4 * Copyright(c) 2014-2017 Chelsio Communications.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * * Neither the name of Chelsio Communications nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 #include <sys/queue.h>
42 #include <netinet/in.h>
44 #include <rte_byteorder.h>
45 #include <rte_common.h>
46 #include <rte_cycles.h>
47 #include <rte_interrupts.h>
49 #include <rte_debug.h>
51 #include <rte_atomic.h>
52 #include <rte_branch_prediction.h>
53 #include <rte_memory.h>
54 #include <rte_tailq.h>
56 #include <rte_alarm.h>
57 #include <rte_ether.h>
58 #include <rte_ethdev_driver.h>
59 #include <rte_ethdev_pci.h>
60 #include <rte_malloc.h>
61 #include <rte_random.h>
70 * Response queue handler for the FW event queue.
72 static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
73 __rte_unused const struct pkt_gl *gl)
75 u8 opcode = ((const struct rss_header *)rsp)->opcode;
77 rsp++; /* skip RSS header */
80 * FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG.
82 if (unlikely(opcode == CPL_FW4_MSG &&
83 ((const struct cpl_fw4_msg *)rsp)->type ==
86 opcode = ((const struct rss_header *)rsp)->opcode;
88 if (opcode != CPL_SGE_EGR_UPDATE) {
89 dev_err(q->adapter, "unexpected FW4/CPL %#x on FW event queue\n",
95 if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
97 } else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
98 const struct cpl_fw6_msg *msg = (const void *)rsp;
100 t4_handle_fw_rpl(q->adapter, msg->data);
102 dev_err(adapter, "unexpected CPL %#x on FW event queue\n",
109 int setup_sge_fwevtq(struct adapter *adapter)
111 struct sge *s = &adapter->sge;
115 err = t4_sge_alloc_rxq(adapter, &s->fw_evtq, true, adapter->eth_dev,
116 msi_idx, NULL, fwevtq_handler, -1, NULL, 0,
121 static int closest_timer(const struct sge *s, int time)
123 unsigned int i, match = 0;
124 int delta, min_delta = INT_MAX;
126 for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
127 delta = time - s->timer_val[i];
130 if (delta < min_delta) {
138 static int closest_thres(const struct sge *s, int thres)
140 unsigned int i, match = 0;
141 int delta, min_delta = INT_MAX;
143 for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
144 delta = thres - s->counter_val[i];
147 if (delta < min_delta) {
156 * cxgb4_set_rspq_intr_params - set a queue's interrupt holdoff parameters
158 * @us: the hold-off time in us, or 0 to disable timer
159 * @cnt: the hold-off packet count, or 0 to disable counter
161 * Sets an Rx queue's interrupt hold-off time and packet count. At least
162 * one of the two needs to be enabled for the queue to generate interrupts.
164 int cxgb4_set_rspq_intr_params(struct sge_rspq *q, unsigned int us,
167 struct adapter *adap = q->adapter;
168 unsigned int timer_val;
174 new_idx = closest_thres(&adap->sge, cnt);
175 if (q->desc && q->pktcnt_idx != new_idx) {
176 /* the queue has already been created, update it */
177 v = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
179 FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
180 V_FW_PARAMS_PARAM_YZ(q->cntxt_id);
181 err = t4_set_params(adap, adap->mbox, adap->pf, 0, 1,
186 q->pktcnt_idx = new_idx;
189 timer_val = (us == 0) ? X_TIMERREG_RESTART_COUNTER :
190 closest_timer(&adap->sge, us);
193 q->intr_params = V_QINTR_TIMER_IDX(X_TIMERREG_UPDATE_CIDX);
195 q->intr_params = V_QINTR_TIMER_IDX(timer_val) |
196 V_QINTR_CNT_EN(cnt > 0);
200 static inline bool is_x_1g_port(const struct link_config *lc)
202 return (lc->pcaps & FW_PORT_CAP32_SPEED_1G) != 0;
205 static inline bool is_x_10g_port(const struct link_config *lc)
207 unsigned int speeds, high_speeds;
209 speeds = V_FW_PORT_CAP32_SPEED(G_FW_PORT_CAP32_SPEED(lc->pcaps));
210 high_speeds = speeds &
211 ~(FW_PORT_CAP32_SPEED_100M | FW_PORT_CAP32_SPEED_1G);
213 return high_speeds != 0;
216 inline void init_rspq(struct adapter *adap, struct sge_rspq *q,
217 unsigned int us, unsigned int cnt,
218 unsigned int size, unsigned int iqe_size)
221 cxgb4_set_rspq_intr_params(q, us, cnt);
222 q->iqe_len = iqe_size;
226 int cfg_queue_count(struct rte_eth_dev *eth_dev)
228 struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
229 struct adapter *adap = pi->adapter;
230 struct sge *s = &adap->sge;
231 unsigned int max_queues = s->max_ethqsets / adap->params.nports;
233 if ((eth_dev->data->nb_rx_queues < 1) ||
234 (eth_dev->data->nb_tx_queues < 1))
237 if ((eth_dev->data->nb_rx_queues > max_queues) ||
238 (eth_dev->data->nb_tx_queues > max_queues))
241 if (eth_dev->data->nb_rx_queues > pi->rss_size)
244 /* We must configure RSS, since config has changed*/
245 pi->flags &= ~PORT_RSS_DONE;
247 pi->n_rx_qsets = eth_dev->data->nb_rx_queues;
248 pi->n_tx_qsets = eth_dev->data->nb_tx_queues;
253 void cfg_queues(struct rte_eth_dev *eth_dev)
255 struct rte_config *config = rte_eal_get_configuration();
256 struct port_info *pi = (struct port_info *)(eth_dev->data->dev_private);
257 struct adapter *adap = pi->adapter;
258 struct sge *s = &adap->sge;
259 unsigned int i, nb_ports = 0, qidx = 0;
260 unsigned int q_per_port = 0;
262 if (!(adap->flags & CFG_QUEUES)) {
263 for_each_port(adap, i) {
264 struct port_info *tpi = adap2pinfo(adap, i);
266 nb_ports += (is_x_10g_port(&tpi->link_cfg)) ||
267 is_x_1g_port(&tpi->link_cfg) ? 1 : 0;
271 * We default up to # of cores queues per 1G/10G port.
