pktsched_cmd(sc, p->sched, p->idx, p->min, p->max,
2858 case CHELSIO_IFCONF_GETREGS: {
2859 struct ch_ifconf_regs *regs = (struct ch_ifconf_regs *)data;
2860 int reglen = cxgb_get_regs_len();
2861 uint8_t *buf = malloc(reglen, M_DEVBUF, M_NOWAIT);
2865 if (regs->len > reglen)
2867 else if (regs->len < reglen)
2871 cxgb_get_regs(sc, regs, buf);
2872 error = copyout(buf, regs->data, reglen);
2874 free(buf, M_DEVBUF);
2878 case CHELSIO_SET_HW_SCHED: {
2879 struct ch_hw_sched *t = (struct ch_hw_sched *)data;
2880 unsigned int ticks_per_usec = core_ticks_per_usec(sc);
2882 if ((sc->flags & FULL_INIT_DONE) == 0)
2883 return (EAGAIN); /* need TP to be initialized */
2884 if (t->sched >= NTX_SCHED || !in_range(t->mode, 0, 1) ||
2885 !in_range(t->channel, 0, 1) ||
2886 !in_range(t->kbps, 0, 10000000) ||
2887 !in_range(t->class_ipg, 0, 10000 * 65535 / ticks_per_usec) ||
2888 !in_range(t->flow_ipg, 0,
2889 dack_ticks_to_usec(sc, 0x7ff)))
2893 error = t3_config_sched(sc, t->kbps, t->sched);
2897 if (t->class_ipg >= 0)
2898 t3_set_sched_ipg(sc, t->sched, t->class_ipg);
2899 if (t->flow_ipg >= 0) {
2900 t->flow_ipg *= 1000; /* us -> ns */
2901 t3_set_pace_tbl(sc, &t->flow_ipg, t->sched, 1);
2904 int bit = 1 << (S_TX_MOD_TIMER_MODE + t->sched);
2906 t3_set_reg_field(sc, A_TP_TX_MOD_QUEUE_REQ_MAP,
2907 bit, t->mode ? bit : 0);
2909 if (t->channel >= 0)
2910 t3_set_reg_field(sc, A_TP_TX_MOD_QUEUE_REQ_MAP,
2911 1 << t->sched, t->channel << t->sched);
2914 case CHELSIO_GET_EEPROM: {
2916 struct ch_eeprom *e = (struct ch_eeprom *)data;
2917 uint8_t *buf = malloc(EEPROMSIZE, M_DEVBUF, M_NOWAIT);
2922 e->magic = EEPROM_MAGIC;
2923 for (i = e->offset & ~3; !error && i < e->offset + e->len; i += 4)
2924 error = -t3_seeprom_read(sc, i, (uint32_t *)&buf[i]);
2927 error = copyout(buf + e->offset, e->data, e->len);
2929 free(buf, M_DEVBUF);
2932 case CHELSIO_CLEAR_STATS: {
2933 if (!(sc->flags & FULL_INIT_DONE))
2937 t3_mac_update_stats(&pi->mac);
2938 memset(&pi->mac.stats, 0, sizeof(pi->mac.stats));
2942 case CHELSIO_GET_UP_LA: {
2943 struct ch_up_la *la = (struct ch_up_la *)data;
2944 uint8_t *buf = malloc(LA_BUFSIZE, M_DEVBUF, M_NOWAIT);
2948 if (la->bufsize < LA_BUFSIZE)
2952 error = -t3_get_up_la(sc, &la->stopped, &la->idx,
2955 error = copyout(buf, la->data, la->bufsize);
2957 free(buf, M_DEVBUF);
2960 case CHELSIO_GET_UP_IOQS: {
2961 struct ch_up_ioqs *ioqs = (struct ch_up_ioqs *)data;
2962 uint8_t *buf = malloc(IOQS_BUFSIZE, M_DEVBUF, M_NOWAIT);
2968 if (ioqs->bufsize < IOQS_BUFSIZE)
2972 error = -t3_get_up_ioqs(sc, &ioqs->bufsize, buf);
2975 v = (uint32_t *)buf;
2977 ioqs->ioq_rx_enable = *v++;
2978 ioqs->ioq_tx_enable = *v++;
2979 ioqs->ioq_rx_status = *v++;
2980 ioqs->ioq_tx_status = *v++;
2982 error = copyout(v, ioqs->data, ioqs->bufsize);
2985 free(buf, M_DEVBUF);
2988 case CHELSIO_SET_FILTER: {
2989 struct ch_filter *f = (struct ch_filter *)data;
