2 * This file is part of the Chelsio T4 Ethernet driver for Linux.
4 * Copyright (c) 2003-2016 Chelsio Communications, Inc. All rights reserved.
6 * This software is available to you under a choice of one of two
7 * licenses. You may choose to be licensed under the terms of the GNU
8 * General Public License (GPL) Version 2, available from the file
9 * COPYING in the main directory of this source tree, or the
10 * OpenIB.org BSD license below:
12 * Redistribution and use in source and binary forms, with or
13 * without modification, are permitted provided that the following
16 * - Redistributions of source code must retain the above
17 * copyright notice, this list of conditions and the following
20 * - Redistributions in binary form must reproduce the above
21 * copyright notice, this list of conditions and the following
22 * disclaimer in the documentation and/or other materials
23 * provided with the distribution.
25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
35 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
37 #include <linux/bitmap.h>
38 #include <linux/crc32.h>
39 #include <linux/ctype.h>
40 #include <linux/debugfs.h>
41 #include <linux/err.h>
42 #include <linux/etherdevice.h>
43 #include <linux/firmware.h>
45 #include <linux/if_vlan.h>
46 #include <linux/init.h>
47 #include <linux/log2.h>
48 #include <linux/mdio.h>
49 #include <linux/module.h>
50 #include <linux/moduleparam.h>
51 #include <linux/mutex.h>
52 #include <linux/netdevice.h>
53 #include <linux/pci.h>
54 #include <linux/aer.h>
55 #include <linux/rtnetlink.h>
56 #include <linux/sched.h>
57 #include <linux/seq_file.h>
58 #include <linux/sockios.h>
59 #include <linux/vmalloc.h>
60 #include <linux/workqueue.h>
61 #include <net/neighbour.h>
62 #include <net/netevent.h>
63 #include <net/addrconf.h>
64 #include <net/bonding.h>
65 #include <net/addrconf.h>
66 #include <linux/uaccess.h>
67 #include <linux/crash_dump.h>
68 #include <net/udp_tunnel.h>
71 #include "cxgb4_filter.h"
73 #include "t4_values.h"
76 #include "t4fw_version.h"
77 #include "cxgb4_dcb.h"
78 #include "cxgb4_debugfs.h"
83 #include "cxgb4_tc_u32.h"
84 #include "cxgb4_tc_flower.h"
85 #include "cxgb4_ptp.h"
86 #include "cxgb4_cudbg.h"
88 char cxgb4_driver_name[] = KBUILD_MODNAME;
93 #define DRV_VERSION "2.0.0-ko"
94 const char cxgb4_driver_version[] = DRV_VERSION;
95 #define DRV_DESC "Chelsio T4/T5/T6 Network Driver"
97 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
98 NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
99 NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
101 /* Macros needed to support the PCI Device ID Table ...
103 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_BEGIN \
104 static const struct pci_device_id cxgb4_pci_tbl[] = {
105 #define CXGB4_UNIFIED_PF 0x4
107 #define CH_PCI_DEVICE_ID_FUNCTION CXGB4_UNIFIED_PF
109 /* Include PCI Device IDs for both PF4 and PF0-3 so our PCI probe() routine is
112 #define CH_PCI_DEVICE_ID_FUNCTION2 0x0
114 #define CH_PCI_ID_TABLE_ENTRY(devid) \
115 {PCI_VDEVICE(CHELSIO, (devid)), CXGB4_UNIFIED_PF}
117 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_END \
121 #include "t4_pci_id_tbl.h"
123 #define FW4_FNAME "cxgb4/t4fw.bin"
124 #define FW5_FNAME "cxgb4/t5fw.bin"
125 #define FW6_FNAME "cxgb4/t6fw.bin"
126 #define FW4_CFNAME "cxgb4/t4-config.txt"
127 #define FW5_CFNAME "cxgb4/t5-config.txt"
128 #define FW6_CFNAME "cxgb4/t6-config.txt"
129 #define PHY_AQ1202_FIRMWARE "cxgb4/aq1202_fw.cld"
130 #define PHY_BCM84834_FIRMWARE "cxgb4/bcm8483.bin"
131 #define PHY_AQ1202_DEVICEID 0x4409
132 #define PHY_BCM84834_DEVICEID 0x4486
134 MODULE_DESCRIPTION(DRV_DESC);
135 MODULE_AUTHOR("Chelsio Communications");
136 MODULE_LICENSE("Dual BSD/GPL");
137 MODULE_VERSION(DRV_VERSION);
138 MODULE_DEVICE_TABLE(pci, cxgb4_pci_tbl);
139 MODULE_FIRMWARE(FW4_FNAME);
140 MODULE_FIRMWARE(FW5_FNAME);
141 MODULE_FIRMWARE(FW6_FNAME);
144 * The driver uses the best interrupt scheme available on a platform in the
145 * order MSI-X, MSI, legacy INTx interrupts. This parameter determines which
146 * of these schemes the driver may consider as follows:
148 * msi = 2: choose from among all three options
149 * msi = 1: only consider MSI and INTx interrupts
150 * msi = 0: force INTx interrupts
154 module_param(msi, int, 0644);
155 MODULE_PARM_DESC(msi, "whether to use INTx (0), MSI (1) or MSI-X (2)");
158 * Normally we tell the chip to deliver Ingress Packets into our DMA buffers
159 * offset by 2 bytes in order to have the IP headers line up on 4-byte
160 * boundaries. This is a requirement for many architectures which will throw
161 * a machine check fault if an attempt is made to access one of the 4-byte IP
162 * header fields on a non-4-byte boundary. And it's a major performance issue
163 * even on some architectures which allow it like some implementations of the
164 * x86 ISA. However, some architectures don't mind this and for some very
165 * edge-case performance sensitive applications (like forwarding large volumes
166 * of small packets), setting this DMA offset to 0 will decrease the number of
167 * PCI-E Bus transfers enough to measurably affect performance.
169 static int rx_dma_offset = 2;
171 /* TX Queue select used to determine what algorithm to use for selecting TX
172 * queue. Select between the kernel provided function (select_queue=0) or user
173 * cxgb_select_queue function (select_queue=1)
175 * Default: select_queue=0
177 static int select_queue;
178 module_param(select_queue, int, 0644);
179 MODULE_PARM_DESC(select_queue,
180 "Select between kernel provided method of selecting or driver method of selecting TX queue. Default is kernel method.");
182 static struct dentry *cxgb4_debugfs_root;
184 LIST_HEAD(adapter_list);
185 DEFINE_MUTEX(uld_mutex);
187 static void link_report(struct net_device *dev)
189 if (!netif_carrier_ok(dev))
190 netdev_info(dev, "link down\n");
192 static const char *fc[] = { "no", "Rx", "Tx", "Tx/Rx" };
195 const struct port_info *p = netdev_priv(dev);
197 switch (p->link_cfg.speed) {
217 pr_info("%s: unsupported speed: %d\n",
218 dev->name, p->link_cfg.speed);
222 netdev_info(dev, "link up, %s, full-duplex, %s PAUSE\n", s,
227 #ifdef CONFIG_CHELSIO_T4_DCB
228 /* Set up/tear down Data Center Bridging Priority mapping for a net device. */
229 static void dcb_tx_queue_prio_enable(struct net_device *dev, int enable)
231 struct port_info *pi = netdev_priv(dev);
232 struct adapter *adap = pi->adapter;
233 struct sge_eth_txq *txq = &adap->sge.ethtxq[pi->first_qset];
236 /* We use a simple mapping of Port TX Queue Index to DCB
237 * Priority when we're enabling DCB.
239 for (i = 0; i < pi->nqsets; i++, txq++) {
243 name = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DMAQ) |
245 FW_PARAMS_PARAM_DMAQ_EQ_DCBPRIO_ETH) |
246 FW_PARAMS_PARAM_YZ_V(txq->q.cntxt_id));
247 value = enable ? i : 0xffffffff;
249 /* Since we can be called while atomic (from "interrupt
250 * level") we need to issue the Set Parameters Commannd
251 * without sleeping (timeout < 0).
253 err = t4_set_params_timeout(adap, adap->mbox, adap->pf, 0, 1,
255 -FW_CMD_MAX_TIMEOUT);
258 dev_err(adap->pdev_dev,
259 "Can't %s DCB Priority on port %d, TX Queue %d: err=%d\n",
260 enable ? "set" : "unset", pi->port_id, i, -err);
262 txq->dcb_prio = value;
266 static int cxgb4_dcb_enabled(const struct net_device *dev)
268 struct port_info *pi = netdev_priv(dev);
270 if (!pi->dcb.enabled)
273 return ((pi->dcb.state == CXGB4_DCB_STATE_FW_ALLSYNCED) ||
274 (pi->dcb.state == CXGB4_DCB_STATE_HOST));
276 #endif /* CONFIG_CHELSIO_T4_DCB */
278 void t4_os_link_changed(struct adapter *adapter, int port_id, int link_stat)
280 struct net_device *dev = adapter->port[port_id];
282 /* Skip changes from disabled ports. */
283 if (netif_running(dev) && link_stat != netif_carrier_ok(dev)) {
285 netif_carrier_on(dev);
287 #ifdef CONFIG_CHELSIO_T4_DCB
288 if (cxgb4_dcb_enabled(dev)) {
289 cxgb4_dcb_reset(dev);
290 dcb_tx_queue_prio_enable(dev, false);
292 #endif /* CONFIG_CHELSIO_T4_DCB */
293 netif_carrier_off(dev);
300 void t4_os_portmod_changed(const struct adapter *adap, int port_id)
302 static const char *mod_str[] = {
303 NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
306 const struct net_device *dev = adap->port[port_id];
307 const struct port_info *pi = netdev_priv(dev);
309 if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
310 netdev_info(dev, "port module unplugged\n");
311 else if (pi->mod_type < ARRAY_SIZE(mod_str))
312 netdev_info(dev, "%s module inserted\n", mod_str[pi->mod_type]);
313 else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
314 netdev_info(dev, "%s: unsupported port module inserted\n",
316 else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
317 netdev_info(dev, "%s: unknown port module inserted\n",
319 else if (pi->mod_type == FW_PORT_MOD_TYPE_ERROR)
320 netdev_info(dev, "%s: transceiver module error\n", dev->name);
322 netdev_info(dev, "%s: unknown module type %d inserted\n",
323 dev->name, pi->mod_type);
326 int dbfifo_int_thresh = 10; /* 10 == 640 entry threshold */
327 module_param(dbfifo_int_thresh, int, 0644);
328 MODULE_PARM_DESC(dbfifo_int_thresh, "doorbell fifo interrupt threshold");
331 * usecs to sleep while draining the dbfifo
333 static int dbfifo_drain_delay = 1000;
334 module_param(dbfifo_drain_delay, int, 0644);
335 MODULE_PARM_DESC(dbfifo_drain_delay,
336 "usecs to sleep while draining the dbfifo");
338 static inline int cxgb4_set_addr_hash(struct port_info *pi)
340 struct adapter *adap = pi->adapter;
343 struct hash_mac_addr *entry;
345 /* Calculate the hash vector for the updated list and program it */
346 list_for_each_entry(entry, &adap->mac_hlist, list) {
347 ucast |= is_unicast_ether_addr(entry->addr);
348 vec |= (1ULL << hash_mac_addr(entry->addr));
350 return t4_set_addr_hash(adap, adap->mbox, pi->viid, ucast,
354 static int cxgb4_mac_sync(struct net_device *netdev, const u8 *mac_addr)
356 struct port_info *pi = netdev_priv(netdev);
357 struct adapter *adap = pi->adapter;
362 bool ucast = is_unicast_ether_addr(mac_addr);
363 const u8 *maclist[1] = {mac_addr};
364 struct hash_mac_addr *new_entry;
366 ret = t4_alloc_mac_filt(adap, adap->mbox, pi->viid, free, 1, maclist,
367 NULL, ucast ? &uhash : &mhash, false);
370 /* if hash != 0, then add the addr to hash addr list
371 * so on the end we will calculate the hash for the
372 * list and program it
374 if (uhash || mhash) {
375 new_entry = kzalloc(sizeof(*new_entry), GFP_ATOMIC);
378 ether_addr_copy(new_entry->addr, mac_addr);
379 list_add_tail(&new_entry->list, &adap->mac_hlist);
380 ret = cxgb4_set_addr_hash(pi);
383 return ret < 0 ? ret : 0;
386 static int cxgb4_mac_unsync(struct net_device *netdev, const u8 *mac_addr)
388 struct port_info *pi = netdev_priv(netdev);
389 struct adapter *adap = pi->adapter;
391 const u8 *maclist[1] = {mac_addr};
392 struct hash_mac_addr *entry, *tmp;
394 /* If the MAC address to be removed is in the hash addr
395 * list, delete it from the list and update hash vector
397 list_for_each_entry_safe(entry, tmp, &adap->mac_hlist, list) {
398 if (ether_addr_equal(entry->addr, mac_addr)) {
399 list_del(&entry->list);
401 return cxgb4_set_addr_hash(pi);
405 ret = t4_free_mac_filt(adap, adap->mbox, pi->viid, 1, maclist, false);
406 return ret < 0 ? -EINVAL : 0;
410 * Set Rx properties of a port, such as promiscruity, address filters, and MTU.
411 * If @mtu is -1 it is left unchanged.
413 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
415 struct port_info *pi = netdev_priv(dev);
416 struct adapter *adapter = pi->adapter;
418 __dev_uc_sync(dev, cxgb4_mac_sync, cxgb4_mac_unsync);
419 __dev_mc_sync(dev, cxgb4_mac_sync, cxgb4_mac_unsync);
421 return t4_set_rxmode(adapter, adapter->mbox, pi->viid, mtu,
422 (dev->flags & IFF_PROMISC) ? 1 : 0,
423 (dev->flags & IFF_ALLMULTI) ? 1 : 0, 1, -1,
428 * link_start - enable a port
429 * @dev: the port to enable
431 * Performs the MAC and PHY actions needed to enable a port.
433 static int link_start(struct net_device *dev)
436 struct port_info *pi = netdev_priv(dev);
437 unsigned int mb = pi->adapter->pf;
440 * We do not set address filters and promiscuity here, the stack does
441 * that step explicitly.
443 ret = t4_set_rxmode(pi->adapter, mb, pi->viid, dev->mtu, -1, -1, -1,
444 !!(dev->features & NETIF_F_HW_VLAN_CTAG_RX), true);
446 ret = t4_change_mac(pi->adapter, mb, pi->viid,
447 pi->xact_addr_filt, dev->dev_addr, true,
450 pi->xact_addr_filt = ret;
455 ret = t4_link_l1cfg(pi->adapter, mb, pi->tx_chan,
459 ret = t4_enable_vi_params(pi->adapter, mb, pi->viid, true,
460 true, CXGB4_DCB_ENABLED);
467 #ifdef CONFIG_CHELSIO_T4_DCB
468 /* Handle a Data Center Bridging update message from the firmware. */
469 static void dcb_rpl(struct adapter *adap, const struct fw_port_cmd *pcmd)
471 int port = FW_PORT_CMD_PORTID_G(ntohl(pcmd->op_to_portid));
472 struct net_device *dev = adap->port[adap->chan_map[port]];
473 int old_dcb_enabled = cxgb4_dcb_enabled(dev);
476 cxgb4_dcb_handle_fw_update(adap, pcmd);
477 new_dcb_enabled = cxgb4_dcb_enabled(dev);
479 /* If the DCB has become enabled or disabled on the port then we're
480 * going to need to set up/tear down DCB Priority parameters for the
481 * TX Queues associated with the port.
483 if (new_dcb_enabled != old_dcb_enabled)
484 dcb_tx_queue_prio_enable(dev, new_dcb_enabled);
486 #endif /* CONFIG_CHELSIO_T4_DCB */
488 /* Response queue handler for the FW event queue.
490 static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
491 const struct pkt_gl *gl)
493 u8 opcode = ((const struct rss_header *)rsp)->opcode;
495 rsp++; /* skip RSS header */
497 /* FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG.
499 if (unlikely(opcode == CPL_FW4_MSG &&
500 ((const struct cpl_fw4_msg *)rsp)->type == FW_TYPE_RSSCPL)) {
502 opcode = ((const struct rss_header *)rsp)->opcode;
504 if (opcode != CPL_SGE_EGR_UPDATE) {
505 dev_err(q->adap->pdev_dev, "unexpected FW4/CPL %#x on FW event queue\n"
511 if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
512 const struct cpl_sge_egr_update *p = (void *)rsp;
513 unsigned int qid = EGR_QID_G(ntohl(p->opcode_qid));
516 txq = q->adap->sge.egr_map[qid - q->adap->sge.egr_start];
518 if (txq->q_type == CXGB4_TXQ_ETH) {
519 struct sge_eth_txq *eq;
521 eq = container_of(txq, struct sge_eth_txq, q);
522 netif_tx_wake_queue(eq->txq);
524 struct sge_uld_txq *oq;
526 oq = container_of(txq, struct sge_uld_txq, q);
527 tasklet_schedule(&oq->qresume_tsk);
529 } else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
530 const struct cpl_fw6_msg *p = (void *)rsp;
532 #ifdef CONFIG_CHELSIO_T4_DCB
533 const struct fw_port_cmd *pcmd = (const void *)p->data;
534 unsigned int cmd = FW_CMD_OP_G(ntohl(pcmd->op_to_portid));
535 unsigned int action =
536 FW_PORT_CMD_ACTION_G(ntohl(pcmd->action_to_len16));
538 if (cmd == FW_PORT_CMD &&
539 (action == FW_PORT_ACTION_GET_PORT_INFO ||
540 action == FW_PORT_ACTION_GET_PORT_INFO32)) {
541 int port = FW_PORT_CMD_PORTID_G(
542 be32_to_cpu(pcmd->op_to_portid));
543 struct net_device *dev;
544 int dcbxdis, state_input;
546 dev = q->adap->port[q->adap->chan_map[port]];
547 dcbxdis = (action == FW_PORT_ACTION_GET_PORT_INFO
548 ? !!(pcmd->u.info.dcbxdis_pkd &
549 FW_PORT_CMD_DCBXDIS_F)
550 : !!(pcmd->u.info32.lstatus32_to_cbllen32 &
551 FW_PORT_CMD_DCBXDIS32_F));
552 state_input = (dcbxdis
553 ? CXGB4_DCB_INPUT_FW_DISABLED
554 : CXGB4_DCB_INPUT_FW_ENABLED);
556 cxgb4_dcb_state_fsm(dev, state_input);
559 if (cmd == FW_PORT_CMD &&
560 action == FW_PORT_ACTION_L2_DCB_CFG)
561 dcb_rpl(q->adap, pcmd);
565 t4_handle_fw_rpl(q->adap, p->data);
566 } else if (opcode == CPL_L2T_WRITE_RPL) {
567 const struct cpl_l2t_write_rpl *p = (void *)rsp;
569 do_l2t_write_rpl(q->adap, p);
570 } else if (opcode == CPL_SMT_WRITE_RPL) {
571 const struct cpl_smt_write_rpl *p = (void *)rsp;
573 do_smt_write_rpl(q->adap, p);
574 } else if (opcode == CPL_SET_TCB_RPL) {
575 const struct cpl_set_tcb_rpl *p = (void *)rsp;
577 filter_rpl(q->adap, p);
578 } else if (opcode == CPL_ACT_OPEN_RPL) {
579 const struct cpl_act_open_rpl *p = (void *)rsp;
581 hash_filter_rpl(q->adap, p);
582 } else if (opcode == CPL_ABORT_RPL_RSS) {
583 const struct cpl_abort_rpl_rss *p = (void *)rsp;
585 hash_del_filter_rpl(q->adap, p);
587 dev_err(q->adap->pdev_dev,
588 "unexpected CPL %#x on FW event queue\n", opcode);
593 static void disable_msi(struct adapter *adapter)
595 if (adapter->flags & USING_MSIX) {
596 pci_disable_msix(adapter->pdev);
597 adapter->flags &= ~USING_MSIX;
598 } else if (adapter->flags & USING_MSI) {
599 pci_disable_msi(adapter->pdev);
600 adapter->flags &= ~USING_MSI;
605 * Interrupt handler for non-data events used with MSI-X.