274 q_per_port = (MAX_ETH_QSETS -
275 (adap->params.nports - nb_ports)) /
278 if (q_per_port > config->lcore_count)
279 q_per_port = config->lcore_count;
281 for_each_port(adap, i) {
282 struct port_info *pi = adap2pinfo(adap, i);
284 pi->first_qset = qidx;
286 /* Initially n_rx_qsets == n_tx_qsets */
287 pi->n_rx_qsets = (is_x_10g_port(&pi->link_cfg) ||
288 is_x_1g_port(&pi->link_cfg)) ?
290 pi->n_tx_qsets = pi->n_rx_qsets;
292 if (pi->n_rx_qsets > pi->rss_size)
293 pi->n_rx_qsets = pi->rss_size;
295 qidx += pi->n_rx_qsets;
298 s->max_ethqsets = qidx;
300 for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
301 struct sge_eth_rxq *r = &s->ethrxq[i];
303 init_rspq(adap, &r->rspq, 5, 32, 1024, 64);
305 r->fl.size = (r->usembufs ? 1024 : 72);
308 for (i = 0; i < ARRAY_SIZE(s->ethtxq); i++)
309 s->ethtxq[i].q.size = 1024;
311 init_rspq(adap, &adap->sge.fw_evtq, 0, 0, 1024, 64);
312 adap->flags |= CFG_QUEUES;
316 void cxgbe_stats_get(struct port_info *pi, struct port_stats *stats)
318 t4_get_port_stats_offset(pi->adapter, pi->tx_chan, stats,
322 void cxgbe_stats_reset(struct port_info *pi)
324 t4_clr_port_stats(pi->adapter, pi->tx_chan);
327 static void setup_memwin(struct adapter *adap)
331 /* For T5, only relative offset inside the PCIe BAR is passed */
332 mem_win0_base = MEMWIN0_BASE;
335 * Set up memory window for accessing adapter memory ranges. (Read
336 * back MA register to ensure that changes propagate before we attempt
337 * to use the new values.)
340 PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN,
342 mem_win0_base | V_BIR(0) |
343 V_WINDOW(ilog2(MEMWIN0_APERTURE) - X_WINDOW_SHIFT));
345 PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN,
349 int init_rss(struct adapter *adap)
356 err = t4_init_rss_mode(adap, adap->mbox);
361 for_each_port(adap, i) {
362 struct port_info *pi = adap2pinfo(adap, i);
364 pi->rss = rte_zmalloc(NULL, pi->rss_size * sizeof(u16), 0);
368 pi->rss_hf = CXGBE_RSS_HF_ALL;
374 * Dump basic information about the adapter.
376 void print_adapter_info(struct adapter *adap)
379 * Hardware/Firmware/etc. Version/Revision IDs.
381 t4_dump_version_info(adap);
384 void print_port_info(struct adapter *adap)
388 struct rte_pci_addr *loc = &adap->pdev->addr;
390 for_each_port(adap, i) {
391 const struct port_info *pi = adap2pinfo(adap, i);
394 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_100M)
395 bufp += sprintf(bufp, "100M/");
396 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_1G)
397 bufp += sprintf(bufp, "1G/");
398 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_10G)
399 bufp += sprintf(bufp, "10G/");
400 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_25G)
401 bufp += sprintf(bufp, "25G/");
402 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_40G)
403 bufp += sprintf(bufp, "40G/");
404 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_50G)
405 bufp += sprintf(bufp, "50G/");
406 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_100G)
407 bufp += sprintf(bufp, "100G/");
410 sprintf(bufp, "BASE-%s",
411 t4_get_port_type_description(
412 (enum fw_port_type)pi->port_type));
415 " " PCI_PRI_FMT " Chelsio rev %d %s %s\n",
416 loc->domain, loc->bus, loc->devid, loc->function,
417 CHELSIO_CHIP_RELEASE(adap->params.chip), buf,
418 (adap->flags & USING_MSIX) ? " MSI-X" :
419 (adap->flags & USING_MSI) ? " MSI" : "");
423 static void configure_pcie_ext_tag(struct adapter *adapter)
426 int pos = t4_os_find_pci_capability(adapter, PCI_CAP_ID_EXP);
432 t4_os_pci_read_cfg2(adapter, pos + PCI_EXP_DEVCTL, &v);
433 v |= PCI_EXP_DEVCTL_EXT_TAG;
434 t4_os_pci_write_cfg2(adapter, pos + PCI_EXP_DEVCTL, v);
435 if (is_t6(adapter->params.chip)) {
436 t4_set_reg_field(adapter, A_PCIE_CFG2,
437 V_T6_TOTMAXTAG(M_T6_TOTMAXTAG),
439 t4_set_reg_field(adapter, A_PCIE_CMD_CFG,
440 V_T6_MINTAG(M_T6_MINTAG),
443 t4_set_reg_field(adapter, A_PCIE_CFG2,
444 V_TOTMAXTAG(M_TOTMAXTAG),
446 t4_set_reg_field(adapter, A_PCIE_CMD_CFG,
454 * Tweak configuration based on system architecture, etc. Most of these have
455 * defaults assigned to them by Firmware Configuration Files (if we're using
456 * them) but need to be explicitly set if we're using hard-coded
457 * initialization. So these are essentially common tweaks/settings for
458 * Configuration Files and hard-coded initialization ...