2990 struct filter_info *p;
2991 unsigned int nfilters = sc->params.mc5.nfilters;
2993 if (!is_offload(sc))
2994 return (EOPNOTSUPP); /* No TCAM */
2995 if (!(sc->flags & FULL_INIT_DONE))
2996 return (EAGAIN); /* mc5 not setup yet */
2998 return (EBUSY); /* TOE will use TCAM */
3001 if (f->filter_id >= nfilters ||
3002 (f->val.dip && f->mask.dip != 0xffffffff) ||
3003 (f->val.sport && f->mask.sport != 0xffff) ||
3004 (f->val.dport && f->mask.dport != 0xffff) ||
3005 (f->val.vlan && f->mask.vlan != 0xfff) ||
3006 (f->val.vlan_prio &&
3007 f->mask.vlan_prio != FILTER_NO_VLAN_PRI) ||
3008 (f->mac_addr_idx != 0xffff && f->mac_addr_idx > 15) ||
3009 f->qset >= SGE_QSETS ||
3010 sc->rrss_map[f->qset] >= RSS_TABLE_SIZE)
3013 /* Was allocated with M_WAITOK */
3014 KASSERT(sc->filters, ("filter table NULL\n"));
3016 p = &sc->filters[f->filter_id];
3020 bzero(p, sizeof(*p));
3021 p->sip = f->val.sip;
3022 p->sip_mask = f->mask.sip;
3023 p->dip = f->val.dip;
3024 p->sport = f->val.sport;
3025 p->dport = f->val.dport;
3026 p->vlan = f->mask.vlan ? f->val.vlan : 0xfff;
3027 p->vlan_prio = f->mask.vlan_prio ? (f->val.vlan_prio & 6) :
3029 p->mac_hit = f->mac_hit;
3030 p->mac_vld = f->mac_addr_idx != 0xffff;
3031 p->mac_idx = f->mac_addr_idx;
3032 p->pkt_type = f->proto;
3033 p->report_filter_id = f->want_filter_id;
3038 error = set_filter(sc, f->filter_id, p);
3043 case CHELSIO_DEL_FILTER: {
3044 struct ch_filter *f = (struct ch_filter *)data;
3045 struct filter_info *p;
3046 unsigned int nfilters = sc->params.mc5.nfilters;
3048 if (!is_offload(sc))
3049 return (EOPNOTSUPP);
3050 if (!(sc->flags & FULL_INIT_DONE))
3052 if (nfilters == 0 || sc->filters == NULL)
3054 if (f->filter_id >= nfilters)
3057 p = &sc->filters[f->filter_id];
3061 return (EFAULT); /* Read "Bad address" as "Bad index" */
3063 bzero(p, sizeof(*p));
3064 p->sip = p->sip_mask = 0xffffffff;
3066 p->vlan_prio = FILTER_NO_VLAN_PRI;
3068 error = set_filter(sc, f->filter_id, p);
3071 case CHELSIO_GET_FILTER: {
3072 struct ch_filter *f = (struct ch_filter *)data;
3073 struct filter_info *p;
3074 unsigned int i, nfilters = sc->params.mc5.nfilters;
3076 if (!is_offload(sc))
3077 return (EOPNOTSUPP);
3078 if (!(sc->flags & FULL_INIT_DONE))
3080 if (nfilters == 0 || sc->filters == NULL)
3083 i = f->filter_id == 0xffffffff ? 0 : f->filter_id + 1;
3084 for (; i < nfilters; i++) {
3085 p = &sc->filters[i];
3089 bzero(f, sizeof(*f));
3092 f->val.sip = p->sip;
3093 f->mask.sip = p->sip_mask;
3094 f->val.dip = p->dip;
3095 f->mask.dip = p->dip ? 0xffffffff : 0;
3096 f->val.sport = p->sport;
3097 f->mask.sport = p->sport ? 0xffff : 0;
3098 f->val.dport = p->dport;
3099 f->mask.dport = p->dport ? 0xffff : 0;
3100 f->val.vlan = p->vlan == 0xfff ? 0 : p->vlan;
3101 f->mask.vlan = p->vlan == 0xfff ? 0 : 0xfff;
3102 f->val.vlan_prio = p->vlan_prio == FILTER_NO_VLAN_PRI ?