607 static irqreturn_t t4_nondata_intr(int irq, void *cookie)
609 struct adapter *adap = cookie;
610 u32 v = t4_read_reg(adap, MYPF_REG(PL_PF_INT_CAUSE_A));
614 t4_write_reg(adap, MYPF_REG(PL_PF_INT_CAUSE_A), v);
616 if (adap->flags & MASTER_PF)
617 t4_slow_intr_handler(adap);
622 * Name the MSI-X interrupts.
624 static void name_msix_vecs(struct adapter *adap)
626 int i, j, msi_idx = 2, n = sizeof(adap->msix_info[0].desc);
628 /* non-data interrupts */
629 snprintf(adap->msix_info[0].desc, n, "%s", adap->port[0]->name);
632 snprintf(adap->msix_info[1].desc, n, "%s-FWeventq",
633 adap->port[0]->name);
635 /* Ethernet queues */
636 for_each_port(adap, j) {
637 struct net_device *d = adap->port[j];
638 const struct port_info *pi = netdev_priv(d);
640 for (i = 0; i < pi->nqsets; i++, msi_idx++)
641 snprintf(adap->msix_info[msi_idx].desc, n, "%s-Rx%d",
646 static int request_msix_queue_irqs(struct adapter *adap)
648 struct sge *s = &adap->sge;
652 err = request_irq(adap->msix_info[1].vec, t4_sge_intr_msix, 0,
653 adap->msix_info[1].desc, &s->fw_evtq);
657 for_each_ethrxq(s, ethqidx) {
658 err = request_irq(adap->msix_info[msi_index].vec,
660 adap->msix_info[msi_index].desc,
661 &s->ethrxq[ethqidx].rspq);
669 while (--ethqidx >= 0)
670 free_irq(adap->msix_info[--msi_index].vec,
671 &s->ethrxq[ethqidx].rspq);
672 free_irq(adap->msix_info[1].vec, &s->fw_evtq);
676 static void free_msix_queue_irqs(struct adapter *adap)
678 int i, msi_index = 2;
679 struct sge *s = &adap->sge;
681 free_irq(adap->msix_info[1].vec, &s->fw_evtq);
682 for_each_ethrxq(s, i)
683 free_irq(adap->msix_info[msi_index++].vec, &s->ethrxq[i].rspq);
687 * cxgb4_write_rss - write the RSS table for a given port
689 * @queues: array of queue indices for RSS
691 * Sets up the portion of the HW RSS table for the port's VI to distribute
692 * packets to the Rx queues in @queues.
693 * Should never be called before setting up sge eth rx queues
695 int cxgb4_write_rss(const struct port_info *pi, const u16 *queues)
699 struct adapter *adapter = pi->adapter;
700 const struct sge_eth_rxq *rxq;
702 rxq = &adapter->sge.ethrxq[pi->first_qset];
703 rss = kmalloc(pi->rss_size * sizeof(u16), GFP_KERNEL);
707 /* map the queue indices to queue ids */
708 for (i = 0; i < pi->rss_size; i++, queues++)
709 rss[i] = rxq[*queues].rspq.abs_id;
711 err = t4_config_rss_range(adapter, adapter->pf, pi->viid, 0,
712 pi->rss_size, rss, pi->rss_size);
713 /* If Tunnel All Lookup isn't specified in the global RSS
714 * Configuration, then we need to specify a default Ingress
715 * Queue for any ingress packets which aren't hashed. We'll
716 * use our first ingress queue ...
719 err = t4_config_vi_rss(adapter, adapter->mbox, pi->viid,
720 FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN_F |
721 FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN_F |
722 FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN_F |
723 FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN_F |
724 FW_RSS_VI_CONFIG_CMD_UDPEN_F,
731 * setup_rss - configure RSS
734 * Sets up RSS for each port.
736 static int setup_rss(struct adapter *adap)
740 for_each_port(adap, i) {
741 const struct port_info *pi = adap2pinfo(adap, i);
743 /* Fill default values with equal distribution */
744 for (j = 0; j < pi->rss_size; j++)
745 pi->rss[j] = j % pi->nqsets;
747 err = cxgb4_write_rss(pi, pi->rss);
755 * Return the channel of the ingress queue with the given qid.
757 static unsigned int rxq_to_chan(const struct sge *p, unsigned int qid)
759 qid -= p->ingr_start;
760 return netdev2pinfo(p->ingr_map[qid]->netdev)->tx_chan;
764 * Wait until all NAPI handlers are descheduled.
766 static void quiesce_rx(struct adapter *adap)
770 for (i = 0; i < adap->sge.ingr_sz; i++) {
771 struct sge_rspq *q = adap->sge.ingr_map[i];
774 napi_disable(&q->napi);
778 /* Disable interrupt and napi handler */
779 static void disable_interrupts(struct adapter *adap)
781 if (adap->flags & FULL_INIT_DONE) {
782 t4_intr_disable(adap);
783 if (adap->flags & USING_MSIX) {
784 free_msix_queue_irqs(adap);
785 free_irq(adap->msix_info[0].vec, adap);
787 free_irq(adap->pdev->irq, adap);
794 * Enable NAPI scheduling and interrupt generation for all Rx queues.
796 static void enable_rx(struct adapter *adap)
800 for (i = 0; i < adap->sge.ingr_sz; i++) {
801 struct sge_rspq *q = adap->sge.ingr_map[i];
806 napi_enable(&q->napi);
808 /* 0-increment GTS to start the timer and enable interrupts */
809 t4_write_reg(adap, MYPF_REG(SGE_PF_GTS_A),
810 SEINTARM_V(q->intr_params) |
811 INGRESSQID_V(q->cntxt_id));
816 static int setup_fw_sge_queues(struct adapter *adap)
818 struct sge *s = &adap->sge;
821 bitmap_zero(s->starving_fl, s->egr_sz);
822 bitmap_zero(s->txq_maperr, s->egr_sz);
824 if (adap->flags & USING_MSIX)
825 adap->msi_idx = 1; /* vector 0 is for non-queue interrupts */
827 err = t4_sge_alloc_rxq(adap, &s->intrq, false, adap->port[0], 0,
828 NULL, NULL, NULL, -1);
831 adap->msi_idx = -((int)s->intrq.abs_id + 1);
834 err = t4_sge_alloc_rxq(adap, &s->fw_evtq, true, adap->port[0],
835 adap->msi_idx, NULL, fwevtq_handler, NULL, -1);
837 t4_free_sge_resources(adap);
842 * setup_sge_queues - configure SGE Tx/Rx/response queues
845 * Determines how many sets of SGE queues to use and initializes them.
846 * We support multiple queue sets per port if we have MSI-X, otherwise
847 * just one queue set per port.
849 static int setup_sge_queues(struct adapter *adap)
852 struct sge *s = &adap->sge;
853 struct sge_uld_rxq_info *rxq_info = NULL;
854 unsigned int cmplqid = 0;
857 rxq_info = s->uld_rxq_info[CXGB4_ULD_RDMA];
859 for_each_port(adap, i) {
860 struct net_device *dev = adap->port[i];
861 struct port_info *pi = netdev_priv(dev);
862 struct sge_eth_rxq *q = &s->ethrxq[pi->first_qset];
863 struct sge_eth_txq *t = &s->ethtxq[pi->first_qset];
865 for (j = 0; j < pi->nqsets; j++, q++) {
866 if (adap->msi_idx > 0)
868 err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev,
869 adap->msi_idx, &q->fl,
872 t4_get_tp_ch_map(adap,
877 memset(&q->stats, 0, sizeof(q->stats));
879 for (j = 0; j < pi->nqsets; j++, t++) {
880 err = t4_sge_alloc_eth_txq(adap, t, dev,
881 netdev_get_tx_queue(dev, j),
882 s->fw_evtq.cntxt_id);
888 for_each_port(adap, i) {
889 /* Note that cmplqid below is 0 if we don't
890 * have RDMA queues, and that's the right value.
893 cmplqid = rxq_info->uldrxq[i].rspq.cntxt_id;
895 err = t4_sge_alloc_ctrl_txq(adap, &s->ctrlq[i], adap->port[i],
896 s->fw_evtq.cntxt_id, cmplqid);
901 if (!is_t4(adap->params.chip)) {
902 err = t4_sge_alloc_eth_txq(adap, &s->ptptxq, adap->port[0],
903 netdev_get_tx_queue(adap->port[0], 0)
904 , s->fw_evtq.cntxt_id);
909 t4_write_reg(adap, is_t4(adap->params.chip) ?
910 MPS_TRC_RSS_CONTROL_A :
911 MPS_T5_TRC_RSS_CONTROL_A,
912 RSSCONTROL_V(netdev2pinfo(adap->port[0])->tx_chan) |
913 QUEUENUMBER_V(s->ethrxq[0].rspq.abs_id));
916 t4_free_sge_resources(adap);
920 static u16 cxgb_select_queue(struct net_device *dev, struct sk_buff *skb,
921 void *accel_priv, select_queue_fallback_t fallback)
925 #ifdef CONFIG_CHELSIO_T4_DCB
926 /* If a Data Center Bridging has been successfully negotiated on this
927 * link then we'll use the skb's priority to map it to a TX Queue.
928 * The skb's priority is determined via the VLAN Tag Priority Code
931 if (cxgb4_dcb_enabled(dev) && !is_kdump_kernel()) {
935 err = vlan_get_tag(skb, &vlan_tci);
939 "TX Packet without VLAN Tag on DCB Link\n");
942 txq = (vlan_tci & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
943 #ifdef CONFIG_CHELSIO_T4_FCOE
944 if (skb->protocol == htons(ETH_P_FCOE))
945 txq = skb->priority & 0x7;
946 #endif /* CONFIG_CHELSIO_T4_FCOE */
950 #endif /* CONFIG_CHELSIO_T4_DCB */
953 txq = (skb_rx_queue_recorded(skb)
954 ? skb_get_rx_queue(skb)
955 : smp_processor_id());
957 while (unlikely(txq >= dev->real_num_tx_queues))
958 txq -= dev->real_num_tx_queues;
963 return fallback(dev, skb) % dev->real_num_tx_queues;
966 static int closest_timer(const struct sge *s, int time)
968 int i, delta, match = 0, min_delta = INT_MAX;
970 for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
971 delta = time - s->timer_val[i];
974 if (delta < min_delta) {
982 static int closest_thres(const struct sge *s, int thres)
984 int i, delta, match = 0, min_delta = INT_MAX;
986 for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
987 delta = thres - s->counter_val[i];
990 if (delta < min_delta) {
999 * cxgb4_set_rspq_intr_params - set a queue's interrupt holdoff parameters
1001 * @us: the hold-off time in us, or 0 to disable timer
1002 * @cnt: the hold-off packet count, or 0 to disable counter
1004 * Sets an Rx queue's interrupt hold-off time and packet count. At least
1005 * one of the two needs to be enabled for the queue to generate interrupts.
1007 int cxgb4_set_rspq_intr_params(struct sge_rspq *q,
1008 unsigned int us, unsigned int cnt)
1010 struct adapter *adap = q->adap;
1012 if ((us | cnt) == 0)
1019 new_idx = closest_thres(&adap->sge, cnt);
1020 if (q->desc && q->pktcnt_idx != new_idx) {
1021 /* the queue has already been created, update it */
1022 v = FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DMAQ) |
1023 FW_PARAMS_PARAM_X_V(
1024 FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
1025 FW_PARAMS_PARAM_YZ_V(q->cntxt_id);
1026 err = t4_set_params(adap, adap->mbox, adap->pf, 0, 1,
1031 q->pktcnt_idx = new_idx;
1034 us = us == 0 ? 6 : closest_timer(&adap->sge, us);
1035 q->intr_params = QINTR_TIMER_IDX_V(us) | QINTR_CNT_EN_V(cnt > 0);
1039 static int cxgb_set_features(struct net_device *dev, netdev_features_t features)
1041 const struct port_info *pi = netdev_priv(dev);
1042 netdev_features_t changed = dev->features ^ features;
1045 if (!(changed & NETIF_F_HW_VLAN_CTAG_RX))
1048 err = t4_set_rxmode(pi->adapter, pi->adapter->pf, pi->viid, -1,
1050 !!(features & NETIF_F_HW_VLAN_CTAG_RX), true);
1052 dev->features = features ^ NETIF_F_HW_VLAN_CTAG_RX;
1056 static int setup_debugfs(struct adapter *adap)
1058 if (IS_ERR_OR_NULL(adap->debugfs_root))
1061 #ifdef CONFIG_DEBUG_FS
1062 t4_setup_debugfs(adap);
1068 * upper-layer driver support
1072 * Allocate an active-open TID and set it to the supplied value.
1074 int cxgb4_alloc_atid(struct tid_info *t, void *data)
1078 spin_lock_bh(&t->atid_lock);
1080 union aopen_entry *p = t->afree;
1082 atid = (p - t->atid_tab) + t->atid_base;
1087 spin_unlock_bh(&t->atid_lock);
1090 EXPORT_SYMBOL(cxgb4_alloc_atid);
1093 * Release an active-open TID.
1095 void cxgb4_free_atid(struct tid_info *t, unsigned int atid)
1097 union aopen_entry *p = &t->atid_tab[atid - t->atid_base];
1099 spin_lock_bh(&t->atid_lock);
1103 spin_unlock_bh(&t->atid_lock);
1105 EXPORT_SYMBOL(cxgb4_free_atid);
1108 * Allocate a server TID and set it to the supplied value.
1110 int cxgb4_alloc_stid(struct tid_info *t, int family, void *data)
1114 spin_lock_bh(&t->stid_lock);
1115 if (family == PF_INET) {
1116 stid = find_first_zero_bit(t->stid_bmap, t->nstids);
1117 if (stid < t->nstids)
1118 __set_bit(stid, t->stid_bmap);
1122 stid = bitmap_find_free_region(t->stid_bmap, t->nstids, 1);
1127 t->stid_tab[stid].data = data;
1128 stid += t->stid_base;
1129 /* IPv6 requires max of 520 bits or 16 cells in TCAM
1130 * This is equivalent to 4 TIDs. With CLIP enabled it
1133 if (family == PF_INET6) {
1134 t->stids_in_use += 2;
1135 t->v6_stids_in_use += 2;
1140 spin_unlock_bh(&t->stid_lock);
1143 EXPORT_SYMBOL(cxgb4_alloc_stid);
1145 /* Allocate a server filter TID and set it to the supplied value.
1147 int cxgb4_alloc_sftid(struct tid_info *t, int family, void *data)
1151 spin_lock_bh(&t->stid_lock);
1152 if (family == PF_INET) {
1153 stid = find_next_zero_bit(t->stid_bmap,
1154 t->nstids + t->nsftids, t->nstids);
1155 if (stid < (t->nstids + t->nsftids))
1156 __set_bit(stid, t->stid_bmap);
1163 t->stid_tab[stid].data = data;
1165 stid += t->sftid_base;
1168 spin_unlock_bh(&t->stid_lock);
1171 EXPORT_SYMBOL(cxgb4_alloc_sftid);
1173 /* Release a server TID.
1175 void cxgb4_free_stid(struct tid_info *t, unsigned int stid, int family)
1177 /* Is it a server filter TID? */
1178 if (t->nsftids && (stid >= t->sftid_base)) {
1179 stid -= t->sftid_base;
1182 stid -= t->stid_base;
1185 spin_lock_bh(&t->stid_lock);
1186 if (family == PF_INET)
1187 __clear_bit(stid, t->stid_bmap);
1189 bitmap_release_region(t->stid_bmap, stid, 1);
1190 t->stid_tab[stid].data = NULL;
1191 if (stid < t->nstids) {
1192 if (family == PF_INET6) {
1193 t->stids_in_use -= 2;
1194 t->v6_stids_in_use -= 2;
1202 spin_unlock_bh(&t->stid_lock);
1204 EXPORT_SYMBOL(cxgb4_free_stid);
1207 * Populate a TID_RELEASE WR. Caller must properly size the skb.
1209 static void mk_tid_release(struct sk_buff *skb, unsigned int chan,
1212 struct cpl_tid_release *req;
1214 set_wr_txq(skb, CPL_PRIORITY_SETUP, chan);
1215 req = __skb_put(skb, sizeof(*req));
1216 INIT_TP_WR(req, tid);
1217 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, tid));
1221 * Queue a TID release request and if necessary schedule a work queue to
1224 static void cxgb4_queue_tid_release(struct tid_info *t, unsigned int chan,
1227 void **p = &t->tid_tab[tid];
1228 struct adapter *adap = container_of(t, struct adapter, tids);
1230 spin_lock_bh(&adap->tid_release_lock);
1231 *p = adap->tid_release_head;
1232 /* Low 2 bits encode the Tx channel number */
1233 adap->tid_release_head = (void **)((uintptr_t)p | chan);
1234 if (!adap->tid_release_task_busy) {
1235 adap->tid_release_task_busy = true;
1236 queue_work(adap->workq, &adap->tid_release_task);
1238 spin_unlock_bh(&adap->tid_release_lock);
1242 * Process the list of pending TID release requests.