460 static int adap_init0_tweaks(struct adapter *adapter)
465 * Fix up various Host-Dependent Parameters like Page Size, Cache
466 * Line Size, etc. The firmware default is for a 4KB Page Size and
467 * 64B Cache Line Size ...
469 t4_fixup_host_params_compat(adapter, CXGBE_PAGE_SIZE, L1_CACHE_BYTES,
473 * Keep the chip default offset to deliver Ingress packets into our
474 * DMA buffers to zero
477 t4_set_reg_field(adapter, A_SGE_CONTROL, V_PKTSHIFT(M_PKTSHIFT),
478 V_PKTSHIFT(rx_dma_offset));
480 t4_set_reg_field(adapter, A_SGE_FLM_CFG,
481 V_CREDITCNT(M_CREDITCNT) | M_CREDITCNTPACKING,
482 V_CREDITCNT(3) | V_CREDITCNTPACKING(1));
484 t4_set_reg_field(adapter, A_SGE_INGRESS_RX_THRESHOLD,
485 V_THRESHOLD_3(M_THRESHOLD_3), V_THRESHOLD_3(32U));
487 t4_set_reg_field(adapter, A_SGE_CONTROL2, V_IDMAARBROUNDROBIN(1U),
488 V_IDMAARBROUNDROBIN(1U));
491 * Don't include the "IP Pseudo Header" in CPL_RX_PKT checksums: Linux
492 * adds the pseudo header itself.
494 t4_tp_wr_bits_indirect(adapter, A_TP_INGRESS_CONFIG,
495 F_CSUM_HAS_PSEUDO_HDR, 0);
501 * Attempt to initialize the adapter via a Firmware Configuration File.
503 static int adap_init0_config(struct adapter *adapter, int reset)
505 struct fw_caps_config_cmd caps_cmd;
506 unsigned long mtype = 0, maddr = 0;
507 u32 finiver, finicsum, cfcsum;
509 int config_issued = 0;
511 char config_name[20];
514 * Reset device if necessary.
517 ret = t4_fw_reset(adapter, adapter->mbox,
518 F_PIORSTMODE | F_PIORST);
520 dev_warn(adapter, "Firmware reset failed, error %d\n",
526 cfg_addr = t4_flash_cfg_addr(adapter);
529 dev_warn(adapter, "Finding address for firmware config file in flash failed, error %d\n",
534 strcpy(config_name, "On Flash");
535 mtype = FW_MEMTYPE_CF_FLASH;
539 * Issue a Capability Configuration command to the firmware to get it
540 * to parse the Configuration File. We don't use t4_fw_config_file()
541 * because we want the ability to modify various features after we've
542 * processed the configuration file ...
544 memset(&caps_cmd, 0, sizeof(caps_cmd));
545 caps_cmd.op_to_write = cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
546 F_FW_CMD_REQUEST | F_FW_CMD_READ);
547 caps_cmd.cfvalid_to_len16 =
548 cpu_to_be32(F_FW_CAPS_CONFIG_CMD_CFVALID |
549 V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
550 V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(maddr >> 16) |
552 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
555 * If the CAPS_CONFIG failed with an ENOENT (for a Firmware
556 * Configuration File in FLASH), our last gasp effort is to use the
557 * Firmware Configuration File which is embedded in the firmware. A
558 * very few early versions of the firmware didn't have one embedded
559 * but we can ignore those.
561 if (ret == -ENOENT) {
562 dev_info(adapter, "%s: Going for embedded config in firmware..\n",
565 memset(&caps_cmd, 0, sizeof(caps_cmd));
566 caps_cmd.op_to_write =
567 cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
568 F_FW_CMD_REQUEST | F_FW_CMD_READ);
569 caps_cmd.cfvalid_to_len16 = cpu_to_be32(FW_LEN16(caps_cmd));
570 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd,
571 sizeof(caps_cmd), &caps_cmd);
572 strcpy(config_name, "Firmware Default");
579 finiver = be32_to_cpu(caps_cmd.finiver);
580 finicsum = be32_to_cpu(caps_cmd.finicsum);
581 cfcsum = be32_to_cpu(caps_cmd.cfcsum);
582 if (finicsum != cfcsum)
583 dev_warn(adapter, "Configuration File checksum mismatch: [fini] csum=%#x, computed csum=%#x\n",
587 * If we're a pure NIC driver then disable all offloading facilities.