3104 f->mask.vlan_prio = p->vlan_prio == FILTER_NO_VLAN_PRI ?
3105 0 : FILTER_NO_VLAN_PRI;
3106 f->mac_hit = p->mac_hit;
3107 f->mac_addr_idx = p->mac_vld ? p->mac_idx : 0xffff;
3108 f->proto = p->pkt_type;
3109 f->want_filter_id = p->report_filter_id;
3118 f->filter_id = 0xffffffff;
3122 return (EOPNOTSUPP);
3129 static __inline void
3130 reg_block_dump(struct adapter *ap, uint8_t *buf, unsigned int start,
3133 uint32_t *p = (uint32_t *)(buf + start);
3135 for ( ; start <= end; start += sizeof(uint32_t))
3136 *p++ = t3_read_reg(ap, start);
3139 #define T3_REGMAP_SIZE (3 * 1024)
3141 cxgb_get_regs_len(void)
3143 return T3_REGMAP_SIZE;
3147 cxgb_get_regs(adapter_t *sc, struct ch_ifconf_regs *regs, uint8_t *buf)
3152 * bits 0..9: chip version
3153 * bits 10..15: chip revision
3154 * bit 31: set for PCIe cards
3156 regs->version = 3 | (sc->params.rev << 10) | (is_pcie(sc) << 31);
3159 * We skip the MAC statistics registers because they are clear-on-read.
3160 * Also reading multi-register stats would need to synchronize with the
3161 * periodic mac stats accumulation. Hard to justify the complexity.
3163 memset(buf, 0, cxgb_get_regs_len());
3164 reg_block_dump(sc, buf, 0, A_SG_RSPQ_CREDIT_RETURN);
3165 reg_block_dump(sc, buf, A_SG_HI_DRB_HI_THRSH, A_ULPRX_PBL_ULIMIT);
3166 reg_block_dump(sc, buf, A_ULPTX_CONFIG, A_MPS_INT_CAUSE);
3167 reg_block_dump(sc, buf, A_CPL_SWITCH_CNTRL, A_CPL_MAP_TBL_DATA);
3168 reg_block_dump(sc, buf, A_SMB_GLOBAL_TIME_CFG, A_XGM_SERDES_STAT3);
3169 reg_block_dump(sc, buf, A_XGM_SERDES_STATUS0,
3170 XGM_REG(A_XGM_SERDES_STAT3, 1));
3171 reg_block_dump(sc, buf, XGM_REG(A_XGM_SERDES_STATUS0, 1),
3172 XGM_REG(A_XGM_RX_SPI4_SOP_EOP_CNT, 1));
3176 alloc_filters(struct adapter *sc)
3178 struct filter_info *p;
3179 unsigned int nfilters = sc->params.mc5.nfilters;
3184 p = malloc(sizeof(*p) * nfilters, M_DEVBUF, M_WAITOK | M_ZERO);
3187 p = &sc->filters[nfilters - 1];
3189 p->vlan_prio = FILTER_NO_VLAN_PRI;
3190 p->pass = p->rss = p->valid = p->locked = 1;
3196 setup_hw_filters(struct adapter *sc)
3199 unsigned int nfilters = sc->params.mc5.nfilters;
3204 t3_enable_filters(sc);
3206 for (i = rc = 0; i < nfilters && !rc; i++) {
3207 if (sc->filters[i].locked)
3208 rc = set_filter(sc, i, &sc->filters[i]);
3215 set_filter(struct adapter *sc, int id, const struct filter_info *f)
3219 struct ulp_txpkt *txpkt;
3220 struct work_request_hdr *wr;
3221 struct cpl_pass_open_req *oreq;
3222 struct cpl_set_tcb_field *sreq;
3224 len = sizeof(*wr) + sizeof(*oreq) + 2 * sizeof(*sreq);
3225 KASSERT(len <= MHLEN, ("filter request too big for an mbuf"));
3227 id += t3_mc5_size(&sc->mc5) - sc->params.mc5.nroutes -