1244 static void process_tid_release_list(struct work_struct *work)
1246 struct sk_buff *skb;
1247 struct adapter *adap;
1249 adap = container_of(work, struct adapter, tid_release_task);
1251 spin_lock_bh(&adap->tid_release_lock);
1252 while (adap->tid_release_head) {
1253 void **p = adap->tid_release_head;
1254 unsigned int chan = (uintptr_t)p & 3;
1255 p = (void *)p - chan;
1257 adap->tid_release_head = *p;
1259 spin_unlock_bh(&adap->tid_release_lock);
1261 while (!(skb = alloc_skb(sizeof(struct cpl_tid_release),
1263 schedule_timeout_uninterruptible(1);
1265 mk_tid_release(skb, chan, p - adap->tids.tid_tab);
1266 t4_ofld_send(adap, skb);
1267 spin_lock_bh(&adap->tid_release_lock);
1269 adap->tid_release_task_busy = false;
1270 spin_unlock_bh(&adap->tid_release_lock);
1274 * Release a TID and inform HW. If we are unable to allocate the release
1275 * message we defer to a work queue.
1277 void cxgb4_remove_tid(struct tid_info *t, unsigned int chan, unsigned int tid,
1278 unsigned short family)
1280 struct sk_buff *skb;
1281 struct adapter *adap = container_of(t, struct adapter, tids);
1283 WARN_ON(tid >= t->ntids);
1285 if (t->tid_tab[tid]) {
1286 t->tid_tab[tid] = NULL;
1287 atomic_dec(&t->conns_in_use);
1288 if (t->hash_base && (tid >= t->hash_base)) {
1289 if (family == AF_INET6)
1290 atomic_sub(2, &t->hash_tids_in_use);
1292 atomic_dec(&t->hash_tids_in_use);
1294 if (family == AF_INET6)
1295 atomic_sub(2, &t->tids_in_use);
1297 atomic_dec(&t->tids_in_use);
1301 skb = alloc_skb(sizeof(struct cpl_tid_release), GFP_ATOMIC);
1303 mk_tid_release(skb, chan, tid);
1304 t4_ofld_send(adap, skb);
1306 cxgb4_queue_tid_release(t, chan, tid);
1308 EXPORT_SYMBOL(cxgb4_remove_tid);
1311 * Allocate and initialize the TID tables. Returns 0 on success.
1313 static int tid_init(struct tid_info *t)
1315 struct adapter *adap = container_of(t, struct adapter, tids);
1316 unsigned int max_ftids = t->nftids + t->nsftids;
1317 unsigned int natids = t->natids;
1318 unsigned int stid_bmap_size;
1319 unsigned int ftid_bmap_size;
1322 stid_bmap_size = BITS_TO_LONGS(t->nstids + t->nsftids);
1323 ftid_bmap_size = BITS_TO_LONGS(t->nftids);
1324 size = t->ntids * sizeof(*t->tid_tab) +
1325 natids * sizeof(*t->atid_tab) +
1326 t->nstids * sizeof(*t->stid_tab) +
1327 t->nsftids * sizeof(*t->stid_tab) +
1328 stid_bmap_size * sizeof(long) +
1329 max_ftids * sizeof(*t->ftid_tab) +
1330 ftid_bmap_size * sizeof(long);
1332 t->tid_tab = kvzalloc(size, GFP_KERNEL);
1336 t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
1337 t->stid_tab = (struct serv_entry *)&t->atid_tab[natids];
1338 t->stid_bmap = (unsigned long *)&t->stid_tab[t->nstids + t->nsftids];
1339 t->ftid_tab = (struct filter_entry *)&t->stid_bmap[stid_bmap_size];
1340 t->ftid_bmap = (unsigned long *)&t->ftid_tab[max_ftids];
1341 spin_lock_init(&t->stid_lock);
1342 spin_lock_init(&t->atid_lock);
1343 spin_lock_init(&t->ftid_lock);
1345 t->stids_in_use = 0;
1346 t->v6_stids_in_use = 0;
1347 t->sftids_in_use = 0;
1349 t->atids_in_use = 0;
1350 atomic_set(&t->tids_in_use, 0);
1351 atomic_set(&t->conns_in_use, 0);
1352 atomic_set(&t->hash_tids_in_use, 0);
1354 /* Setup the free list for atid_tab and clear the stid bitmap. */
1357 t->atid_tab[natids - 1].next = &t->atid_tab[natids];
1358 t->afree = t->atid_tab;
1361 if (is_offload(adap)) {
1362 bitmap_zero(t->stid_bmap, t->nstids + t->nsftids);
1363 /* Reserve stid 0 for T4/T5 adapters */
1364 if (!t->stid_base &&
1365 CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5)
1366 __set_bit(0, t->stid_bmap);
1369 bitmap_zero(t->ftid_bmap, t->nftids);
1374 * cxgb4_create_server - create an IP server
1376 * @stid: the server TID
1377 * @sip: local IP address to bind server to
1378 * @sport: the server's TCP port
1379 * @queue: queue to direct messages from this server to
1381 * Create an IP server for the given port and address.
1382 * Returns <0 on error and one of the %NET_XMIT_* values on success.
1384 int cxgb4_create_server(const struct net_device *dev, unsigned int stid,
1385 __be32 sip, __be16 sport, __be16 vlan,
1389 struct sk_buff *skb;
1390 struct adapter *adap;
1391 struct cpl_pass_open_req *req;
1394 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
1398 adap = netdev2adap(dev);
1399 req = __skb_put(skb, sizeof(*req));
1401 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, stid));
1402 req->local_port = sport;
1403 req->peer_port = htons(0);
1404 req->local_ip = sip;
1405 req->peer_ip = htonl(0);
1406 chan = rxq_to_chan(&adap->sge, queue);
1407 req->opt0 = cpu_to_be64(TX_CHAN_V(chan));
1408 req->opt1 = cpu_to_be64(CONN_POLICY_V(CPL_CONN_POLICY_ASK) |
1409 SYN_RSS_ENABLE_F | SYN_RSS_QUEUE_V(queue));
1410 ret = t4_mgmt_tx(adap, skb);
1411 return net_xmit_eval(ret);
1413 EXPORT_SYMBOL(cxgb4_create_server);
1415 /* cxgb4_create_server6 - create an IPv6 server
1417 * @stid: the server TID
1418 * @sip: local IPv6 address to bind server to
1419 * @sport: the server's TCP port
1420 * @queue: queue to direct messages from this server to
1422 * Create an IPv6 server for the given port and address.
1423 * Returns <0 on error and one of the %NET_XMIT_* values on success.
1425 int cxgb4_create_server6(const struct net_device *dev, unsigned int stid,
1426 const struct in6_addr *sip, __be16 sport,
1430 struct sk_buff *skb;
1431 struct adapter *adap;
1432 struct cpl_pass_open_req6 *req;
1435 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
1439 adap = netdev2adap(dev);
1440 req = __skb_put(skb, sizeof(*req));
1442 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ6, stid));
1443 req->local_port = sport;
1444 req->peer_port = htons(0);
1445 req->local_ip_hi = *(__be64 *)(sip->s6_addr);
1446 req->local_ip_lo = *(__be64 *)(sip->s6_addr + 8);
1447 req->peer_ip_hi = cpu_to_be64(0);
1448 req->peer_ip_lo = cpu_to_be64(0);
1449 chan = rxq_to_chan(&adap->sge, queue);
1450 req->opt0 = cpu_to_be64(TX_CHAN_V(chan));
1451 req->opt1 = cpu_to_be64(CONN_POLICY_V(CPL_CONN_POLICY_ASK) |
1452 SYN_RSS_ENABLE_F | SYN_RSS_QUEUE_V(queue));
1453 ret = t4_mgmt_tx(adap, skb);
1454 return net_xmit_eval(ret);
1456 EXPORT_SYMBOL(cxgb4_create_server6);
1458 int cxgb4_remove_server(const struct net_device *dev, unsigned int stid,
1459 unsigned int queue, bool ipv6)
1461 struct sk_buff *skb;
1462 struct adapter *adap;
1463 struct cpl_close_listsvr_req *req;
1466 adap = netdev2adap(dev);
1468 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
1472 req = __skb_put(skb, sizeof(*req));
1474 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, stid));
1475 req->reply_ctrl = htons(NO_REPLY_V(0) | (ipv6 ? LISTSVR_IPV6_V(1) :
1476 LISTSVR_IPV6_V(0)) | QUEUENO_V(queue));
1477 ret = t4_mgmt_tx(adap, skb);
1478 return net_xmit_eval(ret);
1480 EXPORT_SYMBOL(cxgb4_remove_server);
1483 * cxgb4_best_mtu - find the entry in the MTU table closest to an MTU
1484 * @mtus: the HW MTU table
1485 * @mtu: the target MTU
1486 * @idx: index of selected entry in the MTU table
1488 * Returns the index and the value in the HW MTU table that is closest to
1489 * but does not exceed @mtu, unless @mtu is smaller than any value in the
1490 * table, in which case that smallest available value is selected.
1492 unsigned int cxgb4_best_mtu(const unsigned short *mtus, unsigned short mtu,
1497 while (i < NMTUS - 1 && mtus[i + 1] <= mtu)
1503 EXPORT_SYMBOL(cxgb4_best_mtu);
1506 * cxgb4_best_aligned_mtu - find best MTU, [hopefully] data size aligned
1507 * @mtus: the HW MTU table
1508 * @header_size: Header Size
1509 * @data_size_max: maximum Data Segment Size
1510 * @data_size_align: desired Data Segment Size Alignment (2^N)
1511 * @mtu_idxp: HW MTU Table Index return value pointer (possibly NULL)
1513 * Similar to cxgb4_best_mtu() but instead of searching the Hardware
1514 * MTU Table based solely on a Maximum MTU parameter, we break that
1515 * parameter up into a Header Size and Maximum Data Segment Size, and
1516 * provide a desired Data Segment Size Alignment. If we find an MTU in
1517 * the Hardware MTU Table which will result in a Data Segment Size with
1518 * the requested alignment _and_ that MTU isn't "too far" from the
1519 * closest MTU, then we'll return that rather than the closest MTU.
1521 unsigned int cxgb4_best_aligned_mtu(const unsigned short *mtus,
1522 unsigned short header_size,
1523 unsigned short data_size_max,
1524 unsigned short data_size_align,
1525 unsigned int *mtu_idxp)
1527 unsigned short max_mtu = header_size + data_size_max;
1528 unsigned short data_size_align_mask = data_size_align - 1;
1529 int mtu_idx, aligned_mtu_idx;
1531 /* Scan the MTU Table till we find an MTU which is larger than our
1532 * Maximum MTU or we reach the end of the table. Along the way,
1533 * record the last MTU found, if any, which will result in a Data
1534 * Segment Length matching the requested alignment.
1536 for (mtu_idx = 0, aligned_mtu_idx = -1; mtu_idx < NMTUS; mtu_idx++) {
1537 unsigned short data_size = mtus[mtu_idx] - header_size;
1539 /* If this MTU minus the Header Size would result in a
1540 * Data Segment Size of the desired alignment, remember it.
1542 if ((data_size & data_size_align_mask) == 0)
1543 aligned_mtu_idx = mtu_idx;
1545 /* If we're not at the end of the Hardware MTU Table and the
1546 * next element is larger than our Maximum MTU, drop out of
1549 if (mtu_idx+1 < NMTUS && mtus[mtu_idx+1] > max_mtu)
1553 /* If we fell out of the loop because we ran to the end of the table,
1554 * then we just have to use the last [largest] entry.
1556 if (mtu_idx == NMTUS)
1559 /* If we found an MTU which resulted in the requested Data Segment
1560 * Length alignment and that's "not far" from the largest MTU which is
1561 * less than or equal to the maximum MTU, then use that.
1563 if (aligned_mtu_idx >= 0 &&
1564 mtu_idx - aligned_mtu_idx <= 1)
1565 mtu_idx = aligned_mtu_idx;
1567 /* If the caller has passed in an MTU Index pointer, pass the
1568 * MTU Index back. Return the MTU value.
1571 *mtu_idxp = mtu_idx;
1572 return mtus[mtu_idx];
1574 EXPORT_SYMBOL(cxgb4_best_aligned_mtu);
1577 * cxgb4_tp_smt_idx - Get the Source Mac Table index for this VI
1579 * @viid: VI id of the given port
1581 * Return the SMT index for this VI.
1583 unsigned int cxgb4_tp_smt_idx(enum chip_type chip, unsigned int viid)
1585 /* In T4/T5, SMT contains 256 SMAC entries organized in
1586 * 128 rows of 2 entries each.
1587 * In T6, SMT contains 256 SMAC entries in 256 rows.
1588 * TODO: The below code needs to be updated when we add support
1591 if (CHELSIO_CHIP_VERSION(chip) <= CHELSIO_T5)
1592 return ((viid & 0x7f) << 1);
1594 return (viid & 0x7f);
1596 EXPORT_SYMBOL(cxgb4_tp_smt_idx);
1599 * cxgb4_port_chan - get the HW channel of a port
1600 * @dev: the net device for the port
1602 * Return the HW Tx channel of the given port.
1604 unsigned int cxgb4_port_chan(const struct net_device *dev)
1606 return netdev2pinfo(dev)->tx_chan;
1608 EXPORT_SYMBOL(cxgb4_port_chan);
1610 unsigned int cxgb4_dbfifo_count(const struct net_device *dev, int lpfifo)
1612 struct adapter *adap = netdev2adap(dev);
1613 u32 v1, v2, lp_count, hp_count;
1615 v1 = t4_read_reg(adap, SGE_DBFIFO_STATUS_A);
1616 v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2_A);
1617 if (is_t4(adap->params.chip)) {
1618 lp_count = LP_COUNT_G(v1);
1619 hp_count = HP_COUNT_G(v1);
1621 lp_count = LP_COUNT_T5_G(v1);
1622 hp_count = HP_COUNT_T5_G(v2);
1624 return lpfifo ? lp_count : hp_count;
1626 EXPORT_SYMBOL(cxgb4_dbfifo_count);
1629 * cxgb4_port_viid - get the VI id of a port
1630 * @dev: the net device for the port
1632 * Return the VI id of the given port.
1634 unsigned int cxgb4_port_viid(const struct net_device *dev)
1636 return netdev2pinfo(dev)->viid;
1638 EXPORT_SYMBOL(cxgb4_port_viid);
1641 * cxgb4_port_idx - get the index of a port
1642 * @dev: the net device for the port
1644 * Return the index of the given port.
1646 unsigned int cxgb4_port_idx(const struct net_device *dev)
1648 return netdev2pinfo(dev)->port_id;
1650 EXPORT_SYMBOL(cxgb4_port_idx);
1652 void cxgb4_get_tcp_stats(struct pci_dev *pdev, struct tp_tcp_stats *v4,
1653 struct tp_tcp_stats *v6)
1655 struct adapter *adap = pci_get_drvdata(pdev);
1657 spin_lock(&adap->stats_lock);
1658 t4_tp_get_tcp_stats(adap, v4, v6, false);
1659 spin_unlock(&adap->stats_lock);
1661 EXPORT_SYMBOL(cxgb4_get_tcp_stats);
1663 void cxgb4_iscsi_init(struct net_device *dev, unsigned int tag_mask,
1664 const unsigned int *pgsz_order)
1666 struct adapter *adap = netdev2adap(dev);
1668 t4_write_reg(adap, ULP_RX_ISCSI_TAGMASK_A, tag_mask);
1669 t4_write_reg(adap, ULP_RX_ISCSI_PSZ_A, HPZ0_V(pgsz_order[0]) |
1670 HPZ1_V(pgsz_order[1]) | HPZ2_V(pgsz_order[2]) |
1671 HPZ3_V(pgsz_order[3]));
1673 EXPORT_SYMBOL(cxgb4_iscsi_init);
1675 int cxgb4_flush_eq_cache(struct net_device *dev)
1677 struct adapter *adap = netdev2adap(dev);
1679 return t4_sge_ctxt_flush(adap, adap->mbox, CTXT_EGRESS);
1681 EXPORT_SYMBOL(cxgb4_flush_eq_cache);
1683 static int read_eq_indices(struct adapter *adap, u16 qid, u16 *pidx, u16 *cidx)
1685 u32 addr = t4_read_reg(adap, SGE_DBQ_CTXT_BADDR_A) + 24 * qid + 8;
1689 spin_lock(&adap->win0_lock);
1690 ret = t4_memory_rw(adap, 0, MEM_EDC0, addr,
1691 sizeof(indices), (__be32 *)&indices,
1693 spin_unlock(&adap->win0_lock);
1695 *cidx = (be64_to_cpu(indices) >> 25) & 0xffff;
1696 *pidx = (be64_to_cpu(indices) >> 9) & 0xffff;
1701 int cxgb4_sync_txq_pidx(struct net_device *dev, u16 qid, u16 pidx,
1704 struct adapter *adap = netdev2adap(dev);
1705 u16 hw_pidx, hw_cidx;
1708 ret = read_eq_indices(adap, qid, &hw_pidx, &hw_cidx);
1712 if (pidx != hw_pidx) {
1716 if (pidx >= hw_pidx)
1717 delta = pidx - hw_pidx;
1719 delta = size - hw_pidx + pidx;
1721 if (is_t4(adap->params.chip))
1722 val = PIDX_V(delta);
1724 val = PIDX_T5_V(delta);
1726 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
1732 EXPORT_SYMBOL(cxgb4_sync_txq_pidx);
1734 int cxgb4_read_tpte(struct net_device *dev, u32 stag, __be32 *tpte)
1736 struct adapter *adap;
1737 u32 offset, memtype, memaddr;
1738 u32 edc0_size, edc1_size, mc0_size, mc1_size, size;
1739 u32 edc0_end, edc1_end, mc0_end, mc1_end;
1742 adap = netdev2adap(dev);
1744 offset = ((stag >> 8) * 32) + adap->vres.stag.start;
1746 /* Figure out where the offset lands in the Memory Type/Address scheme.
1747 * This code assumes that the memory is laid out starting at offset 0
1748 * with no breaks as: EDC0, EDC1, MC0, MC1. All cards have both EDC0
1749 * and EDC1. Some cards will have neither MC0 nor MC1, most cards have
1750 * MC0, and some have both MC0 and MC1.