588 * This will allow the firmware to optimize aspects of the hardware
589 * configuration which will result in improved performance.
591 caps_cmd.niccaps &= cpu_to_be16(~(FW_CAPS_CONFIG_NIC_HASHFILTER |
592 FW_CAPS_CONFIG_NIC_ETHOFLD));
593 caps_cmd.toecaps = 0;
594 caps_cmd.iscsicaps = 0;
595 caps_cmd.rdmacaps = 0;
596 caps_cmd.fcoecaps = 0;
599 * And now tell the firmware to use the configuration we just loaded.
601 caps_cmd.op_to_write = cpu_to_be32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
602 F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
603 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
604 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
607 dev_warn(adapter, "Unable to finalize Firmware Capabilities %d\n",
613 * Tweak configuration based on system architecture, etc.
615 ret = adap_init0_tweaks(adapter);
617 dev_warn(adapter, "Unable to do init0-tweaks %d\n", -ret);
622 * And finally tell the firmware to initialize itself using the
623 * parameters from the Configuration File.
625 ret = t4_fw_initialize(adapter, adapter->mbox);
627 dev_warn(adapter, "Initializing Firmware failed, error %d\n",
633 * Return successfully and note that we're operating with parameters
634 * not supplied by the driver, rather than from hard-wired
635 * initialization constants buried in the driver.
638 "Successfully configured using Firmware Configuration File \"%s\", version %#x, computed checksum %#x\n",
639 config_name, finiver, cfcsum);
644 * Something bad happened. Return the error ... (If the "error"
645 * is that there's no Configuration File on the adapter we don't
646 * want to issue a warning since this is fairly common.)
649 if (config_issued && ret != -ENOENT)
650 dev_warn(adapter, "\"%s\" configuration file error %d\n",
653 dev_debug(adapter, "%s: returning ret = %d ..\n", __func__, ret);
657 static int adap_init0(struct adapter *adap)
661 enum dev_state state;
662 u32 params[7], val[7];
664 int mbox = adap->mbox;
667 * Contact FW, advertising Master capability.
669 ret = t4_fw_hello(adap, adap->mbox, adap->mbox, MASTER_MAY, &state);
671 dev_err(adap, "%s: could not connect to FW, error %d\n",
676 CXGBE_DEBUG_MBOX(adap, "%s: adap->mbox = %d; ret = %d\n", __func__,
680 adap->flags |= MASTER_PF;
682 if (state == DEV_STATE_INIT) {
684 * Force halt and reset FW because a previous instance may have
685 * exited abnormally without properly shutting down
687 ret = t4_fw_halt(adap, adap->mbox, reset);
689 dev_err(adap, "Failed to halt. Exit.\n");
693 ret = t4_fw_restart(adap, adap->mbox, reset);
695 dev_err(adap, "Failed to restart. Exit.\n");
698 state = (enum dev_state)((unsigned)state & ~DEV_STATE_INIT);
701 t4_get_version_info(adap);
703 ret = t4_get_core_clock(adap, &adap->params.vpd);
705 dev_err(adap, "%s: could not get core clock, error %d\n",
711 * If the firmware is initialized already (and we're not forcing a
712 * master initialization), note that we're living with existing
713 * adapter parameters. Otherwise, it's time to try initializing the
716 if (state == DEV_STATE_INIT) {
717 dev_info(adap, "Coming up as %s: Adapter already initialized\n",
718 adap->flags & MASTER_PF ? "MASTER" : "SLAVE");
720 dev_info(adap, "Coming up as MASTER: Initializing adapter\n");
722 ret = adap_init0_config(adap, reset);
723 if (ret == -ENOENT) {
725 "No Configuration File present on adapter. Using hard-wired configuration parameters.\n");
730 dev_err(adap, "could not initialize adapter, error %d\n", -ret);
734 /* Find out what ports are available to us. */
735 v = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
736 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_PORTVEC);
737 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, &v, &port_vec);
739 dev_err(adap, "%s: failure in t4_query_params; error = %d\n",
744 adap->params.nports = hweight32(port_vec);
745 adap->params.portvec = port_vec;
747 dev_debug(adap, "%s: adap->params.nports = %u\n", __func__,
748 adap->params.nports);
751 * Give the SGE code a chance to pull in anything that it needs ...
752 * Note that this must be called after we retrieve our VPD parameters
753 * in order to know how to convert core ticks to seconds, etc.
755 ret = t4_sge_init(adap);
757 dev_err(adap, "t4_sge_init failed with error %d\n",
763 * Grab some of our basic fundamental operating parameters.
765 #define FW_PARAM_DEV(param) \
766 (V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
767 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
769 #define FW_PARAM_PFVF(param) \
770 (V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
771 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param) | \
772 V_FW_PARAMS_PARAM_Y(0) | \
773 V_FW_PARAMS_PARAM_Z(0))
775 /* If we're running on newer firmware, let it know that we're
776 * prepared to deal with encapsulated CPL messages. Older
777 * firmware won't understand this and we'll just get
778 * unencapsulated messages ...