3228 sc->params.mc5.nfilters;
3230 m = m_gethdr(M_WAITOK, MT_DATA);
3231 m->m_len = m->m_pkthdr.len = len;
3232 bzero(mtod(m, char *), len);
3234 wr = mtod(m, struct work_request_hdr *);
3235 wr->wrh_hi = htonl(V_WR_OP(FW_WROPCODE_BYPASS) | F_WR_ATOMIC);
3237 oreq = (struct cpl_pass_open_req *)(wr + 1);
3238 txpkt = (struct ulp_txpkt *)oreq;
3239 txpkt->cmd_dest = htonl(V_ULPTX_CMD(ULP_TXPKT));
3240 txpkt->len = htonl(V_ULPTX_NFLITS(sizeof(*oreq) / 8));
3241 OPCODE_TID(oreq) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, id));
3242 oreq->local_port = htons(f->dport);
3243 oreq->peer_port = htons(f->sport);
3244 oreq->local_ip = htonl(f->dip);
3245 oreq->peer_ip = htonl(f->sip);
3246 oreq->peer_netmask = htonl(f->sip_mask);
3248 oreq->opt0l = htonl(F_NO_OFFLOAD);
3249 oreq->opt1 = htonl(V_MAC_MATCH_VALID(f->mac_vld) |
3250 V_CONN_POLICY(CPL_CONN_POLICY_FILTER) |
3251 V_VLAN_PRI(f->vlan_prio >> 1) |
3252 V_VLAN_PRI_VALID(f->vlan_prio != FILTER_NO_VLAN_PRI) |
3253 V_PKT_TYPE(f->pkt_type) | V_OPT1_VLAN(f->vlan) |
3254 V_MAC_MATCH(f->mac_idx | (f->mac_hit << 4)));
3256 sreq = (struct cpl_set_tcb_field *)(oreq + 1);
3257 set_tcb_field_ulp(sreq, id, 1, 0x1800808000ULL,
3258 (f->report_filter_id << 15) | (1 << 23) |
3259 ((u64)f->pass << 35) | ((u64)!f->rss << 36));
3260 set_tcb_field_ulp(sreq + 1, id, 0, 0xffffffff, (2 << 19) | 1);
3263 if (f->pass && !f->rss) {
3264 len = sizeof(*sreq);
3265 m = m_gethdr(M_WAITOK, MT_DATA);
3266 m->m_len = m->m_pkthdr.len = len;
3267 bzero(mtod(m, char *), len);
3268 sreq = mtod(m, struct cpl_set_tcb_field *);
3269 sreq->wr.wrh_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
3270 mk_set_tcb_field(sreq, id, 25, 0x3f80000,
3271 (u64)sc->rrss_map[f->qset] << 19);
3278 mk_set_tcb_field(struct cpl_set_tcb_field *req, unsigned int tid,
3279 unsigned int word, u64 mask, u64 val)
3281 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, tid));
3282 req->reply = V_NO_REPLY(1);
3284 req->word = htons(word);
3285 req->mask = htobe64(mask);
3286 req->val = htobe64(val);
3290 set_tcb_field_ulp(struct cpl_set_tcb_field *req, unsigned int tid,
3291 unsigned int word, u64 mask, u64 val)
3293 struct ulp_txpkt *txpkt = (struct ulp_txpkt *)req;
3295 txpkt->cmd_dest = htonl(V_ULPTX_CMD(ULP_TXPKT));
3296 txpkt->len = htonl(V_ULPTX_NFLITS(sizeof(*req) / 8));
3297 mk_set_tcb_field(req, tid, word, mask, val);
3301 t3_iterate(void (*func)(struct adapter *, void *), void *arg)
3305 mtx_lock(&t3_list_lock);
3306 SLIST_FOREACH(sc, &t3_list, link) {
3308 * func should not make any assumptions about what state sc is
3309 * in - the only guarantee is that sc->sc_lock is a valid lock.