1752 size = t4_read_reg(adap, MA_EDRAM0_BAR_A);
1753 edc0_size = EDRAM0_SIZE_G(size) << 20;
1754 size = t4_read_reg(adap, MA_EDRAM1_BAR_A);
1755 edc1_size = EDRAM1_SIZE_G(size) << 20;
1756 size = t4_read_reg(adap, MA_EXT_MEMORY0_BAR_A);
1757 mc0_size = EXT_MEM0_SIZE_G(size) << 20;
1759 edc0_end = edc0_size;
1760 edc1_end = edc0_end + edc1_size;
1761 mc0_end = edc1_end + mc0_size;
1763 if (offset < edc0_end) {
1766 } else if (offset < edc1_end) {
1768 memaddr = offset - edc0_end;
1770 if (offset < mc0_end) {
1772 memaddr = offset - edc1_end;
1773 } else if (is_t5(adap->params.chip)) {
1774 size = t4_read_reg(adap, MA_EXT_MEMORY1_BAR_A);
1775 mc1_size = EXT_MEM1_SIZE_G(size) << 20;
1776 mc1_end = mc0_end + mc1_size;
1777 if (offset < mc1_end) {
1779 memaddr = offset - mc0_end;
1781 /* offset beyond the end of any memory */
1785 /* T4/T6 only has a single memory channel */
1790 spin_lock(&adap->win0_lock);
1791 ret = t4_memory_rw(adap, 0, memtype, memaddr, 32, tpte, T4_MEMORY_READ);
1792 spin_unlock(&adap->win0_lock);
1796 dev_err(adap->pdev_dev, "stag %#x, offset %#x out of range\n",
1800 EXPORT_SYMBOL(cxgb4_read_tpte);
1802 u64 cxgb4_read_sge_timestamp(struct net_device *dev)
1805 struct adapter *adap;
1807 adap = netdev2adap(dev);
1808 lo = t4_read_reg(adap, SGE_TIMESTAMP_LO_A);
1809 hi = TSVAL_G(t4_read_reg(adap, SGE_TIMESTAMP_HI_A));
1811 return ((u64)hi << 32) | (u64)lo;
1813 EXPORT_SYMBOL(cxgb4_read_sge_timestamp);
1815 int cxgb4_bar2_sge_qregs(struct net_device *dev,
1817 enum cxgb4_bar2_qtype qtype,
1820 unsigned int *pbar2_qid)
1822 return t4_bar2_sge_qregs(netdev2adap(dev),
1824 (qtype == CXGB4_BAR2_QTYPE_EGRESS
1825 ? T4_BAR2_QTYPE_EGRESS
1826 : T4_BAR2_QTYPE_INGRESS),
1831 EXPORT_SYMBOL(cxgb4_bar2_sge_qregs);
1833 static struct pci_driver cxgb4_driver;
1835 static void check_neigh_update(struct neighbour *neigh)
1837 const struct device *parent;
1838 const struct net_device *netdev = neigh->dev;
1840 if (is_vlan_dev(netdev))
1841 netdev = vlan_dev_real_dev(netdev);
1842 parent = netdev->dev.parent;
1843 if (parent && parent->driver == &cxgb4_driver.driver)
1844 t4_l2t_update(dev_get_drvdata(parent), neigh);
1847 static int netevent_cb(struct notifier_block *nb, unsigned long event,
1851 case NETEVENT_NEIGH_UPDATE:
1852 check_neigh_update(data);
1854 case NETEVENT_REDIRECT:
1861 static bool netevent_registered;
1862 static struct notifier_block cxgb4_netevent_nb = {
1863 .notifier_call = netevent_cb
1866 static void drain_db_fifo(struct adapter *adap, int usecs)
1868 u32 v1, v2, lp_count, hp_count;
1871 v1 = t4_read_reg(adap, SGE_DBFIFO_STATUS_A);
1872 v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2_A);
1873 if (is_t4(adap->params.chip)) {
1874 lp_count = LP_COUNT_G(v1);
1875 hp_count = HP_COUNT_G(v1);
1877 lp_count = LP_COUNT_T5_G(v1);
1878 hp_count = HP_COUNT_T5_G(v2);
1881 if (lp_count == 0 && hp_count == 0)
1883 set_current_state(TASK_UNINTERRUPTIBLE);
1884 schedule_timeout(usecs_to_jiffies(usecs));
1888 static void disable_txq_db(struct sge_txq *q)
1890 unsigned long flags;
1892 spin_lock_irqsave(&q->db_lock, flags);
1894 spin_unlock_irqrestore(&q->db_lock, flags);
1897 static void enable_txq_db(struct adapter *adap, struct sge_txq *q)
1899 spin_lock_irq(&q->db_lock);
1900 if (q->db_pidx_inc) {
1901 /* Make sure that all writes to the TX descriptors
1902 * are committed before we tell HW about them.
1905 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
1906 QID_V(q->cntxt_id) | PIDX_V(q->db_pidx_inc));
1910 spin_unlock_irq(&q->db_lock);
1913 static void disable_dbs(struct adapter *adap)
1917 for_each_ethrxq(&adap->sge, i)
1918 disable_txq_db(&adap->sge.ethtxq[i].q);
1919 if (is_offload(adap)) {
1920 struct sge_uld_txq_info *txq_info =
1921 adap->sge.uld_txq_info[CXGB4_TX_OFLD];
1924 for_each_ofldtxq(&adap->sge, i) {
1925 struct sge_uld_txq *txq = &txq_info->uldtxq[i];
1927 disable_txq_db(&txq->q);
1931 for_each_port(adap, i)
1932 disable_txq_db(&adap->sge.ctrlq[i].q);
1935 static void enable_dbs(struct adapter *adap)
1939 for_each_ethrxq(&adap->sge, i)
1940 enable_txq_db(adap, &adap->sge.ethtxq[i].q);
1941 if (is_offload(adap)) {
1942 struct sge_uld_txq_info *txq_info =
1943 adap->sge.uld_txq_info[CXGB4_TX_OFLD];
1946 for_each_ofldtxq(&adap->sge, i) {
1947 struct sge_uld_txq *txq = &txq_info->uldtxq[i];
1949 enable_txq_db(adap, &txq->q);
1953 for_each_port(adap, i)
1954 enable_txq_db(adap, &adap->sge.ctrlq[i].q);
1957 static void notify_rdma_uld(struct adapter *adap, enum cxgb4_control cmd)
1959 enum cxgb4_uld type = CXGB4_ULD_RDMA;
1961 if (adap->uld && adap->uld[type].handle)
1962 adap->uld[type].control(adap->uld[type].handle, cmd);
1965 static void process_db_full(struct work_struct *work)
1967 struct adapter *adap;
1969 adap = container_of(work, struct adapter, db_full_task);
1971 drain_db_fifo(adap, dbfifo_drain_delay);
1973 notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
1974 if (CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5)
1975 t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
1976 DBFIFO_HP_INT_F | DBFIFO_LP_INT_F,
1977 DBFIFO_HP_INT_F | DBFIFO_LP_INT_F);
1979 t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
1980 DBFIFO_LP_INT_F, DBFIFO_LP_INT_F);
1983 static void sync_txq_pidx(struct adapter *adap, struct sge_txq *q)
1985 u16 hw_pidx, hw_cidx;
1988 spin_lock_irq(&q->db_lock);
1989 ret = read_eq_indices(adap, (u16)q->cntxt_id, &hw_pidx, &hw_cidx);
1992 if (q->db_pidx != hw_pidx) {
1996 if (q->db_pidx >= hw_pidx)
1997 delta = q->db_pidx - hw_pidx;
1999 delta = q->size - hw_pidx + q->db_pidx;
2001 if (is_t4(adap->params.chip))
2002 val = PIDX_V(delta);
2004 val = PIDX_T5_V(delta);
2006 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
2007 QID_V(q->cntxt_id) | val);
2012 spin_unlock_irq(&q->db_lock);
2014 CH_WARN(adap, "DB drop recovery failed.\n");
2017 static void recover_all_queues(struct adapter *adap)
2021 for_each_ethrxq(&adap->sge, i)
2022 sync_txq_pidx(adap, &adap->sge.ethtxq[i].q);
2023 if (is_offload(adap)) {
2024 struct sge_uld_txq_info *txq_info =
2025 adap->sge.uld_txq_info[CXGB4_TX_OFLD];
2027 for_each_ofldtxq(&adap->sge, i) {
2028 struct sge_uld_txq *txq = &txq_info->uldtxq[i];
2030 sync_txq_pidx(adap, &txq->q);
2034 for_each_port(adap, i)
2035 sync_txq_pidx(adap, &adap->sge.ctrlq[i].q);
2038 static void process_db_drop(struct work_struct *work)
2040 struct adapter *adap;
2042 adap = container_of(work, struct adapter, db_drop_task);
2044 if (is_t4(adap->params.chip)) {
2045 drain_db_fifo(adap, dbfifo_drain_delay);
2046 notify_rdma_uld(adap, CXGB4_CONTROL_DB_DROP);
2047 drain_db_fifo(adap, dbfifo_drain_delay);
2048 recover_all_queues(adap);
2049 drain_db_fifo(adap, dbfifo_drain_delay);
2051 notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
2052 } else if (is_t5(adap->params.chip)) {
2053 u32 dropped_db = t4_read_reg(adap, 0x010ac);
2054 u16 qid = (dropped_db >> 15) & 0x1ffff;
2055 u16 pidx_inc = dropped_db & 0x1fff;
2057 unsigned int bar2_qid;
2060 ret = t4_bar2_sge_qregs(adap, qid, T4_BAR2_QTYPE_EGRESS,
2061 0, &bar2_qoffset, &bar2_qid);
2063 dev_err(adap->pdev_dev, "doorbell drop recovery: "
2064 "qid=%d, pidx_inc=%d\n", qid, pidx_inc);
2066 writel(PIDX_T5_V(pidx_inc) | QID_V(bar2_qid),
2067 adap->bar2 + bar2_qoffset + SGE_UDB_KDOORBELL);
2069 /* Re-enable BAR2 WC */
2070 t4_set_reg_field(adap, 0x10b0, 1<<15, 1<<15);
2073 if (CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5)
2074 t4_set_reg_field(adap, SGE_DOORBELL_CONTROL_A, DROPPED_DB_F, 0);
2077 void t4_db_full(struct adapter *adap)
2079 if (is_t4(adap->params.chip)) {
2081 notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
2082 t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
2083 DBFIFO_HP_INT_F | DBFIFO_LP_INT_F, 0);
2084 queue_work(adap->workq, &adap->db_full_task);
2088 void t4_db_dropped(struct adapter *adap)
2090 if (is_t4(adap->params.chip)) {
2092 notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
2094 queue_work(adap->workq, &adap->db_drop_task);
2097 void t4_register_netevent_notifier(void)
2099 if (!netevent_registered) {
2100 register_netevent_notifier(&cxgb4_netevent_nb);
2101 netevent_registered = true;
2105 static void detach_ulds(struct adapter *adap)
2109 mutex_lock(&uld_mutex);
2110 list_del(&adap->list_node);
2112 for (i = 0; i < CXGB4_ULD_MAX; i++)
2113 if (adap->uld && adap->uld[i].handle)
2114 adap->uld[i].state_change(adap->uld[i].handle,
2115 CXGB4_STATE_DETACH);
2117 if (netevent_registered && list_empty(&adapter_list)) {
2118 unregister_netevent_notifier(&cxgb4_netevent_nb);
2119 netevent_registered = false;
2121 mutex_unlock(&uld_mutex);
2124 static void notify_ulds(struct adapter *adap, enum cxgb4_state new_state)
2128 mutex_lock(&uld_mutex);
2129 for (i = 0; i < CXGB4_ULD_MAX; i++)
2130 if (adap->uld && adap->uld[i].handle)
2131 adap->uld[i].state_change(adap->uld[i].handle,
2133 mutex_unlock(&uld_mutex);
2136 #if IS_ENABLED(CONFIG_IPV6)
2137 static int cxgb4_inet6addr_handler(struct notifier_block *this,
2138 unsigned long event, void *data)
2140 struct inet6_ifaddr *ifa = data;
2141 struct net_device *event_dev = ifa->idev->dev;
2142 const struct device *parent = NULL;
2143 #if IS_ENABLED(CONFIG_BONDING)
2144 struct adapter *adap;
2146 if (is_vlan_dev(event_dev))
2147 event_dev = vlan_dev_real_dev(event_dev);
2148 #if IS_ENABLED(CONFIG_BONDING)
2149 if (event_dev->flags & IFF_MASTER) {
2150 list_for_each_entry(adap, &adapter_list, list_node) {
2153 cxgb4_clip_get(adap->port[0],
2154 (const u32 *)ifa, 1);
2157 cxgb4_clip_release(adap->port[0],
2158 (const u32 *)ifa, 1);
2169 parent = event_dev->dev.parent;
2171 if (parent && parent->driver == &cxgb4_driver.driver) {
2174 cxgb4_clip_get(event_dev, (const u32 *)ifa, 1);
2177 cxgb4_clip_release(event_dev, (const u32 *)ifa, 1);
2186 static bool inet6addr_registered;
2187 static struct notifier_block cxgb4_inet6addr_notifier = {
2188 .notifier_call = cxgb4_inet6addr_handler
2191 static void update_clip(const struct adapter *adap)
2194 struct net_device *dev;
2199 for (i = 0; i < MAX_NPORTS; i++) {
2200 dev = adap->port[i];
2204 ret = cxgb4_update_root_dev_clip(dev);
2211 #endif /* IS_ENABLED(CONFIG_IPV6) */
2214 * cxgb_up - enable the adapter
2215 * @adap: adapter being enabled
2217 * Called when the first port is enabled, this function performs the
2218 * actions necessary to make an adapter operational, such as completing
2219 * the initialization of HW modules, and enabling interrupts.
2221 * Must be called with the rtnl lock held.
2223 static int cxgb_up(struct adapter *adap)
2227 mutex_lock(&uld_mutex);
2228 err = setup_sge_queues(adap);
2231 err = setup_rss(adap);
2235 if (adap->flags & USING_MSIX) {
2236 name_msix_vecs(adap);
2237 err = request_irq(adap->msix_info[0].vec, t4_nondata_intr, 0,
2238 adap->msix_info[0].desc, adap);
2241 err = request_msix_queue_irqs(adap);
2243 free_irq(adap->msix_info[0].vec, adap);
2247 err = request_irq(adap->pdev->irq, t4_intr_handler(adap),
2248 (adap->flags & USING_MSI) ? 0 : IRQF_SHARED,
2249 adap->port[0]->name, adap);
2256 t4_intr_enable(adap);
2257 adap->flags |= FULL_INIT_DONE;
2258 mutex_unlock(&uld_mutex);
2260 notify_ulds(adap, CXGB4_STATE_UP);
2261 #if IS_ENABLED(CONFIG_IPV6)
2264 /* Initialize hash mac addr list*/
2265 INIT_LIST_HEAD(&adap->mac_hlist);
2269 dev_err(adap->pdev_dev, "request_irq failed, err %d\n", err);
2271 t4_free_sge_resources(adap);
2273 mutex_unlock(&uld_mutex);
2277 static void cxgb_down(struct adapter *adapter)
2279 cancel_work_sync(&adapter->tid_release_task);
2280 cancel_work_sync(&adapter->db_full_task);
2281 cancel_work_sync(&adapter->db_drop_task);
2282 adapter->tid_release_task_busy = false;
2283 adapter->tid_release_head = NULL;
2285 t4_sge_stop(adapter);
2286 t4_free_sge_resources(adapter);
2287 adapter->flags &= ~FULL_INIT_DONE;
2291 * net_device operations
2293 static int cxgb_open(struct net_device *dev)
2296 struct port_info *pi = netdev_priv(dev);
2297 struct adapter *adapter = pi->adapter;
2299 netif_carrier_off(dev);
2301 if (!(adapter->flags & FULL_INIT_DONE)) {
2302 err = cxgb_up(adapter);
2307 /* It's possible that the basic port information could have
2308 * changed since we first read it.
2310 err = t4_update_port_info(pi);
2314 err = link_start(dev);
2316 netif_tx_start_all_queues(dev);
2320 static int cxgb_close(struct net_device *dev)
2322 struct port_info *pi = netdev_priv(dev);
2323 struct adapter *adapter = pi->adapter;
2326 netif_tx_stop_all_queues(dev);
2327 netif_carrier_off(dev);
2328 ret = t4_enable_vi(adapter, adapter->pf, pi->viid, false, false);
2329 #ifdef CONFIG_CHELSIO_T4_DCB
2330 cxgb4_dcb_reset(dev);
2331 dcb_tx_queue_prio_enable(dev, false);
2336 int cxgb4_create_server_filter(const struct net_device *dev, unsigned int stid,
2337 __be32 sip, __be16 sport, __be16 vlan,
2338 unsigned int queue, unsigned char port, unsigned char mask)
2341 struct filter_entry *f;
2342 struct adapter *adap;
2346 adap = netdev2adap(dev);
2348 /* Adjust stid to correct filter index */
2349 stid -= adap->tids.sftid_base;
2350 stid += adap->tids.nftids;
2352 /* Check to make sure the filter requested is writable ...
2354 f = &adap->tids.ftid_tab[stid];
2355 ret = writable_filter(f);
2359 /* Clear out any old resources being used by the filter before
2360 * we start constructing the new filter.
2363 clear_filter(adap, f);
2365 /* Clear out filter specifications */
2366 memset(&f->fs, 0, sizeof(struct ch_filter_specification));
2367 f->fs.val.lport = cpu_to_be16(sport);
2368 f->fs.mask.lport = ~0;
2370 if ((val[0] | val[1] | val[2] | val[3]) != 0) {
2371 for (i = 0; i < 4; i++) {
2372 f->fs.val.lip[i] = val[i];
2373 f->fs.mask.lip[i] = ~0;
2375 if (adap->params.tp.vlan_pri_map & PORT_F) {
2376 f->fs.val.iport = port;
2377 f->fs.mask.iport = mask;
2381 if (adap->params.tp.vlan_pri_map & PROTOCOL_F) {
2382 f->fs.val.proto = IPPROTO_TCP;
2383 f->fs.mask.proto = ~0;
2388 /* Mark filter as locked */
2392 /* Save the actual tid. We need this to get the corresponding
2393 * filter entry structure in filter_rpl.