780 params[0] = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
782 (void)t4_set_params(adap, adap->mbox, adap->pf, 0, 1, params, val);
785 * Find out whether we're allowed to use the T5+ ULPTX MEMWRITE DSGL
786 * capability. Earlier versions of the firmware didn't have the
787 * ULPTX_MEMWRITE_DSGL so we'll interpret a query failure as no
788 * permission to use ULPTX MEMWRITE DSGL.
790 if (is_t4(adap->params.chip)) {
791 adap->params.ulptx_memwrite_dsgl = false;
793 params[0] = FW_PARAM_DEV(ULPTX_MEMWRITE_DSGL);
794 ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
796 adap->params.ulptx_memwrite_dsgl = (ret == 0 && val[0] != 0);
800 * The MTU/MSS Table is initialized by now, so load their values. If
801 * we're initializing the adapter, then we'll make any modifications
802 * we want to the MTU/MSS Table and also initialize the congestion
805 t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
806 if (state != DEV_STATE_INIT) {
810 * The default MTU Table contains values 1492 and 1500.
811 * However, for TCP, it's better to have two values which are
812 * a multiple of 8 +/- 4 bytes apart near this popular MTU.
813 * This allows us to have a TCP Data Payload which is a
814 * multiple of 8 regardless of what combination of TCP Options
815 * are in use (always a multiple of 4 bytes) which is
816 * important for performance reasons. For instance, if no
817 * options are in use, then we have a 20-byte IP header and a
818 * 20-byte TCP header. In this case, a 1500-byte MSS would
819 * result in a TCP Data Payload of 1500 - 40 == 1460 bytes
820 * which is not a multiple of 8. So using an MSS of 1488 in
821 * this case results in a TCP Data Payload of 1448 bytes which
822 * is a multiple of 8. On the other hand, if 12-byte TCP Time
823 * Stamps have been negotiated, then an MTU of 1500 bytes
824 * results in a TCP Data Payload of 1448 bytes which, as
825 * above, is a multiple of 8 bytes ...
827 for (i = 0; i < NMTUS; i++)
828 if (adap->params.mtus[i] == 1492) {
829 adap->params.mtus[i] = 1488;
833 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
836 t4_init_sge_params(adap);
837 t4_init_tp_params(adap);
838 configure_pcie_ext_tag(adap);
840 adap->params.drv_memwin = MEMWIN_NIC;
841 adap->flags |= FW_OK;
842 dev_debug(adap, "%s: returning zero..\n", __func__);
846 * Something bad happened. If a command timed out or failed with EIO
847 * FW does not operate within its spec or something catastrophic
848 * happened to HW/FW, stop issuing commands.
851 if (ret != -ETIMEDOUT && ret != -EIO)
852 t4_fw_bye(adap, adap->mbox);
857 * t4_os_portmod_changed - handle port module changes
858 * @adap: the adapter associated with the module change
859 * @port_id: the port index whose module status has changed
861 * This is the OS-dependent handler for port module changes. It is
862 * invoked when a port module is removed or inserted for any OS-specific
865 void t4_os_portmod_changed(const struct adapter *adap, int port_id)
867 static const char * const mod_str[] = {
868 NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
871 const struct port_info *pi = adap2pinfo(adap, port_id);
873 if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
874 dev_info(adap, "Port%d: port module unplugged\n", pi->port_id);
875 else if (pi->mod_type < ARRAY_SIZE(mod_str))
876 dev_info(adap, "Port%d: %s port module inserted\n", pi->port_id,
877 mod_str[pi->mod_type]);
878 else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
879 dev_info(adap, "Port%d: unsupported port module inserted\n",
881 else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
882 dev_info(adap, "Port%d: unknown port module inserted\n",
884 else if (pi->mod_type == FW_PORT_MOD_TYPE_ERROR)
885 dev_info(adap, "Port%d: transceiver module error\n",
888 dev_info(adap, "Port%d: unknown module type %d inserted\n",
889 pi->port_id, pi->mod_type);
893 * link_start - enable a port
894 * @dev: the port to enable
896 * Performs the MAC and PHY actions needed to enable a port.
898 int link_start(struct port_info *pi)
900 struct adapter *adapter = pi->adapter;
904 mtu = pi->eth_dev->data->dev_conf.rxmode.max_rx_pkt_len -
905 (ETHER_HDR_LEN + ETHER_CRC_LEN);
908 * We do not set address filters and promiscuity here, the stack does
909 * that step explicitly.
911 ret = t4_set_rxmode(adapter, adapter->mbox, pi->viid, mtu, -1, -1,
914 ret = t4_change_mac(adapter, adapter->mbox, pi->viid,
916 (u8 *)&pi->eth_dev->data->mac_addrs[0],
919 pi->xact_addr_filt = ret;
923 if (ret == 0 && is_pf4(adapter))
924 ret = t4_link_l1cfg(adapter, adapter->mbox, pi->tx_chan,
928 * Enabling a Virtual Interface can result in an interrupt
929 * during the processing of the VI Enable command and, in some
930 * paths, result in an attempt to issue another command in the
931 * interrupt context. Thus, we disable interrupts during the
932 * course of the VI Enable command ...