3313 mtx_unlock(&t3_list_lock);
3318 toe_capability(struct port_info *pi, int enable)
3321 struct adapter *sc = pi->adapter;
3323 ADAPTER_LOCK_ASSERT_OWNED(sc);
3325 if (!is_offload(sc))
3329 if (!(sc->flags & FULL_INIT_DONE)) {
3331 "You must enable a cxgb interface first\n");
3335 if (isset(&sc->offload_map, pi->port_id))
3338 if (!(sc->flags & TOM_INIT_DONE)) {
3339 rc = t3_activate_uld(sc, ULD_TOM);
3342 "You must kldload t3_tom.ko before trying "
3343 "to enable TOE on a cxgb interface.\n");
3347 KASSERT(sc->tom_softc != NULL,
3348 ("%s: TOM activated but softc NULL", __func__));
3349 KASSERT(sc->flags & TOM_INIT_DONE,
3350 ("%s: TOM activated but flag not set", __func__));
3353 setbit(&sc->offload_map, pi->port_id);
3356 * XXX: Temporary code to allow iWARP to be enabled when TOE is
3357 * enabled on any port. Need to figure out how to enable,
3358 * disable, load, and unload iWARP cleanly.
3360 if (!isset(&sc->offload_map, MAX_NPORTS) &&
3361 t3_activate_uld(sc, ULD_IWARP) == 0)
3362 setbit(&sc->offload_map, MAX_NPORTS);
3364 if (!isset(&sc->offload_map, pi->port_id))
3367 KASSERT(sc->flags & TOM_INIT_DONE,
3368 ("%s: TOM never initialized?", __func__));
3369 clrbit(&sc->offload_map, pi->port_id);
3376 * Add an upper layer driver to the global list.
3379 t3_register_uld(struct uld_info *ui)
3384 mtx_lock(&t3_uld_list_lock);
3385 SLIST_FOREACH(u, &t3_uld_list, link) {
3386 if (u->uld_id == ui->uld_id) {
3392 SLIST_INSERT_HEAD(&t3_uld_list, ui, link);
3395 mtx_unlock(&t3_uld_list_lock);
3400 t3_unregister_uld(struct uld_info *ui)
3405 mtx_lock(&t3_uld_list_lock);
3407 SLIST_FOREACH(u, &t3_uld_list, link) {
3409 if (ui->refcount > 0) {
3414 SLIST_REMOVE(&t3_uld_list, ui, uld_info, link);
3420 mtx_unlock(&t3_uld_list_lock);
3425 t3_activate_uld(struct adapter *sc, int id)
3428 struct uld_info *ui;
3430 mtx_lock(&t3_uld_list_lock);
3432 SLIST_FOREACH(ui, &t3_uld_list, link) {
3433 if (ui->uld_id == id) {
3434 rc = ui->activate(sc);
3441 mtx_unlock(&t3_uld_list_lock);
3447 t3_deactivate_uld(struct adapter *sc, int id)
3450 struct uld_info *ui;
3452 mtx_lock(&t3_uld_list_lock);
3454 SLIST_FOREACH(ui, &t3_uld_list, link) {
3455 if (ui->uld_id == id) {
3456 rc = ui->deactivate(sc);
3463 mtx_unlock(&t3_uld_list_lock);
3469 cpl_not_handled(struct sge_qset *qs __unused, struct rsp_desc *r __unused,
3477 t3_register_cpl_handler(struct adapter *sc, int opcode, cpl_handler_t h)
3479 uintptr_t *loc, new;
3481 if (opcode >= NUM_CPL_HANDLERS)
3484 new = h ? (uintptr_t)h : (uintptr_t)cpl_not_handled;
3485 loc = (uintptr_t *) &sc->cpl_handler[opcode];
3486 atomic_store_rel_ptr(loc, new);
3493 cxgbc_mod_event(module_t mod, int cmd, void *arg)
3499 mtx_init(&t3_list_lock, "T3 adapters", 0, MTX_DEF);
3500 SLIST_INIT(&t3_list);
3502 mtx_init(&t3_uld_list_lock, "T3 ULDs", 0, MTX_DEF);
3503 SLIST_INIT(&t3_uld_list);
3509 mtx_lock(&t3_uld_list_lock);
3510 if (!SLIST_EMPTY(&t3_uld_list)) {
3512 mtx_unlock(&t3_uld_list_lock);
3515 mtx_unlock(&t3_uld_list_lock);
3516 mtx_destroy(&t3_uld_list_lock);
3518 mtx_lock(&t3_list_lock);
3519 if (!SLIST_EMPTY(&t3_list)) {
3521 mtx_unlock(&t3_list_lock);
3524 mtx_unlock(&t3_list_lock);
3525 mtx_destroy(&t3_list_lock);