2395 f->tid = stid + adap->tids.ftid_base;
2396 ret = set_filter_wr(adap, stid);
2398 clear_filter(adap, f);
2404 EXPORT_SYMBOL(cxgb4_create_server_filter);
2406 int cxgb4_remove_server_filter(const struct net_device *dev, unsigned int stid,
2407 unsigned int queue, bool ipv6)
2409 struct filter_entry *f;
2410 struct adapter *adap;
2412 adap = netdev2adap(dev);
2414 /* Adjust stid to correct filter index */
2415 stid -= adap->tids.sftid_base;
2416 stid += adap->tids.nftids;
2418 f = &adap->tids.ftid_tab[stid];
2419 /* Unlock the filter */
2422 return delete_filter(adap, stid);
2424 EXPORT_SYMBOL(cxgb4_remove_server_filter);
2426 static void cxgb_get_stats(struct net_device *dev,
2427 struct rtnl_link_stats64 *ns)
2429 struct port_stats stats;
2430 struct port_info *p = netdev_priv(dev);
2431 struct adapter *adapter = p->adapter;
2433 /* Block retrieving statistics during EEH error
2434 * recovery. Otherwise, the recovery might fail
2435 * and the PCI device will be removed permanently
2437 spin_lock(&adapter->stats_lock);
2438 if (!netif_device_present(dev)) {
2439 spin_unlock(&adapter->stats_lock);
2442 t4_get_port_stats_offset(adapter, p->tx_chan, &stats,
2444 spin_unlock(&adapter->stats_lock);
2446 ns->tx_bytes = stats.tx_octets;
2447 ns->tx_packets = stats.tx_frames;
2448 ns->rx_bytes = stats.rx_octets;
2449 ns->rx_packets = stats.rx_frames;
2450 ns->multicast = stats.rx_mcast_frames;
2452 /* detailed rx_errors */
2453 ns->rx_length_errors = stats.rx_jabber + stats.rx_too_long +
2455 ns->rx_over_errors = 0;
2456 ns->rx_crc_errors = stats.rx_fcs_err;
2457 ns->rx_frame_errors = stats.rx_symbol_err;
2458 ns->rx_dropped = stats.rx_ovflow0 + stats.rx_ovflow1 +
2459 stats.rx_ovflow2 + stats.rx_ovflow3 +
2460 stats.rx_trunc0 + stats.rx_trunc1 +
2461 stats.rx_trunc2 + stats.rx_trunc3;
2462 ns->rx_missed_errors = 0;
2464 /* detailed tx_errors */
2465 ns->tx_aborted_errors = 0;
2466 ns->tx_carrier_errors = 0;
2467 ns->tx_fifo_errors = 0;
2468 ns->tx_heartbeat_errors = 0;
2469 ns->tx_window_errors = 0;
2471 ns->tx_errors = stats.tx_error_frames;
2472 ns->rx_errors = stats.rx_symbol_err + stats.rx_fcs_err +
2473 ns->rx_length_errors + stats.rx_len_err + ns->rx_fifo_errors;
2476 static int cxgb_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
2479 int ret = 0, prtad, devad;
2480 struct port_info *pi = netdev_priv(dev);
2481 struct adapter *adapter = pi->adapter;
2482 struct mii_ioctl_data *data = (struct mii_ioctl_data *)&req->ifr_data;
2486 if (pi->mdio_addr < 0)
2488 data->phy_id = pi->mdio_addr;
2492 if (mdio_phy_id_is_c45(data->phy_id)) {
2493 prtad = mdio_phy_id_prtad(data->phy_id);
2494 devad = mdio_phy_id_devad(data->phy_id);
2495 } else if (data->phy_id < 32) {
2496 prtad = data->phy_id;
2498 data->reg_num &= 0x1f;
2502 mbox = pi->adapter->pf;
2503 if (cmd == SIOCGMIIREG)
2504 ret = t4_mdio_rd(pi->adapter, mbox, prtad, devad,
2505 data->reg_num, &data->val_out);
2507 ret = t4_mdio_wr(pi->adapter, mbox, prtad, devad,
2508 data->reg_num, data->val_in);
2511 return copy_to_user(req->ifr_data, &pi->tstamp_config,
2512 sizeof(pi->tstamp_config)) ?
2515 if (copy_from_user(&pi->tstamp_config, req->ifr_data,
2516 sizeof(pi->tstamp_config)))
2519 if (!is_t4(adapter->params.chip)) {
2520 switch (pi->tstamp_config.tx_type) {
2521 case HWTSTAMP_TX_OFF:
2522 case HWTSTAMP_TX_ON:
2528 switch (pi->tstamp_config.rx_filter) {
2529 case HWTSTAMP_FILTER_NONE:
2530 pi->rxtstamp = false;
2532 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
2533 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
2534 cxgb4_ptprx_timestamping(pi, pi->port_id,
2537 case HWTSTAMP_FILTER_PTP_V2_EVENT:
2538 cxgb4_ptprx_timestamping(pi, pi->port_id,
2541 case HWTSTAMP_FILTER_ALL:
2542 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
2543 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
2544 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
2545 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
2546 pi->rxtstamp = true;
2549 pi->tstamp_config.rx_filter =
2550 HWTSTAMP_FILTER_NONE;
2554 if ((pi->tstamp_config.tx_type == HWTSTAMP_TX_OFF) &&
2555 (pi->tstamp_config.rx_filter ==
2556 HWTSTAMP_FILTER_NONE)) {
2557 if (cxgb4_ptp_txtype(adapter, pi->port_id) >= 0)
2558 pi->ptp_enable = false;
2561 if (pi->tstamp_config.rx_filter !=
2562 HWTSTAMP_FILTER_NONE) {
2563 if (cxgb4_ptp_redirect_rx_packet(adapter,
2565 pi->ptp_enable = true;
2568 /* For T4 Adapters */
2569 switch (pi->tstamp_config.rx_filter) {
2570 case HWTSTAMP_FILTER_NONE:
2571 pi->rxtstamp = false;
2573 case HWTSTAMP_FILTER_ALL:
2574 pi->rxtstamp = true;
2577 pi->tstamp_config.rx_filter =
2578 HWTSTAMP_FILTER_NONE;
2582 return copy_to_user(req->ifr_data, &pi->tstamp_config,
2583 sizeof(pi->tstamp_config)) ?
2591 static void cxgb_set_rxmode(struct net_device *dev)
2593 /* unfortunately we can't return errors to the stack */
2594 set_rxmode(dev, -1, false);
2597 static int cxgb_change_mtu(struct net_device *dev, int new_mtu)
2600 struct port_info *pi = netdev_priv(dev);
2602 ret = t4_set_rxmode(pi->adapter, pi->adapter->pf, pi->viid, new_mtu, -1,
2609 #ifdef CONFIG_PCI_IOV
2610 static int cxgb4_mgmt_open(struct net_device *dev)
2612 /* Turn carrier off since we don't have to transmit anything on this
2615 netif_carrier_off(dev);
2619 /* Fill MAC address that will be assigned by the FW */
2620 static void cxgb4_mgmt_fill_vf_station_mac_addr(struct adapter *adap)
2622 u8 hw_addr[ETH_ALEN], macaddr[ETH_ALEN];
2623 unsigned int i, vf, nvfs;
2628 adap->params.pci.vpd_cap_addr = pci_find_capability(adap->pdev,
2630 err = t4_get_raw_vpd_params(adap, &adap->params.vpd);
2634 na = adap->params.vpd.na;
2635 for (i = 0; i < ETH_ALEN; i++)
2636 hw_addr[i] = (hex2val(na[2 * i + 0]) * 16 +
2637 hex2val(na[2 * i + 1]));
2639 a = (hw_addr[0] << 8) | hw_addr[1];
2640 b = (hw_addr[1] << 8) | hw_addr[2];
2642 a |= 0x0200; /* locally assigned Ethernet MAC address */
2643 a &= ~0x0100; /* not a multicast Ethernet MAC address */
2644 macaddr[0] = a >> 8;
2645 macaddr[1] = a & 0xff;
2647 for (i = 2; i < 5; i++)
2648 macaddr[i] = hw_addr[i + 1];
2650 for (vf = 0, nvfs = pci_sriov_get_totalvfs(adap->pdev);
2652 macaddr[5] = adap->pf * 16 + vf;
2653 ether_addr_copy(adap->vfinfo[vf].vf_mac_addr, macaddr);
2657 static int cxgb4_mgmt_set_vf_mac(struct net_device *dev, int vf, u8 *mac)
2659 struct port_info *pi = netdev_priv(dev);
2660 struct adapter *adap = pi->adapter;
2663 /* verify MAC addr is valid */
2664 if (!is_valid_ether_addr(mac)) {
2665 dev_err(pi->adapter->pdev_dev,
2666 "Invalid Ethernet address %pM for VF %d\n",
2671 dev_info(pi->adapter->pdev_dev,
2672 "Setting MAC %pM on VF %d\n", mac, vf);
2673 ret = t4_set_vf_mac_acl(adap, vf + 1, 1, mac);
2675 ether_addr_copy(adap->vfinfo[vf].vf_mac_addr, mac);
2679 static int cxgb4_mgmt_get_vf_config(struct net_device *dev,
2680 int vf, struct ifla_vf_info *ivi)
2682 struct port_info *pi = netdev_priv(dev);
2683 struct adapter *adap = pi->adapter;
2685 if (vf >= adap->num_vfs)
2688 ivi->max_tx_rate = adap->vfinfo[vf].tx_rate;
2689 ivi->min_tx_rate = 0;
2690 ether_addr_copy(ivi->mac, adap->vfinfo[vf].vf_mac_addr);
2694 static int cxgb4_mgmt_get_phys_port_id(struct net_device *dev,
2695 struct netdev_phys_item_id *ppid)
2697 struct port_info *pi = netdev_priv(dev);
2698 unsigned int phy_port_id;
2700 phy_port_id = pi->adapter->adap_idx * 10 + pi->port_id;
2701 ppid->id_len = sizeof(phy_port_id);
2702 memcpy(ppid->id, &phy_port_id, ppid->id_len);
2706 static int cxgb4_mgmt_set_vf_rate(struct net_device *dev, int vf,
2707 int min_tx_rate, int max_tx_rate)
2709 struct port_info *pi = netdev_priv(dev);
2710 struct adapter *adap = pi->adapter;
2711 unsigned int link_ok, speed, mtu;
2712 u32 fw_pfvf, fw_class;
2717 if (vf >= adap->num_vfs)
2721 dev_err(adap->pdev_dev,
2722 "Min tx rate (%d) (> 0) for VF %d is Invalid.\n",
2727 ret = t4_get_link_params(pi, &link_ok, &speed, &mtu);
2728 if (ret != FW_SUCCESS) {
2729 dev_err(adap->pdev_dev,
2730 "Failed to get link information for VF %d\n", vf);
2735 dev_err(adap->pdev_dev, "Link down for VF %d\n", vf);
2739 if (max_tx_rate > speed) {
2740 dev_err(adap->pdev_dev,
2741 "Max tx rate %d for VF %d can't be > link-speed %u",
2742 max_tx_rate, vf, speed);
2747 /* subtract ethhdr size and 4 bytes crc since, f/w appends it */
2748 pktsize = pktsize - sizeof(struct ethhdr) - 4;
2749 /* subtract ipv4 hdr size, tcp hdr size to get typical IPv4 MSS size */
2750 pktsize = pktsize - sizeof(struct iphdr) - sizeof(struct tcphdr);
2751 /* configure Traffic Class for rate-limiting */
2752 ret = t4_sched_params(adap, SCHED_CLASS_TYPE_PACKET,
2753 SCHED_CLASS_LEVEL_CL_RL,
2754 SCHED_CLASS_MODE_CLASS,
2755 SCHED_CLASS_RATEUNIT_BITS,
2756 SCHED_CLASS_RATEMODE_ABS,
2757 pi->tx_chan, class_id, 0,
2758 max_tx_rate * 1000, 0, pktsize);
2760 dev_err(adap->pdev_dev, "Err %d for Traffic Class config\n",
2764 dev_info(adap->pdev_dev,
2765 "Class %d with MSS %u configured with rate %u\n",
2766 class_id, pktsize, max_tx_rate);
2768 /* bind VF to configured Traffic Class */
2769 fw_pfvf = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) |
2770 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_SCHEDCLASS_ETH));
2771 fw_class = class_id;
2772 ret = t4_set_params(adap, adap->mbox, adap->pf, vf + 1, 1, &fw_pfvf,
2775 dev_err(adap->pdev_dev,
2776 "Err %d in binding VF %d to Traffic Class %d\n",
2780 dev_info(adap->pdev_dev, "PF %d VF %d is bound to Class %d\n",
2781 adap->pf, vf, class_id);
2782 adap->vfinfo[vf].tx_rate = max_tx_rate;
2786 static int cxgb4_mgmt_set_vf_vlan(struct net_device *dev, int vf,
2787 u16 vlan, u8 qos, __be16 vlan_proto)
2789 struct port_info *pi = netdev_priv(dev);
2790 struct adapter *adap = pi->adapter;
2793 if (vf >= adap->num_vfs || vlan > 4095 || qos > 7)
2796 if (vlan_proto != htons(ETH_P_8021Q) || qos != 0)
2797 return -EPROTONOSUPPORT;
2799 ret = t4_set_vlan_acl(adap, adap->mbox, vf + 1, vlan);
2801 adap->vfinfo[vf].vlan = vlan;
2805 dev_err(adap->pdev_dev, "Err %d %s VLAN ACL for PF/VF %d/%d\n",
2806 ret, (vlan ? "setting" : "clearing"), adap->pf, vf);
2809 #endif /* CONFIG_PCI_IOV */
2811 static int cxgb_set_mac_addr(struct net_device *dev, void *p)
2814 struct sockaddr *addr = p;
2815 struct port_info *pi = netdev_priv(dev);
2817 if (!is_valid_ether_addr(addr->sa_data))
2818 return -EADDRNOTAVAIL;
2820 ret = t4_change_mac(pi->adapter, pi->adapter->pf, pi->viid,
2821 pi->xact_addr_filt, addr->sa_data, true, true);
2825 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
2826 pi->xact_addr_filt = ret;
2830 #ifdef CONFIG_NET_POLL_CONTROLLER
2831 static void cxgb_netpoll(struct net_device *dev)
2833 struct port_info *pi = netdev_priv(dev);
2834 struct adapter *adap = pi->adapter;
2836 if (adap->flags & USING_MSIX) {
2838 struct sge_eth_rxq *rx = &adap->sge.ethrxq[pi->first_qset];
2840 for (i = pi->nqsets; i; i--, rx++)
2841 t4_sge_intr_msix(0, &rx->rspq);
2843 t4_intr_handler(adap)(0, adap);
2847 static int cxgb_set_tx_maxrate(struct net_device *dev, int index, u32 rate)
2849 struct port_info *pi = netdev_priv(dev);
2850 struct adapter *adap = pi->adapter;
2851 struct sched_class *e;
2852 struct ch_sched_params p;
2853 struct ch_sched_queue qe;
2857 if (!can_sched(dev))
2860 if (index < 0 || index > pi->nqsets - 1)
2863 if (!(adap->flags & FULL_INIT_DONE)) {
2864 dev_err(adap->pdev_dev,
2865 "Failed to rate limit on queue %d. Link Down?\n",
2870 /* Convert from Mbps to Kbps */
2871 req_rate = rate << 10;
2873 /* Max rate is 10 Gbps */
2874 if (req_rate >= SCHED_MAX_RATE_KBPS) {
2875 dev_err(adap->pdev_dev,
2876 "Invalid rate %u Mbps, Max rate is %u Gbps\n",
2877 rate, SCHED_MAX_RATE_KBPS);
2881 /* First unbind the queue from any existing class */
2882 memset(&qe, 0, sizeof(qe));
2884 qe.class = SCHED_CLS_NONE;
2886 err = cxgb4_sched_class_unbind(dev, (void *)(&qe), SCHED_QUEUE);
2888 dev_err(adap->pdev_dev,
2889 "Unbinding Queue %d on port %d fail. Err: %d\n",
2890 index, pi->port_id, err);
2894 /* Queue already unbound */
2898 /* Fetch any available unused or matching scheduling class */
2899 memset(&p, 0, sizeof(p));
2900 p.type = SCHED_CLASS_TYPE_PACKET;
2901 p.u.params.level = SCHED_CLASS_LEVEL_CL_RL;
2902 p.u.params.mode = SCHED_CLASS_MODE_CLASS;
2903 p.u.params.rateunit = SCHED_CLASS_RATEUNIT_BITS;
2904 p.u.params.ratemode = SCHED_CLASS_RATEMODE_ABS;
2905 p.u.params.channel = pi->tx_chan;
2906 p.u.params.class = SCHED_CLS_NONE;
2907 p.u.params.minrate = 0;
2908 p.u.params.maxrate = req_rate;
2909 p.u.params.weight = 0;
2910 p.u.params.pktsize = dev->mtu;
2912 e = cxgb4_sched_class_alloc(dev, &p);
2916 /* Bind the queue to a scheduling class */
2917 memset(&qe, 0, sizeof(qe));
2921 err = cxgb4_sched_class_bind(dev, (void *)(&qe), SCHED_QUEUE);
2923 dev_err(adap->pdev_dev,
2924 "Queue rate limiting failed. Err: %d\n", err);
2928 static int cxgb_setup_tc_flower(struct net_device *dev,
2929 struct tc_cls_flower_offload *cls_flower)
2931 switch (cls_flower->command) {
2932 case TC_CLSFLOWER_REPLACE:
2933 return cxgb4_tc_flower_replace(dev, cls_flower);
2934 case TC_CLSFLOWER_DESTROY:
2935 return cxgb4_tc_flower_destroy(dev, cls_flower);
2936 case TC_CLSFLOWER_STATS:
2937 return cxgb4_tc_flower_stats(dev, cls_flower);
2943 static int cxgb_setup_tc_cls_u32(struct net_device *dev,
2944 struct tc_cls_u32_offload *cls_u32)
2946 switch (cls_u32->command) {
2947 case TC_CLSU32_NEW_KNODE:
2948 case TC_CLSU32_REPLACE_KNODE:
2949 return cxgb4_config_knode(dev, cls_u32);
2950 case TC_CLSU32_DELETE_KNODE:
2951 return cxgb4_delete_knode(dev, cls_u32);
2957 static int cxgb_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
2960 struct net_device *dev = cb_priv;
2961 struct port_info *pi = netdev2pinfo(dev);
2962 struct adapter *adap = netdev2adap(dev);
2964 if (!(adap->flags & FULL_INIT_DONE)) {
2965 dev_err(adap->pdev_dev,
2966 "Failed to setup tc on port %d. Link Down?\n",
2971 if (!tc_cls_can_offload_and_chain0(dev, type_data))
2975 case TC_SETUP_CLSU32:
2976 return cxgb_setup_tc_cls_u32(dev, type_data);
2977 case TC_SETUP_CLSFLOWER:
2978 return cxgb_setup_tc_flower(dev, type_data);
2984 static int cxgb_setup_tc_block(struct net_device *dev,
2985 struct tc_block_offload *f)
2987 struct port_info *pi = netdev2pinfo(dev);
2989 if (f->binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_INGRESS)
2992 switch (f->command) {
2994 return tcf_block_cb_register(f->block, cxgb_setup_tc_block_cb,
2996 case TC_BLOCK_UNBIND:
2997 tcf_block_cb_unregister(f->block, cxgb_setup_tc_block_cb, pi);
3004 static int cxgb_setup_tc(struct net_device *dev, enum tc_setup_type type,
3008 case TC_SETUP_BLOCK:
3009 return cxgb_setup_tc_block(dev, type_data);
3015 static void cxgb_del_udp_tunnel(struct net_device *netdev,
3016 struct udp_tunnel_info *ti)
3018 struct port_info *pi = netdev_priv(netdev);
3019 struct adapter *adapter = pi->adapter;
3020 unsigned int chip_ver = CHELSIO_CHIP_VERSION(adapter->params.chip);
3021 u8 match_all_mac[] = { 0, 0, 0, 0, 0, 0 };
3024 if (chip_ver < CHELSIO_T6)
3028 case UDP_TUNNEL_TYPE_VXLAN:
3029 if (!adapter->vxlan_port_cnt ||
3030 adapter->vxlan_port != ti->port)
3031 return; /* Invalid VxLAN destination port */
3033 adapter->vxlan_port_cnt--;
3034 if (adapter->vxlan_port_cnt)
3037 adapter->vxlan_port = 0;
3038 t4_write_reg(adapter, MPS_RX_VXLAN_TYPE_A, 0);
3040 case UDP_TUNNEL_TYPE_GENEVE:
3041 if (!adapter->geneve_port_cnt ||
3042 adapter->geneve_port != ti->port)
3043 return; /* Invalid GENEVE destination port */
3045 adapter->geneve_port_cnt--;
3046 if (adapter->geneve_port_cnt)
3049 adapter->geneve_port = 0;
3050 t4_write_reg(adapter, MPS_RX_GENEVE_TYPE_A, 0);
3055 /* Matchall mac entries can be deleted only after all tunnel ports
3056 * are brought down or removed.