934 ret = t4_enable_vi_params(adapter, adapter->mbox, pi->viid,
941 * cxgbe_write_rss_conf - flash the RSS configuration for a given port
943 * @rss_hf: Hash configuration to apply
945 int cxgbe_write_rss_conf(const struct port_info *pi, uint64_t rss_hf)
947 struct adapter *adapter = pi->adapter;
948 const struct sge_eth_rxq *rxq;
953 /* Should never be called before setting up sge eth rx queues */
954 if (!(adapter->flags & FULL_INIT_DONE)) {
955 dev_err(adap, "%s No RXQs available on port %d\n",
956 __func__, pi->port_id);
960 /* Don't allow unsupported hash functions */
961 if (rss_hf & ~CXGBE_RSS_HF_ALL)
964 if (rss_hf & ETH_RSS_IPV4)
965 flags |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
967 if (rss_hf & ETH_RSS_NONFRAG_IPV4_TCP)
968 flags |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
970 if (rss_hf & ETH_RSS_NONFRAG_IPV4_UDP)
971 flags |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
972 F_FW_RSS_VI_CONFIG_CMD_UDPEN;
974 if (rss_hf & ETH_RSS_IPV6)
975 flags |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
977 if (rss_hf & ETH_RSS_NONFRAG_IPV6_TCP)
978 flags |= F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
980 if (rss_hf & ETH_RSS_NONFRAG_IPV6_UDP)
981 flags |= F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
982 F_FW_RSS_VI_CONFIG_CMD_UDPEN;
984 rxq = &adapter->sge.ethrxq[pi->first_qset];
985 rss = rxq[0].rspq.abs_id;
987 /* If Tunnel All Lookup isn't specified in the global RSS
988 * Configuration, then we need to specify a default Ingress
989 * Queue for any ingress packets which aren't hashed. We'll
990 * use our first ingress queue ...
992 err = t4_config_vi_rss(adapter, adapter->mbox, pi->viid,
998 * cxgbe_write_rss - write the RSS table for a given port
1000 * @queues: array of queue indices for RSS
1002 * Sets up the portion of the HW RSS table for the port's VI to distribute
1003 * packets to the Rx queues in @queues.
1005 int cxgbe_write_rss(const struct port_info *pi, const u16 *queues)
1009 struct adapter *adapter = pi->adapter;
1010 const struct sge_eth_rxq *rxq;
1012 /* Should never be called before setting up sge eth rx queues */
1013 BUG_ON(!(adapter->flags & FULL_INIT_DONE));
1015 rxq = &adapter->sge.ethrxq[pi->first_qset];
1016 rss = rte_zmalloc(NULL, pi->rss_size * sizeof(u16), 0);
1020 /* map the queue indices to queue ids */
1021 for (i = 0; i < pi->rss_size; i++, queues++)
1022 rss[i] = rxq[*queues].rspq.abs_id;
1024 err = t4_config_rss_range(adapter, adapter->pf, pi->viid, 0,
1025 pi->rss_size, rss, pi->rss_size);
1031 * setup_rss - configure RSS
1032 * @adapter: the adapter
1034 * Sets up RSS to distribute packets to multiple receive queues. We
1035 * configure the RSS CPU lookup table to distribute to the number of HW
1036 * receive queues, and the response queue lookup table to narrow that
1037 * down to the response queues actually configured for each port.
1038 * We always configure the RSS mapping for all ports since the mapping
1039 * table has plenty of entries.
1041 int setup_rss(struct port_info *pi)
1044 struct adapter *adapter = pi->adapter;
1046 dev_debug(adapter, "%s: pi->rss_size = %u; pi->n_rx_qsets = %u\n",
1047 __func__, pi->rss_size, pi->n_rx_qsets);
1049 if (!(pi->flags & PORT_RSS_DONE)) {
1050 if (adapter->flags & FULL_INIT_DONE) {
1051 /* Fill default values with equal distribution */
1052 for (j = 0; j < pi->rss_size; j++)
1053 pi->rss[j] = j % pi->n_rx_qsets;
1055 err = cxgbe_write_rss(pi, pi->rss);
1059 err = cxgbe_write_rss_conf(pi, pi->rss_hf);
1062 pi->flags |= PORT_RSS_DONE;
1069 * Enable NAPI scheduling and interrupt generation for all Rx queues.
1071 static void enable_rx(struct adapter *adap, struct sge_rspq *q)
1073 /* 0-increment GTS to start the timer and enable interrupts */
1074 t4_write_reg(adap, is_pf4(adap) ? MYPF_REG(A_SGE_PF_GTS) :
1075 T4VF_SGE_BASE_ADDR + A_SGE_VF_GTS,
1076 V_SEINTARM(q->intr_params) |
1077 V_INGRESSQID(q->cntxt_id));
1080 void cxgbe_enable_rx_queues(struct port_info *pi)
1082 struct adapter *adap = pi->adapter;
1083 struct sge *s = &adap->sge;
1086 for (i = 0; i < pi->n_rx_qsets; i++)
1087 enable_rx(adap, &s->ethrxq[pi->first_qset + i].rspq);
1091 * fw_caps_to_speed_caps - translate Firmware Port Caps to Speed Caps.
1092 * @port_type: Firmware Port Type
1093 * @fw_caps: Firmware Port Capabilities
1094 * @speed_caps: Device Info Speed Capabilities
1096 * Translate a Firmware Port Capabilities specification to Device Info
1097 * Speed Capabilities.