3058 if (!adapter->rawf_cnt)
3060 for_each_port(adapter, i) {
3061 pi = adap2pinfo(adapter, i);
3062 ret = t4_free_raw_mac_filt(adapter, pi->viid,
3063 match_all_mac, match_all_mac,
3064 adapter->rawf_start +
3066 1, pi->port_id, true);
3068 netdev_info(netdev, "Failed to free mac filter entry, for port %d\n",
3072 atomic_dec(&adapter->mps_encap[adapter->rawf_start +
3073 pi->port_id].refcnt);
3077 static void cxgb_add_udp_tunnel(struct net_device *netdev,
3078 struct udp_tunnel_info *ti)
3080 struct port_info *pi = netdev_priv(netdev);
3081 struct adapter *adapter = pi->adapter;
3082 unsigned int chip_ver = CHELSIO_CHIP_VERSION(adapter->params.chip);
3083 u8 match_all_mac[] = { 0, 0, 0, 0, 0, 0 };
3086 if (chip_ver < CHELSIO_T6 || !adapter->rawf_cnt)
3090 case UDP_TUNNEL_TYPE_VXLAN:
3091 /* Callback for adding vxlan port can be called with the same
3092 * port for both IPv4 and IPv6. We should not disable the
3093 * offloading when the same port for both protocols is added
3094 * and later one of them is removed.
3096 if (adapter->vxlan_port_cnt &&
3097 adapter->vxlan_port == ti->port) {
3098 adapter->vxlan_port_cnt++;
3102 /* We will support only one VxLAN port */
3103 if (adapter->vxlan_port_cnt) {
3104 netdev_info(netdev, "UDP port %d already offloaded, not adding port %d\n",
3105 be16_to_cpu(adapter->vxlan_port),
3106 be16_to_cpu(ti->port));
3110 adapter->vxlan_port = ti->port;
3111 adapter->vxlan_port_cnt = 1;
3113 t4_write_reg(adapter, MPS_RX_VXLAN_TYPE_A,
3114 VXLAN_V(be16_to_cpu(ti->port)) | VXLAN_EN_F);
3116 case UDP_TUNNEL_TYPE_GENEVE:
3117 if (adapter->geneve_port_cnt &&
3118 adapter->geneve_port == ti->port) {
3119 adapter->geneve_port_cnt++;
3123 /* We will support only one GENEVE port */
3124 if (adapter->geneve_port_cnt) {
3125 netdev_info(netdev, "UDP port %d already offloaded, not adding port %d\n",
3126 be16_to_cpu(adapter->geneve_port),
3127 be16_to_cpu(ti->port));
3131 adapter->geneve_port = ti->port;
3132 adapter->geneve_port_cnt = 1;
3134 t4_write_reg(adapter, MPS_RX_GENEVE_TYPE_A,
3135 GENEVE_V(be16_to_cpu(ti->port)) | GENEVE_EN_F);
3140 /* Create a 'match all' mac filter entry for inner mac,
3141 * if raw mac interface is supported. Once the linux kernel provides
3142 * driver entry points for adding/deleting the inner mac addresses,
3143 * we will remove this 'match all' entry and fallback to adding
3144 * exact match filters.
3146 for_each_port(adapter, i) {
3147 pi = adap2pinfo(adapter, i);
3149 ret = t4_alloc_raw_mac_filt(adapter, pi->viid,
3152 adapter->rawf_start +
3154 1, pi->port_id, true);
3156 netdev_info(netdev, "Failed to allocate a mac filter entry, not adding port %d\n",
3157 be16_to_cpu(ti->port));
3158 cxgb_del_udp_tunnel(netdev, ti);
3161 atomic_inc(&adapter->mps_encap[ret].refcnt);
3165 static netdev_features_t cxgb_features_check(struct sk_buff *skb,
3166 struct net_device *dev,
3167 netdev_features_t features)
3169 struct port_info *pi = netdev_priv(dev);
3170 struct adapter *adapter = pi->adapter;
3172 if (CHELSIO_CHIP_VERSION(adapter->params.chip) < CHELSIO_T6)
3175 /* Check if hw supports offload for this packet */
3176 if (!skb->encapsulation || cxgb_encap_offload_supported(skb))
3179 /* Offload is not supported for this encapsulated packet */
3180 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3183 static netdev_features_t cxgb_fix_features(struct net_device *dev,
3184 netdev_features_t features)
3186 /* Disable GRO, if RX_CSUM is disabled */
3187 if (!(features & NETIF_F_RXCSUM))
3188 features &= ~NETIF_F_GRO;
3193 static const struct net_device_ops cxgb4_netdev_ops = {
3194 .ndo_open = cxgb_open,
3195 .ndo_stop = cxgb_close,
3196 .ndo_start_xmit = t4_eth_xmit,
3197 .ndo_select_queue = cxgb_select_queue,
3198 .ndo_get_stats64 = cxgb_get_stats,
3199 .ndo_set_rx_mode = cxgb_set_rxmode,
3200 .ndo_set_mac_address = cxgb_set_mac_addr,
3201 .ndo_set_features = cxgb_set_features,
3202 .ndo_validate_addr = eth_validate_addr,
3203 .ndo_do_ioctl = cxgb_ioctl,
3204 .ndo_change_mtu = cxgb_change_mtu,
3205 #ifdef CONFIG_NET_POLL_CONTROLLER
3206 .ndo_poll_controller = cxgb_netpoll,
3208 #ifdef CONFIG_CHELSIO_T4_FCOE
3209 .ndo_fcoe_enable = cxgb_fcoe_enable,
3210 .ndo_fcoe_disable = cxgb_fcoe_disable,
3211 #endif /* CONFIG_CHELSIO_T4_FCOE */
3212 .ndo_set_tx_maxrate = cxgb_set_tx_maxrate,
3213 .ndo_setup_tc = cxgb_setup_tc,
3214 .ndo_udp_tunnel_add = cxgb_add_udp_tunnel,
3215 .ndo_udp_tunnel_del = cxgb_del_udp_tunnel,
3216 .ndo_features_check = cxgb_features_check,
3217 .ndo_fix_features = cxgb_fix_features,
3220 #ifdef CONFIG_PCI_IOV
3221 static const struct net_device_ops cxgb4_mgmt_netdev_ops = {
3222 .ndo_open = cxgb4_mgmt_open,
3223 .ndo_set_vf_mac = cxgb4_mgmt_set_vf_mac,
3224 .ndo_get_vf_config = cxgb4_mgmt_get_vf_config,
3225 .ndo_set_vf_rate = cxgb4_mgmt_set_vf_rate,
3226 .ndo_get_phys_port_id = cxgb4_mgmt_get_phys_port_id,
3227 .ndo_set_vf_vlan = cxgb4_mgmt_set_vf_vlan,
3231 static void cxgb4_mgmt_get_drvinfo(struct net_device *dev,
3232 struct ethtool_drvinfo *info)
3234 struct adapter *adapter = netdev2adap(dev);
3236 strlcpy(info->driver, cxgb4_driver_name, sizeof(info->driver));
3237 strlcpy(info->version, cxgb4_driver_version,
3238 sizeof(info->version));
3239 strlcpy(info->bus_info, pci_name(adapter->pdev),
3240 sizeof(info->bus_info));
3243 static const struct ethtool_ops cxgb4_mgmt_ethtool_ops = {
3244 .get_drvinfo = cxgb4_mgmt_get_drvinfo,
3247 void t4_fatal_err(struct adapter *adap)
3251 if (pci_channel_offline(adap->pdev))
3254 /* Disable the SGE since ULDs are going to free resources that
3255 * could be exposed to the adapter. RDMA MWs for example...
3257 t4_shutdown_adapter(adap);
3258 for_each_port(adap, port) {
3259 struct net_device *dev = adap->port[port];
3261 /* If we get here in very early initialization the network
3262 * devices may not have been set up yet.
3267 netif_tx_stop_all_queues(dev);
3268 netif_carrier_off(dev);
3270 dev_alert(adap->pdev_dev, "encountered fatal error, adapter stopped\n");
3273 static void setup_memwin(struct adapter *adap)
3275 u32 nic_win_base = t4_get_util_window(adap);
3277 t4_setup_memwin(adap, nic_win_base, MEMWIN_NIC);
3280 static void setup_memwin_rdma(struct adapter *adap)
3282 if (adap->vres.ocq.size) {
3286 start = t4_read_pcie_cfg4(adap, PCI_BASE_ADDRESS_2);
3287 start &= PCI_BASE_ADDRESS_MEM_MASK;
3288 start += OCQ_WIN_OFFSET(adap->pdev, &adap->vres);
3289 sz_kb = roundup_pow_of_two(adap->vres.ocq.size) >> 10;
3291 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN_A, 3),
3292 start | BIR_V(1) | WINDOW_V(ilog2(sz_kb)));
3294 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET_A, 3),
3295 adap->vres.ocq.start);
3297 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET_A, 3));
3301 static int adap_init1(struct adapter *adap, struct fw_caps_config_cmd *c)
3306 /* get device capabilities */
3307 memset(c, 0, sizeof(*c));
3308 c->op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3309 FW_CMD_REQUEST_F | FW_CMD_READ_F);
3310 c->cfvalid_to_len16 = htonl(FW_LEN16(*c));
3311 ret = t4_wr_mbox(adap, adap->mbox, c, sizeof(*c), c);
3315 c->op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3316 FW_CMD_REQUEST_F | FW_CMD_WRITE_F);
3317 ret = t4_wr_mbox(adap, adap->mbox, c, sizeof(*c), NULL);
3321 ret = t4_config_glbl_rss(adap, adap->pf,
3322 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL,
3323 FW_RSS_GLB_CONFIG_CMD_TNLMAPEN_F |
3324 FW_RSS_GLB_CONFIG_CMD_TNLALLLKP_F);
3328 ret = t4_cfg_pfvf(adap, adap->mbox, adap->pf, 0, adap->sge.egr_sz, 64,
3329 MAX_INGQ, 0, 0, 4, 0xf, 0xf, 16, FW_CMD_CAP_PF,
3336 /* tweak some settings */
3337 t4_write_reg(adap, TP_SHIFT_CNT_A, 0x64f8849);
3338 t4_write_reg(adap, ULP_RX_TDDP_PSZ_A, HPZ0_V(PAGE_SHIFT - 12));
3339 t4_write_reg(adap, TP_PIO_ADDR_A, TP_INGRESS_CONFIG_A);
3340 v = t4_read_reg(adap, TP_PIO_DATA_A);
3341 t4_write_reg(adap, TP_PIO_DATA_A, v & ~CSUM_HAS_PSEUDO_HDR_F);
3343 /* first 4 Tx modulation queues point to consecutive Tx channels */
3344 adap->params.tp.tx_modq_map = 0xE4;
3345 t4_write_reg(adap, TP_TX_MOD_QUEUE_REQ_MAP_A,
3346 TX_MOD_QUEUE_REQ_MAP_V(adap->params.tp.tx_modq_map));
3348 /* associate each Tx modulation queue with consecutive Tx channels */
3350 t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
3351 &v, 1, TP_TX_SCHED_HDR_A);
3352 t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
3353 &v, 1, TP_TX_SCHED_FIFO_A);
3354 t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
3355 &v, 1, TP_TX_SCHED_PCMD_A);
3357 #define T4_TX_MODQ_10G_WEIGHT_DEFAULT 16 /* in KB units */
3358 if (is_offload(adap)) {
3359 t4_write_reg(adap, TP_TX_MOD_QUEUE_WEIGHT0_A,
3360 TX_MODQ_WEIGHT0_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3361 TX_MODQ_WEIGHT1_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3362 TX_MODQ_WEIGHT2_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3363 TX_MODQ_WEIGHT3_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
3364 t4_write_reg(adap, TP_TX_MOD_CHANNEL_WEIGHT_A,
3365 TX_MODQ_WEIGHT0_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3366 TX_MODQ_WEIGHT1_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3367 TX_MODQ_WEIGHT2_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3368 TX_MODQ_WEIGHT3_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
3371 /* get basic stuff going */
3372 return t4_early_init(adap, adap->pf);
3376 * Max # of ATIDs. The absolute HW max is 16K but we keep it lower.
3378 #define MAX_ATIDS 8192U
3381 * Phase 0 of initialization: contact FW, obtain config, perform basic init.
3383 * If the firmware we're dealing with has Configuration File support, then
3384 * we use that to perform all configuration
3388 * Tweak configuration based on module parameters, etc. Most of these have
3389 * defaults assigned to them by Firmware Configuration Files (if we're using
3390 * them) but need to be explicitly set if we're using hard-coded
3391 * initialization. But even in the case of using Firmware Configuration
3392 * Files, we'd like to expose the ability to change these via module
3393 * parameters so these are essentially common tweaks/settings for
3394 * Configuration Files and hard-coded initialization ...
3396 static int adap_init0_tweaks(struct adapter *adapter)
3399 * Fix up various Host-Dependent Parameters like Page Size, Cache
3400 * Line Size, etc. The firmware default is for a 4KB Page Size and
3401 * 64B Cache Line Size ...
3403 t4_fixup_host_params(adapter, PAGE_SIZE, L1_CACHE_BYTES);
3406 * Process module parameters which affect early initialization.
3408 if (rx_dma_offset != 2 && rx_dma_offset != 0) {
3409 dev_err(&adapter->pdev->dev,
3410 "Ignoring illegal rx_dma_offset=%d, using 2\n",
3414 t4_set_reg_field(adapter, SGE_CONTROL_A,
3415 PKTSHIFT_V(PKTSHIFT_M),
3416 PKTSHIFT_V(rx_dma_offset));
3419 * Don't include the "IP Pseudo Header" in CPL_RX_PKT checksums: Linux
3420 * adds the pseudo header itself.
3422 t4_tp_wr_bits_indirect(adapter, TP_INGRESS_CONFIG_A,
3423 CSUM_HAS_PSEUDO_HDR_F, 0);
3428 /* 10Gb/s-BT PHY Support. chip-external 10Gb/s-BT PHYs are complex chips
3429 * unto themselves and they contain their own firmware to perform their
3432 static int phy_aq1202_version(const u8 *phy_fw_data,
3437 /* At offset 0x8 you're looking for the primary image's
3438 * starting offset which is 3 Bytes wide
3440 * At offset 0xa of the primary image, you look for the offset
3441 * of the DRAM segment which is 3 Bytes wide.
3443 * The FW version is at offset 0x27e of the DRAM and is 2 Bytes
3446 #define be16(__p) (((__p)[0] << 8) | (__p)[1])
3447 #define le16(__p) ((__p)[0] | ((__p)[1] << 8))
3448 #define le24(__p) (le16(__p) | ((__p)[2] << 16))
3450 offset = le24(phy_fw_data + 0x8) << 12;
3451 offset = le24(phy_fw_data + offset + 0xa);
3452 return be16(phy_fw_data + offset + 0x27e);
3459 static struct info_10gbt_phy_fw {
3460 unsigned int phy_fw_id; /* PCI Device ID */
3461 char *phy_fw_file; /* /lib/firmware/ PHY Firmware file */
3462 int (*phy_fw_version)(const u8 *phy_fw_data, size_t phy_fw_size);
3463 int phy_flash; /* Has FLASH for PHY Firmware */
3464 } phy_info_array[] = {
3466 PHY_AQ1202_DEVICEID,
3467 PHY_AQ1202_FIRMWARE,
3472 PHY_BCM84834_DEVICEID,
3473 PHY_BCM84834_FIRMWARE,
3480 static struct info_10gbt_phy_fw *find_phy_info(int devid)
3484 for (i = 0; i < ARRAY_SIZE(phy_info_array); i++) {
3485 if (phy_info_array[i].phy_fw_id == devid)
3486 return &phy_info_array[i];
3491 /* Handle updating of chip-external 10Gb/s-BT PHY firmware. This needs to
3492 * happen after the FW_RESET_CMD but before the FW_INITIALIZE_CMD. On error
3493 * we return a negative error number. If we transfer new firmware we return 1
3494 * (from t4_load_phy_fw()). If we don't do anything we return 0.
3496 static int adap_init0_phy(struct adapter *adap)
3498 const struct firmware *phyf;
3500 struct info_10gbt_phy_fw *phy_info;
3502 /* Use the device ID to determine which PHY file to flash.
3504 phy_info = find_phy_info(adap->pdev->device);
3506 dev_warn(adap->pdev_dev,
3507 "No PHY Firmware file found for this PHY\n");
3511 /* If we have a T4 PHY firmware file under /lib/firmware/cxgb4/, then
3512 * use that. The adapter firmware provides us with a memory buffer
3513 * where we can load a PHY firmware file from the host if we want to
3514 * override the PHY firmware File in flash.
3516 ret = request_firmware_direct(&phyf, phy_info->phy_fw_file,
3519 /* For adapters without FLASH attached to PHY for their
3520 * firmware, it's obviously a fatal error if we can't get the
3521 * firmware to the adapter. For adapters with PHY firmware
3522 * FLASH storage, it's worth a warning if we can't find the
3523 * PHY Firmware but we'll neuter the error ...
3525 dev_err(adap->pdev_dev, "unable to find PHY Firmware image "
3526 "/lib/firmware/%s, error %d\n",
3527 phy_info->phy_fw_file, -ret);
3528 if (phy_info->phy_flash) {
3529 int cur_phy_fw_ver = 0;
3531 t4_phy_fw_ver(adap, &cur_phy_fw_ver);
3532 dev_warn(adap->pdev_dev, "continuing with, on-adapter "
3533 "FLASH copy, version %#x\n", cur_phy_fw_ver);
3540 /* Load PHY Firmware onto adapter.
3542 ret = t4_load_phy_fw(adap, MEMWIN_NIC, &adap->win0_lock,
3543 phy_info->phy_fw_version,
3544 (u8 *)phyf->data, phyf->size);
3546 dev_err(adap->pdev_dev, "PHY Firmware transfer error %d\n",
3549 int new_phy_fw_ver = 0;
3551 if (phy_info->phy_fw_version)
3552 new_phy_fw_ver = phy_info->phy_fw_version(phyf->data,
3554 dev_info(adap->pdev_dev, "Successfully transferred PHY "
3555 "Firmware /lib/firmware/%s, version %#x\n",
3556 phy_info->phy_fw_file, new_phy_fw_ver);
3559 release_firmware(phyf);
3565 * Attempt to initialize the adapter via a Firmware Configuration File.
3567 static int adap_init0_config(struct adapter *adapter, int reset)
3569 struct fw_caps_config_cmd caps_cmd;
3570 const struct firmware *cf;
3571 unsigned long mtype = 0, maddr = 0;
3572 u32 finiver, finicsum, cfcsum;
3574 int config_issued = 0;
3575 char *fw_config_file, fw_config_file_path[256];
3576 char *config_name = NULL;
3579 * Reset device if necessary.
3582 ret = t4_fw_reset(adapter, adapter->mbox,
3583 PIORSTMODE_F | PIORST_F);
3588 /* If this is a 10Gb/s-BT adapter make sure the chip-external
3589 * 10Gb/s-BT PHYs have up-to-date firmware. Note that this step needs
3590 * to be performed after any global adapter RESET above since some
3591 * PHYs only have local RAM copies of the PHY firmware.
3593 if (is_10gbt_device(adapter->pdev->device)) {
3594 ret = adap_init0_phy(adapter);
3599 * If we have a T4 configuration file under /lib/firmware/cxgb4/,
3600 * then use that. Otherwise, use the configuration file stored
3601 * in the adapter flash ...
3603 switch (CHELSIO_CHIP_VERSION(adapter->params.chip)) {
3605 fw_config_file = FW4_CFNAME;
3608 fw_config_file = FW5_CFNAME;
3611 fw_config_file = FW6_CFNAME;
3614 dev_err(adapter->pdev_dev, "Device %d is not supported\n",
3615 adapter->pdev->device);
3620 ret = request_firmware(&cf, fw_config_file, adapter->pdev_dev);
3622 config_name = "On FLASH";
3623 mtype = FW_MEMTYPE_CF_FLASH;
3624 maddr = t4_flash_cfg_addr(adapter);
3626 u32 params[7], val[7];
3628 sprintf(fw_config_file_path,
3629 "/lib/firmware/%s", fw_config_file);
3630 config_name = fw_config_file_path;
3632 if (cf->size >= FLASH_CFG_MAX_SIZE)
3635 params[0] = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
3636 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_CF));
3637 ret = t4_query_params(adapter, adapter->mbox,
3638 adapter->pf, 0, 1, params, val);
3641 * For t4_memory_rw() below addresses and
3642 * sizes have to be in terms of multiples of 4
3643 * bytes. So, if the Configuration File isn't
3644 * a multiple of 4 bytes in length we'll have
3645 * to write that out separately since we can't
3646 * guarantee that the bytes following the
3647 * residual byte in the buffer returned by
3648 * request_firmware() are zeroed out ...
3650 size_t resid = cf->size & 0x3;
3651 size_t size = cf->size & ~0x3;
3652 __be32 *data = (__be32 *)cf->data;
3654 mtype = FW_PARAMS_PARAM_Y_G(val[0]);
3655 maddr = FW_PARAMS_PARAM_Z_G(val[0]) << 16;
3657 spin_lock(&adapter->win0_lock);
3658 ret = t4_memory_rw(adapter, 0, mtype, maddr,
3659 size, data, T4_MEMORY_WRITE);
3660 if (ret == 0 && resid != 0) {
3667 last.word = data[size >> 2];
3668 for (i = resid; i < 4; i++)
3670 ret = t4_memory_rw(adapter, 0, mtype,
3675 spin_unlock(&adapter->win0_lock);
3679 release_firmware(cf);
3685 * Issue a Capability Configuration command to the firmware to get it
3686 * to parse the Configuration File. We don't use t4_fw_config_file()
3687 * because we want the ability to modify various features after we've
3688 * processed the configuration file ...
3690 memset(&caps_cmd, 0, sizeof(caps_cmd));
3691 caps_cmd.op_to_write =
3692 htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3695 caps_cmd.cfvalid_to_len16 =
3696 htonl(FW_CAPS_CONFIG_CMD_CFVALID_F |
3697 FW_CAPS_CONFIG_CMD_MEMTYPE_CF_V(mtype) |
3698 FW_CAPS_CONFIG_CMD_MEMADDR64K_CF_V(maddr >> 16) |
3699 FW_LEN16(caps_cmd));
3700 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
3703 /* If the CAPS_CONFIG failed with an ENOENT (for a Firmware
3704 * Configuration File in FLASH), our last gasp effort is to use the
3705 * Firmware Configuration File which is embedded in the firmware. A
3706 * very few early versions of the firmware didn't have one embedded
3707 * but we can ignore those.
3709 if (ret == -ENOENT) {
3710 memset(&caps_cmd, 0, sizeof(caps_cmd));
3711 caps_cmd.op_to_write =
3712 htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3715 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
3716 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd,
3717 sizeof(caps_cmd), &caps_cmd);
3718 config_name = "Firmware Default";
3725 finiver = ntohl(caps_cmd.finiver);
3726 finicsum = ntohl(caps_cmd.finicsum);
3727 cfcsum = ntohl(caps_cmd.cfcsum);
3728 if (finicsum != cfcsum)
3729 dev_warn(adapter->pdev_dev, "Configuration File checksum "\
3730 "mismatch: [fini] csum=%#x, computed csum=%#x\n",
3734 * And now tell the firmware to use the configuration we just loaded.
3736 caps_cmd.op_to_write =
3737 htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3740 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
3741 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
3747 * Tweak configuration based on system architecture, module
3750 ret = adap_init0_tweaks(adapter);
3755 * And finally tell the firmware to initialize itself using the
3756 * parameters from the Configuration File.
3758 ret = t4_fw_initialize(adapter, adapter->mbox);
3762 /* Emit Firmware Configuration File information and return
3765 dev_info(adapter->pdev_dev, "Successfully configured using Firmware "\
3766 "Configuration File \"%s\", version %#x, computed checksum %#x\n",
3767 config_name, finiver, cfcsum);
3771 * Something bad happened. Return the error ... (If the "error"
3772 * is that there's no Configuration File on the adapter we don't
3773 * want to issue a warning since this is fairly common.)
3776 if (config_issued && ret != -ENOENT)
3777 dev_warn(adapter->pdev_dev, "\"%s\" configuration file error %d\n",
3782 static struct fw_info fw_info_array[] = {
3785 .fs_name = FW4_CFNAME,
3786 .fw_mod_name = FW4_FNAME,
3788 .chip = FW_HDR_CHIP_T4,
3789 .fw_ver = __cpu_to_be32(FW_VERSION(T4)),
3790 .intfver_nic = FW_INTFVER(T4, NIC),
3791 .intfver_vnic = FW_INTFVER(T4, VNIC),
3792 .intfver_ri = FW_INTFVER(T4, RI),
3793 .intfver_iscsi = FW_INTFVER(T4, ISCSI),
3794 .intfver_fcoe = FW_INTFVER(T4, FCOE),
3798 .fs_name = FW5_CFNAME,
3799 .fw_mod_name = FW5_FNAME,
3801 .chip = FW_HDR_CHIP_T5,
3802 .fw_ver = __cpu_to_be32(FW_VERSION(T5)),
3803 .intfver_nic = FW_INTFVER(T5, NIC),
3804 .intfver_vnic = FW_INTFVER(T5, VNIC),
3805 .intfver_ri = FW_INTFVER(T5, RI),
3806 .intfver_iscsi = FW_INTFVER(T5, ISCSI),
3807 .intfver_fcoe = FW_INTFVER(T5, FCOE),
3811 .fs_name = FW6_CFNAME,
3812 .fw_mod_name = FW6_FNAME,
3814 .chip = FW_HDR_CHIP_T6,
3815 .fw_ver = __cpu_to_be32(FW_VERSION(T6)),
3816 .intfver_nic = FW_INTFVER(T6, NIC),
3817 .intfver_vnic = FW_INTFVER(T6, VNIC),
3818 .intfver_ofld = FW_INTFVER(T6, OFLD),
3819 .intfver_ri = FW_INTFVER(T6, RI),
3820 .intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
3821 .intfver_iscsi = FW_INTFVER(T6, ISCSI),
3822 .intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
3823 .intfver_fcoe = FW_INTFVER(T6, FCOE),
3829 static struct fw_info *find_fw_info(int chip)
3833 for (i = 0; i < ARRAY_SIZE(fw_info_array); i++) {
3834 if (fw_info_array[i].chip == chip)
3835 return &fw_info_array[i];
3841 * Phase 0 of initialization: contact FW, obtain config, perform basic init.
3843 static int adap_init0(struct adapter *adap)
3847 enum dev_state state;
3848 u32 params[7], val[7];
3849 struct fw_caps_config_cmd caps_cmd;
3852 /* Grab Firmware Device Log parameters as early as possible so we have
3853 * access to it for debugging, etc.
3855 ret = t4_init_devlog_params(adap);
3859 /* Contact FW, advertising Master capability */
3860 ret = t4_fw_hello(adap, adap->mbox, adap->mbox,
3861 is_kdump_kernel() ? MASTER_MUST : MASTER_MAY, &state);
3863 dev_err(adap->pdev_dev, "could not connect to FW, error %d\n",
3867 if (ret == adap->mbox)
3868 adap->flags |= MASTER_PF;
3871 * If we're the Master PF Driver and the device is uninitialized,
3872 * then let's consider upgrading the firmware ... (We always want
3873 * to check the firmware version number in order to A. get it for
3874 * later reporting and B. to warn if the currently loaded firmware
3875 * is excessively mismatched relative to the driver.)
3878 t4_get_version_info(adap);
3879 ret = t4_check_fw_version(adap);
3880 /* If firmware is too old (not supported by driver) force an update. */
3882 state = DEV_STATE_UNINIT;
3883 if ((adap->flags & MASTER_PF) && state != DEV_STATE_INIT) {
3884 struct fw_info *fw_info;
3885 struct fw_hdr *card_fw;
3886 const struct firmware *fw;
3887 const u8 *fw_data = NULL;
3888 unsigned int fw_size = 0;
3890 /* This is the firmware whose headers the driver was compiled
3893 fw_info = find_fw_info(CHELSIO_CHIP_VERSION(adap->params.chip));
3894 if (fw_info == NULL) {
3895 dev_err(adap->pdev_dev,
3896 "unable to get firmware info for chip %d.\n",
3897 CHELSIO_CHIP_VERSION(adap->params.chip));
3901 /* allocate memory to read the header of the firmware on the
3904 card_fw = kvzalloc(sizeof(*card_fw), GFP_KERNEL);
3906 /* Get FW from from /lib/firmware/ */
3907 ret = request_firmware(&fw, fw_info->fw_mod_name,
3910 dev_err(adap->pdev_dev,
3911 "unable to load firmware image %s, error %d\n",
3912 fw_info->fw_mod_name, ret);
3918 /* upgrade FW logic */
3919 ret = t4_prep_fw(adap, fw_info, fw_data, fw_size, card_fw,
3923 release_firmware(fw);
3931 * Grab VPD parameters. This should be done after we establish a
3932 * connection to the firmware since some of the VPD parameters
3933 * (notably the Core Clock frequency) are retrieved via requests to
3934 * the firmware. On the other hand, we need these fairly early on
3935 * so we do this right after getting ahold of the firmware.
3937 ret = t4_get_vpd_params(adap, &adap->params.vpd);
3942 * Find out what ports are available to us. Note that we need to do
3943 * this before calling adap_init0_no_config() since it needs nports
3947 FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
3948 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_PORTVEC);
3949 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, &v, &port_vec);
3953 adap->params.nports = hweight32(port_vec);
3954 adap->params.portvec = port_vec;
3956 /* If the firmware is initialized already, emit a simply note to that
3957 * effect. Otherwise, it's time to try initializing the adapter.
3959 if (state == DEV_STATE_INIT) {
3960 dev_info(adap->pdev_dev, "Coming up as %s: "\
3961 "Adapter already initialized\n",
3962 adap->flags & MASTER_PF ? "MASTER" : "SLAVE");
3964 dev_info(adap->pdev_dev, "Coming up as MASTER: "\
3965 "Initializing adapter\n");
3967 /* Find out whether we're dealing with a version of the
3968 * firmware which has configuration file support.
3970 params[0] = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
3971 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_CF));
3972 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1,
3975 /* If the firmware doesn't support Configuration Files,
3979 dev_err(adap->pdev_dev, "firmware doesn't support "
3980 "Firmware Configuration Files\n");
3984 /* The firmware provides us with a memory buffer where we can
3985 * load a Configuration File from the host if we want to
3986 * override the Configuration File in flash.
3988 ret = adap_init0_config(adap, reset);
3989 if (ret == -ENOENT) {
3990 dev_err(adap->pdev_dev, "no Configuration File "
3991 "present on adapter.\n");
3995 dev_err(adap->pdev_dev, "could not initialize "
3996 "adapter, error %d\n", -ret);
4001 /* Give the SGE code a chance to pull in anything that it needs ...
4002 * Note that this must be called after we retrieve our VPD parameters
4003 * in order to know how to convert core ticks to seconds, etc.
4005 ret = t4_sge_init(adap);
4009 if (is_bypass_device(adap->pdev->device))
4010 adap->params.bypass = 1;
4013 * Grab some of our basic fundamental operating parameters.
4015 #define FW_PARAM_DEV(param) \
4016 (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) | \
4017 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_##param))
4019 #define FW_PARAM_PFVF(param) \
4020 FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) | \
4021 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_##param)| \
4022 FW_PARAMS_PARAM_Y_V(0) | \
4023 FW_PARAMS_PARAM_Z_V(0)
4025 params[0] = FW_PARAM_PFVF(EQ_START);
4026 params[1] = FW_PARAM_PFVF(L2T_START);
4027 params[2] = FW_PARAM_PFVF(L2T_END);
4028 params[3] = FW_PARAM_PFVF(FILTER_START);
4029 params[4] = FW_PARAM_PFVF(FILTER_END);
4030 params[5] = FW_PARAM_PFVF(IQFLINT_START);
4031 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6, params, val);
4034 adap->sge.egr_start = val[0];
4035 adap->l2t_start = val[1];
4036 adap->l2t_end = val[2];
4037 adap->tids.ftid_base = val[3];
4038 adap->tids.nftids = val[4] - val[3] + 1;
4039 adap->sge.ingr_start = val[5];
4041 /* qids (ingress/egress) returned from firmware can be anywhere
4042 * in the range from EQ(IQFLINT)_START to EQ(IQFLINT)_END.
4043 * Hence driver needs to allocate memory for this range to
4044 * store the queue info. Get the highest IQFLINT/EQ index returned
4045 * in FW_EQ_*_CMD.alloc command.
4047 params[0] = FW_PARAM_PFVF(EQ_END);
4048 params[1] = FW_PARAM_PFVF(IQFLINT_END);
4049 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
4052 adap->sge.egr_sz = val[0] - adap->sge.egr_start + 1;
4053 adap->sge.ingr_sz = val[1] - adap->sge.ingr_start + 1;
4055 adap->sge.egr_map = kcalloc(adap->sge.egr_sz,
4056 sizeof(*adap->sge.egr_map), GFP_KERNEL);
4057 if (!adap->sge.egr_map) {
4062 adap->sge.ingr_map = kcalloc(adap->sge.ingr_sz,
4063 sizeof(*adap->sge.ingr_map), GFP_KERNEL);
4064 if (!adap->sge.ingr_map) {
4069 /* Allocate the memory for the vaious egress queue bitmaps
4070 * ie starving_fl, txq_maperr and blocked_fl.
4072 adap->sge.starving_fl = kcalloc(BITS_TO_LONGS(adap->sge.egr_sz),
4073 sizeof(long), GFP_KERNEL);
4074 if (!adap->sge.starving_fl) {
4079 adap->sge.txq_maperr = kcalloc(BITS_TO_LONGS(adap->sge.egr_sz),
4080 sizeof(long), GFP_KERNEL);
4081 if (!adap->sge.txq_maperr) {
4086 #ifdef CONFIG_DEBUG_FS
4087 adap->sge.blocked_fl = kcalloc(BITS_TO_LONGS(adap->sge.egr_sz),
4088 sizeof(long), GFP_KERNEL);
4089 if (!adap->sge.blocked_fl) {
4095 params[0] = FW_PARAM_PFVF(CLIP_START);
4096 params[1] = FW_PARAM_PFVF(CLIP_END);
4097 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
4100 adap->clipt_start = val[0];
4101 adap->clipt_end = val[1];
4103 /* We don't yet have a PARAMs calls to retrieve the number of Traffic
4104 * Classes supported by the hardware/firmware so we hard code it here
4107 adap->params.nsched_cls = is_t4(adap->params.chip) ? 15 : 16;
4109 /* query params related to active filter region */
4110 params[0] = FW_PARAM_PFVF(ACTIVE_FILTER_START);
4111 params[1] = FW_PARAM_PFVF(ACTIVE_FILTER_END);
4112 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
4113 /* If Active filter size is set we enable establishing
4114 * offload connection through firmware work request
4116 if ((val[0] != val[1]) && (ret >= 0)) {
4117 adap->flags |= FW_OFLD_CONN;
4118 adap->tids.aftid_base = val[0];
4119 adap->tids.aftid_end = val[1];
4122 /* If we're running on newer firmware, let it know that we're
4123 * prepared to deal with encapsulated CPL messages. Older
4124 * firmware won't understand this and we'll just get
4125 * unencapsulated messages ...
4127 params[0] = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
4129 (void)t4_set_params(adap, adap->mbox, adap->pf, 0, 1, params, val);
4132 * Find out whether we're allowed to use the T5+ ULPTX MEMWRITE DSGL
4133 * capability. Earlier versions of the firmware didn't have the
4134 * ULPTX_MEMWRITE_DSGL so we'll interpret a query failure as no
4135 * permission to use ULPTX MEMWRITE DSGL.