1099 static void fw_caps_to_speed_caps(enum fw_port_type port_type,
1100 unsigned int fw_caps,
1103 #define SET_SPEED(__speed_name) \
1105 *speed_caps |= ETH_LINK_ ## __speed_name; \
1108 #define FW_CAPS_TO_SPEED(__fw_name) \
1110 if (fw_caps & FW_PORT_CAP32_ ## __fw_name) \
1111 SET_SPEED(__fw_name); \
1114 switch (port_type) {
1115 case FW_PORT_TYPE_BT_SGMII:
1116 case FW_PORT_TYPE_BT_XFI:
1117 case FW_PORT_TYPE_BT_XAUI:
1118 FW_CAPS_TO_SPEED(SPEED_100M);
1119 FW_CAPS_TO_SPEED(SPEED_1G);
1120 FW_CAPS_TO_SPEED(SPEED_10G);
1123 case FW_PORT_TYPE_KX4:
1124 case FW_PORT_TYPE_KX:
1125 case FW_PORT_TYPE_FIBER_XFI:
1126 case FW_PORT_TYPE_FIBER_XAUI:
1127 case FW_PORT_TYPE_SFP:
1128 case FW_PORT_TYPE_QSFP_10G:
1129 case FW_PORT_TYPE_QSA:
1130 FW_CAPS_TO_SPEED(SPEED_1G);
1131 FW_CAPS_TO_SPEED(SPEED_10G);
1134 case FW_PORT_TYPE_KR:
1135 SET_SPEED(SPEED_10G);
1138 case FW_PORT_TYPE_BP_AP:
1139 case FW_PORT_TYPE_BP4_AP:
1140 SET_SPEED(SPEED_1G);
1141 SET_SPEED(SPEED_10G);
1144 case FW_PORT_TYPE_BP40_BA:
1145 case FW_PORT_TYPE_QSFP:
1146 SET_SPEED(SPEED_40G);
1149 case FW_PORT_TYPE_CR_QSFP:
1150 case FW_PORT_TYPE_SFP28:
1151 case FW_PORT_TYPE_KR_SFP28:
1152 FW_CAPS_TO_SPEED(SPEED_1G);
1153 FW_CAPS_TO_SPEED(SPEED_10G);
1154 FW_CAPS_TO_SPEED(SPEED_25G);
1157 case FW_PORT_TYPE_CR2_QSFP:
1158 SET_SPEED(SPEED_50G);
1161 case FW_PORT_TYPE_KR4_100G:
1162 case FW_PORT_TYPE_CR4_QSFP:
1163 FW_CAPS_TO_SPEED(SPEED_25G);
1164 FW_CAPS_TO_SPEED(SPEED_40G);
1165 FW_CAPS_TO_SPEED(SPEED_50G);
1166 FW_CAPS_TO_SPEED(SPEED_100G);
1173 #undef FW_CAPS_TO_SPEED
1178 * cxgbe_get_speed_caps - Fetch supported speed capabilities
1179 * @pi: Underlying port's info
1180 * @speed_caps: Device Info speed capabilities
1182 * Fetch supported speed capabilities of the underlying port.
1184 void cxgbe_get_speed_caps(struct port_info *pi, u32 *speed_caps)
1188 fw_caps_to_speed_caps(pi->port_type, pi->link_cfg.pcaps,
1191 if (!(pi->link_cfg.pcaps & FW_PORT_CAP32_ANEG))
1192 *speed_caps |= ETH_LINK_SPEED_FIXED;
1196 * cxgb_up - enable the adapter
1197 * @adap: adapter being enabled
1199 * Called when the first port is enabled, this function performs the
1200 * actions necessary to make an adapter operational, such as completing
1201 * the initialization of HW modules, and enabling interrupts.
1203 int cxgbe_up(struct adapter *adap)
1205 enable_rx(adap, &adap->sge.fw_evtq);
1206 t4_sge_tx_monitor_start(adap);
1208 t4_intr_enable(adap);
1209 adap->flags |= FULL_INIT_DONE;
1211 /* TODO: deadman watchdog ?? */
1218 int cxgbe_down(struct port_info *pi)
1220 struct adapter *adapter = pi->adapter;
1223 err = t4_enable_vi(adapter, adapter->mbox, pi->viid, false, false);
1225 dev_err(adapter, "%s: disable_vi failed: %d\n", __func__, err);
1229 t4_reset_link_config(adapter, pi->pidx);
1234 * Release resources when all the ports have been stopped.
1236 void cxgbe_close(struct adapter *adapter)
1238 struct port_info *pi;
1241 if (adapter->flags & FULL_INIT_DONE) {
1242 if (is_pf4(adapter))
1243 t4_intr_disable(adapter);
1244 t4_sge_tx_monitor_stop(adapter);
1245 t4_free_sge_resources(adapter);
1246 for_each_port(adapter, i) {
1247 pi = adap2pinfo(adapter, i);
1249 t4_free_vi(adapter, adapter->mbox,
1250 adapter->pf, 0, pi->viid);
1251 rte_free(pi->eth_dev->data->mac_addrs);
1252 /* Skip first port since it'll be freed by DPDK stack */
1254 rte_free(pi->eth_dev->data->dev_private);
1255 rte_eth_dev_release_port(pi->eth_dev);
1258 adapter->flags &= ~FULL_INIT_DONE;
1261 if (is_pf4(adapter) && (adapter->flags & FW_OK))
1262 t4_fw_bye(adapter, adapter->mbox);
1265 int cxgbe_probe(struct adapter *adapter)
1267 struct port_info *pi;
1273 whoami = t4_read_reg(adapter, A_PL_WHOAMI);
1274 chip = t4_get_chip_type(adapter,
1275 CHELSIO_PCI_ID_VER(adapter->pdev->id.device_id));
1279 func = CHELSIO_CHIP_VERSION(chip) <= CHELSIO_T5 ?
1280 G_SOURCEPF(whoami) : G_T6_SOURCEPF(whoami);
1282 adapter->mbox = func;
1285 t4_os_lock_init(&adapter->mbox_lock);
1286 TAILQ_INIT(&adapter->mbox_list);
1288 err = t4_prep_adapter(adapter);
1292 setup_memwin(adapter);
1293 err = adap_init0(adapter);
1295 dev_err(adapter, "%s: Adapter initialization failed, error %d\n",
1300 if (!is_t4(adapter->params.chip)) {
1302 * The userspace doorbell BAR is split evenly into doorbell
1303 * regions, each associated with an egress queue. If this
1304 * per-queue region is large enough (at least UDBS_SEG_SIZE)
1305 * then it can be used to submit a tx work request with an
1306 * implied doorbell. Enable write combining on the BAR if
1307 * there is room for such work requests.
1309 int s_qpp, qpp, num_seg;
1311 s_qpp = (S_QUEUESPERPAGEPF0 +
1312 (S_QUEUESPERPAGEPF1 - S_QUEUESPERPAGEPF0) *
1314 qpp = 1 << ((t4_read_reg(adapter,
1315 A_SGE_EGRESS_QUEUES_PER_PAGE_PF) >> s_qpp)
1316 & M_QUEUESPERPAGEPF0);
1317 num_seg = CXGBE_PAGE_SIZE / UDBS_SEG_SIZE;
1319 dev_warn(adapter, "Incorrect SGE EGRESS QUEUES_PER_PAGE configuration, continuing in debug mode\n");
1321 adapter->bar2 = (void *)adapter->pdev->mem_resource[2].addr;
1322 if (!adapter->bar2) {
1323 dev_err(adapter, "cannot map device bar2 region\n");
1327 t4_write_reg(adapter, A_SGE_STAT_CFG, V_STATSOURCE_T5(7) |
1331 for_each_port(adapter, i) {
1332 const unsigned int numa_node = rte_socket_id();
1333 char name[RTE_ETH_NAME_MAX_LEN];
1334 struct rte_eth_dev *eth_dev;
1336 snprintf(name, sizeof(name), "%s_%d",
1337 adapter->pdev->device.name, i);
1340 /* First port is already allocated by DPDK */
1341 eth_dev = adapter->eth_dev;
1346 * now do all data allocation - for eth_dev structure,
1347 * and internal (private) data for the remaining ports
1350 /* reserve an ethdev entry */
1351 eth_dev = rte_eth_dev_allocate(name);
1355 eth_dev->data->dev_private =
1356 rte_zmalloc_socket(name, sizeof(struct port_info),
1357 RTE_CACHE_LINE_SIZE, numa_node);
1358 if (!eth_dev->data->dev_private)
1362 pi = (struct port_info *)eth_dev->data->dev_private;
1363 adapter->port[i] = pi;
1364 pi->eth_dev = eth_dev;
1365 pi->adapter = adapter;
1366 pi->xact_addr_filt = -1;
1370 pi->eth_dev->device = &adapter->pdev->device;
1371 pi->eth_dev->dev_ops = adapter->eth_dev->dev_ops;
1372 pi->eth_dev->tx_pkt_burst = adapter->eth_dev->tx_pkt_burst;
1373 pi->eth_dev->rx_pkt_burst = adapter->eth_dev->rx_pkt_burst;
1375 rte_eth_copy_pci_info(pi->eth_dev, adapter->pdev);
1377 pi->eth_dev->data->mac_addrs = rte_zmalloc(name,
1379 if (!pi->eth_dev->data->mac_addrs) {
1380 dev_err(adapter, "%s: Mem allocation failed for storing mac addr, aborting\n",
1387 if (adapter->flags & FW_OK) {
1388 err = t4_port_init(adapter, adapter->mbox, adapter->pf, 0);
1390 dev_err(adapter, "%s: t4_port_init failed with err %d\n",
1396 cfg_queues(adapter->eth_dev);
1398 print_adapter_info(adapter);
1399 print_port_info(adapter);
1401 err = init_rss(adapter);
1408 for_each_port(adapter, i) {
1409 pi = adap2pinfo(adapter, i);
1411 t4_free_vi(adapter, adapter->mbox, adapter->pf,
1413 /* Skip first port since it'll be de-allocated by DPDK */
1417 if (pi->eth_dev->data->dev_private)
1418 rte_free(pi->eth_dev->data->dev_private);
1419 rte_eth_dev_release_port(pi->eth_dev);
1423 if (adapter->flags & FW_OK)
1424 t4_fw_bye(adapter, adapter->mbox);