4137 if (is_t4(adap->params.chip)) {
4138 adap->params.ulptx_memwrite_dsgl = false;
4140 params[0] = FW_PARAM_DEV(ULPTX_MEMWRITE_DSGL);
4141 ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
4143 adap->params.ulptx_memwrite_dsgl = (ret == 0 && val[0] != 0);
4146 /* See if FW supports FW_RI_FR_NSMR_TPTE_WR work request */
4147 params[0] = FW_PARAM_DEV(RI_FR_NSMR_TPTE_WR);
4148 ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
4150 adap->params.fr_nsmr_tpte_wr_support = (ret == 0 && val[0] != 0);
4152 /* See if FW supports FW_FILTER2 work request */
4153 if (is_t4(adap->params.chip)) {
4154 adap->params.filter2_wr_support = 0;
4156 params[0] = FW_PARAM_DEV(FILTER2_WR);
4157 ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
4159 adap->params.filter2_wr_support = (ret == 0 && val[0] != 0);
4163 * Get device capabilities so we can determine what resources we need
4166 memset(&caps_cmd, 0, sizeof(caps_cmd));
4167 caps_cmd.op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
4168 FW_CMD_REQUEST_F | FW_CMD_READ_F);
4169 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
4170 ret = t4_wr_mbox(adap, adap->mbox, &caps_cmd, sizeof(caps_cmd),
4175 if (caps_cmd.ofldcaps ||
4176 (caps_cmd.niccaps & htons(FW_CAPS_CONFIG_NIC_HASHFILTER))) {
4177 /* query offload-related parameters */
4178 params[0] = FW_PARAM_DEV(NTID);
4179 params[1] = FW_PARAM_PFVF(SERVER_START);
4180 params[2] = FW_PARAM_PFVF(SERVER_END);
4181 params[3] = FW_PARAM_PFVF(TDDP_START);
4182 params[4] = FW_PARAM_PFVF(TDDP_END);
4183 params[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
4184 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6,
4188 adap->tids.ntids = val[0];
4189 adap->tids.natids = min(adap->tids.ntids / 2, MAX_ATIDS);
4190 adap->tids.stid_base = val[1];
4191 adap->tids.nstids = val[2] - val[1] + 1;
4193 * Setup server filter region. Divide the available filter
4194 * region into two parts. Regular filters get 1/3rd and server
4195 * filters get 2/3rd part. This is only enabled if workarond
4197 * 1. For regular filters.
4198 * 2. Server filter: This are special filters which are used
4199 * to redirect SYN packets to offload queue.
4201 if (adap->flags & FW_OFLD_CONN && !is_bypass(adap)) {
4202 adap->tids.sftid_base = adap->tids.ftid_base +
4203 DIV_ROUND_UP(adap->tids.nftids, 3);
4204 adap->tids.nsftids = adap->tids.nftids -
4205 DIV_ROUND_UP(adap->tids.nftids, 3);
4206 adap->tids.nftids = adap->tids.sftid_base -
4207 adap->tids.ftid_base;
4209 adap->vres.ddp.start = val[3];
4210 adap->vres.ddp.size = val[4] - val[3] + 1;
4211 adap->params.ofldq_wr_cred = val[5];
4213 if (caps_cmd.niccaps & htons(FW_CAPS_CONFIG_NIC_HASHFILTER)) {
4214 if (init_hash_filter(adap) < 0)
4217 adap->params.offload = 1;
4218 adap->num_ofld_uld += 1;
4221 if (caps_cmd.rdmacaps) {
4222 params[0] = FW_PARAM_PFVF(STAG_START);
4223 params[1] = FW_PARAM_PFVF(STAG_END);
4224 params[2] = FW_PARAM_PFVF(RQ_START);
4225 params[3] = FW_PARAM_PFVF(RQ_END);
4226 params[4] = FW_PARAM_PFVF(PBL_START);
4227 params[5] = FW_PARAM_PFVF(PBL_END);
4228 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6,
4232 adap->vres.stag.start = val[0];
4233 adap->vres.stag.size = val[1] - val[0] + 1;
4234 adap->vres.rq.start = val[2];
4235 adap->vres.rq.size = val[3] - val[2] + 1;
4236 adap->vres.pbl.start = val[4];
4237 adap->vres.pbl.size = val[5] - val[4] + 1;
4239 params[0] = FW_PARAM_PFVF(SQRQ_START);
4240 params[1] = FW_PARAM_PFVF(SQRQ_END);
4241 params[2] = FW_PARAM_PFVF(CQ_START);
4242 params[3] = FW_PARAM_PFVF(CQ_END);
4243 params[4] = FW_PARAM_PFVF(OCQ_START);
4244 params[5] = FW_PARAM_PFVF(OCQ_END);
4245 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6, params,
4249 adap->vres.qp.start = val[0];
4250 adap->vres.qp.size = val[1] - val[0] + 1;
4251 adap->vres.cq.start = val[2];
4252 adap->vres.cq.size = val[3] - val[2] + 1;
4253 adap->vres.ocq.start = val[4];
4254 adap->vres.ocq.size = val[5] - val[4] + 1;
4256 params[0] = FW_PARAM_DEV(MAXORDIRD_QP);
4257 params[1] = FW_PARAM_DEV(MAXIRD_ADAPTER);
4258 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params,
4261 adap->params.max_ordird_qp = 8;
4262 adap->params.max_ird_adapter = 32 * adap->tids.ntids;
4265 adap->params.max_ordird_qp = val[0];
4266 adap->params.max_ird_adapter = val[1];
4268 dev_info(adap->pdev_dev,
4269 "max_ordird_qp %d max_ird_adapter %d\n",
4270 adap->params.max_ordird_qp,
4271 adap->params.max_ird_adapter);
4272 adap->num_ofld_uld += 2;
4274 if (caps_cmd.iscsicaps) {
4275 params[0] = FW_PARAM_PFVF(ISCSI_START);
4276 params[1] = FW_PARAM_PFVF(ISCSI_END);
4277 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
4281 adap->vres.iscsi.start = val[0];
4282 adap->vres.iscsi.size = val[1] - val[0] + 1;
4283 /* LIO target and cxgb4i initiaitor */
4284 adap->num_ofld_uld += 2;
4286 if (caps_cmd.cryptocaps) {
4287 /* Should query params here...TODO */
4288 params[0] = FW_PARAM_PFVF(NCRYPTO_LOOKASIDE);
4289 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
4295 adap->vres.ncrypto_fc = val[0];
4297 adap->params.crypto = ntohs(caps_cmd.cryptocaps);
4300 #undef FW_PARAM_PFVF
4303 /* The MTU/MSS Table is initialized by now, so load their values. If
4304 * we're initializing the adapter, then we'll make any modifications
4305 * we want to the MTU/MSS Table and also initialize the congestion
4308 t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
4309 if (state != DEV_STATE_INIT) {
4312 /* The default MTU Table contains values 1492 and 1500.
4313 * However, for TCP, it's better to have two values which are
4314 * a multiple of 8 +/- 4 bytes apart near this popular MTU.
4315 * This allows us to have a TCP Data Payload which is a
4316 * multiple of 8 regardless of what combination of TCP Options
4317 * are in use (always a multiple of 4 bytes) which is
4318 * important for performance reasons. For instance, if no
4319 * options are in use, then we have a 20-byte IP header and a
4320 * 20-byte TCP header. In this case, a 1500-byte MSS would
4321 * result in a TCP Data Payload of 1500 - 40 == 1460 bytes
4322 * which is not a multiple of 8. So using an MSS of 1488 in
4323 * this case results in a TCP Data Payload of 1448 bytes which
4324 * is a multiple of 8. On the other hand, if 12-byte TCP Time
4325 * Stamps have been negotiated, then an MTU of 1500 bytes
4326 * results in a TCP Data Payload of 1448 bytes which, as
4327 * above, is a multiple of 8 bytes ...
4329 for (i = 0; i < NMTUS; i++)
4330 if (adap->params.mtus[i] == 1492) {
4331 adap->params.mtus[i] = 1488;
4335 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
4336 adap->params.b_wnd);
4338 t4_init_sge_params(adap);
4339 adap->flags |= FW_OK;
4340 t4_init_tp_params(adap, true);
4344 * Something bad happened. If a command timed out or failed with EIO
4345 * FW does not operate within its spec or something catastrophic
4346 * happened to HW/FW, stop issuing commands.
4349 kfree(adap->sge.egr_map);
4350 kfree(adap->sge.ingr_map);
4351 kfree(adap->sge.starving_fl);
4352 kfree(adap->sge.txq_maperr);
4353 #ifdef CONFIG_DEBUG_FS
4354 kfree(adap->sge.blocked_fl);
4356 if (ret != -ETIMEDOUT && ret != -EIO)
4357 t4_fw_bye(adap, adap->mbox);
4363 static pci_ers_result_t eeh_err_detected(struct pci_dev *pdev,
4364 pci_channel_state_t state)
4367 struct adapter *adap = pci_get_drvdata(pdev);
4373 adap->flags &= ~FW_OK;
4374 notify_ulds(adap, CXGB4_STATE_START_RECOVERY);
4375 spin_lock(&adap->stats_lock);
4376 for_each_port(adap, i) {
4377 struct net_device *dev = adap->port[i];
4379 netif_device_detach(dev);
4380 netif_carrier_off(dev);
4383 spin_unlock(&adap->stats_lock);
4384 disable_interrupts(adap);
4385 if (adap->flags & FULL_INIT_DONE)
4388 if ((adap->flags & DEV_ENABLED)) {
4389 pci_disable_device(pdev);
4390 adap->flags &= ~DEV_ENABLED;
4392 out: return state == pci_channel_io_perm_failure ?
4393 PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
4396 static pci_ers_result_t eeh_slot_reset(struct pci_dev *pdev)
4399 struct fw_caps_config_cmd c;
4400 struct adapter *adap = pci_get_drvdata(pdev);
4403 pci_restore_state(pdev);
4404 pci_save_state(pdev);
4405 return PCI_ERS_RESULT_RECOVERED;
4408 if (!(adap->flags & DEV_ENABLED)) {
4409 if (pci_enable_device(pdev)) {
4410 dev_err(&pdev->dev, "Cannot reenable PCI "
4411 "device after reset\n");
4412 return PCI_ERS_RESULT_DISCONNECT;
4414 adap->flags |= DEV_ENABLED;
4417 pci_set_master(pdev);
4418 pci_restore_state(pdev);
4419 pci_save_state(pdev);
4420 pci_cleanup_aer_uncorrect_error_status(pdev);
4422 if (t4_wait_dev_ready(adap->regs) < 0)
4423 return PCI_ERS_RESULT_DISCONNECT;
4424 if (t4_fw_hello(adap, adap->mbox, adap->pf, MASTER_MUST, NULL) < 0)
4425 return PCI_ERS_RESULT_DISCONNECT;
4426 adap->flags |= FW_OK;
4427 if (adap_init1(adap, &c))
4428 return PCI_ERS_RESULT_DISCONNECT;
4430 for_each_port(adap, i) {
4431 struct port_info *p = adap2pinfo(adap, i);
4433 ret = t4_alloc_vi(adap, adap->mbox, p->tx_chan, adap->pf, 0, 1,
4436 return PCI_ERS_RESULT_DISCONNECT;
4438 p->xact_addr_filt = -1;
4441 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
4442 adap->params.b_wnd);
4445 return PCI_ERS_RESULT_DISCONNECT;
4446 return PCI_ERS_RESULT_RECOVERED;
4449 static void eeh_resume(struct pci_dev *pdev)
4452 struct adapter *adap = pci_get_drvdata(pdev);
4458 for_each_port(adap, i) {
4459 struct net_device *dev = adap->port[i];
4461 if (netif_running(dev)) {
4463 cxgb_set_rxmode(dev);
4465 netif_device_attach(dev);
4471 static const struct pci_error_handlers cxgb4_eeh = {
4472 .error_detected = eeh_err_detected,
4473 .slot_reset = eeh_slot_reset,
4474 .resume = eeh_resume,
4477 /* Return true if the Link Configuration supports "High Speeds" (those greater
4480 static inline bool is_x_10g_port(const struct link_config *lc)
4482 unsigned int speeds, high_speeds;
4484 speeds = FW_PORT_CAP32_SPEED_V(FW_PORT_CAP32_SPEED_G(lc->pcaps));
4485 high_speeds = speeds &
4486 ~(FW_PORT_CAP32_SPEED_100M | FW_PORT_CAP32_SPEED_1G);
4488 return high_speeds != 0;
4492 * Perform default configuration of DMA queues depending on the number and type
4493 * of ports we found and the number of available CPUs. Most settings can be
4494 * modified by the admin prior to actual use.
4496 static void cfg_queues(struct adapter *adap)
4498 struct sge *s = &adap->sge;
4499 int i = 0, n10g = 0, qidx = 0;
4500 #ifndef CONFIG_CHELSIO_T4_DCB
4504 /* Reduce memory usage in kdump environment, disable all offload.
4506 if (is_kdump_kernel() || (is_uld(adap) && t4_uld_mem_alloc(adap))) {
4507 adap->params.offload = 0;
4508 adap->params.crypto = 0;
4511 n10g += is_x_10g_port(&adap2pinfo(adap, i)->link_cfg);
4512 #ifdef CONFIG_CHELSIO_T4_DCB
4513 /* For Data Center Bridging support we need to be able to support up
4514 * to 8 Traffic Priorities; each of which will be assigned to its
4515 * own TX Queue in order to prevent Head-Of-Line Blocking.
4517 if (adap->params.nports * 8 > MAX_ETH_QSETS) {
4518 dev_err(adap->pdev_dev, "MAX_ETH_QSETS=%d < %d!\n",
4519 MAX_ETH_QSETS, adap->params.nports * 8);
4523 for_each_port(adap, i) {
4524 struct port_info *pi = adap2pinfo(adap, i);
4526 pi->first_qset = qidx;
4527 pi->nqsets = is_kdump_kernel() ? 1 : 8;
4530 #else /* !CONFIG_CHELSIO_T4_DCB */
4532 * We default to 1 queue per non-10G port and up to # of cores queues
4536 q10g = (MAX_ETH_QSETS - (adap->params.nports - n10g)) / n10g;
4537 if (q10g > netif_get_num_default_rss_queues())
4538 q10g = netif_get_num_default_rss_queues();
4540 if (is_kdump_kernel())
4543 for_each_port(adap, i) {
4544 struct port_info *pi = adap2pinfo(adap, i);
4546 pi->first_qset = qidx;
4547 pi->nqsets = is_x_10g_port(&pi->link_cfg) ? q10g : 1;
4550 #endif /* !CONFIG_CHELSIO_T4_DCB */
4553 s->max_ethqsets = qidx; /* MSI-X may lower it later */
4557 * For offload we use 1 queue/channel if all ports are up to 1G,
4558 * otherwise we divide all available queues amongst the channels
4559 * capped by the number of available cores.
4562 i = min_t(int, MAX_OFLD_QSETS, num_online_cpus());
4563 s->ofldqsets = roundup(i, adap->params.nports);
4565 s->ofldqsets = adap->params.nports;
4569 for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
4570 struct sge_eth_rxq *r = &s->ethrxq[i];
4572 init_rspq(adap, &r->rspq, 5, 10, 1024, 64);
4576 for (i = 0; i < ARRAY_SIZE(s->ethtxq); i++)
4577 s->ethtxq[i].q.size = 1024;
4579 for (i = 0; i < ARRAY_SIZE(s->ctrlq); i++)
4580 s->ctrlq[i].q.size = 512;
4582 if (!is_t4(adap->params.chip))
4583 s->ptptxq.q.size = 8;
4585 init_rspq(adap, &s->fw_evtq, 0, 1, 1024, 64);
4586 init_rspq(adap, &s->intrq, 0, 1, 512, 64);
4590 * Reduce the number of Ethernet queues across all ports to at most n.
4591 * n provides at least one queue per port.
4593 static void reduce_ethqs(struct adapter *adap, int n)
4596 struct port_info *pi;
4598 while (n < adap->sge.ethqsets)
4599 for_each_port(adap, i) {
4600 pi = adap2pinfo(adap, i);
4601 if (pi->nqsets > 1) {
4603 adap->sge.ethqsets--;
4604 if (adap->sge.ethqsets <= n)
4610 for_each_port(adap, i) {
4611 pi = adap2pinfo(adap, i);
4617 static int get_msix_info(struct adapter *adap)
4619 struct uld_msix_info *msix_info;
4620 unsigned int max_ingq = 0;
4622 if (is_offload(adap))
4623 max_ingq += MAX_OFLD_QSETS * adap->num_ofld_uld;
4624 if (is_pci_uld(adap))
4625 max_ingq += MAX_OFLD_QSETS * adap->num_uld;
4630 msix_info = kcalloc(max_ingq, sizeof(*msix_info), GFP_KERNEL);
4634 adap->msix_bmap_ulds.msix_bmap = kcalloc(BITS_TO_LONGS(max_ingq),
4635 sizeof(long), GFP_KERNEL);
4636 if (!adap->msix_bmap_ulds.msix_bmap) {
4640 spin_lock_init(&adap->msix_bmap_ulds.lock);
4641 adap->msix_info_ulds = msix_info;
4646 static void free_msix_info(struct adapter *adap)
4648 if (!(adap->num_uld && adap->num_ofld_uld))
4651 kfree(adap->msix_info_ulds);
4652 kfree(adap->msix_bmap_ulds.msix_bmap);
4655 /* 2 MSI-X vectors needed for the FW queue and non-data interrupts */
4656 #define EXTRA_VECS 2
4658 static int enable_msix(struct adapter *adap)
4660 int ofld_need = 0, uld_need = 0;
4661 int i, j, want, need, allocated;
4662 struct sge *s = &adap->sge;
4663 unsigned int nchan = adap->params.nports;
4664 struct msix_entry *entries;
4665 int max_ingq = MAX_INGQ;
4667 if (is_pci_uld(adap))
4668 max_ingq += (MAX_OFLD_QSETS * adap->num_uld);
4669 if (is_offload(adap))
4670 max_ingq += (MAX_OFLD_QSETS * adap->num_ofld_uld);
4671 entries = kmalloc(sizeof(*entries) * (max_ingq + 1),
4677 if (get_msix_info(adap)) {
4678 adap->params.offload = 0;
4679 adap->params.crypto = 0;
4682 for (i = 0; i < max_ingq + 1; ++i)
4683 entries[i].entry = i;
4685 want = s->max_ethqsets + EXTRA_VECS;
4686 if (is_offload(adap)) {
4687 want += adap->num_ofld_uld * s->ofldqsets;
4688 ofld_need = adap->num_ofld_uld * nchan;
4690 if (is_pci_uld(adap)) {
4691 want += adap->num_uld * s->ofldqsets;
4692 uld_need = adap->num_uld * nchan;
4694 #ifdef CONFIG_CHELSIO_T4_DCB
4695 /* For Data Center Bridging we need 8 Ethernet TX Priority Queues for
4698 need = 8 * adap->params.nports + EXTRA_VECS + ofld_need + uld_need;
4700 need = adap->params.nports + EXTRA_VECS + ofld_need + uld_need;
4702 allocated = pci_enable_msix_range(adap->pdev, entries, need, want);
4703 if (allocated < 0) {
4704 dev_info(adap->pdev_dev, "not enough MSI-X vectors left,"
4705 " not using MSI-X\n");
4710 /* Distribute available vectors to the various queue groups.
4711 * Every group gets its minimum requirement and NIC gets top
4712 * priority for leftovers.
4714 i = allocated - EXTRA_VECS - ofld_need - uld_need;
4715 if (i < s->max_ethqsets) {
4716 s->max_ethqsets = i;
4717 if (i < s->ethqsets)
4718 reduce_ethqs(adap, i);
4721 if (allocated < want)
4722 s->nqs_per_uld = nchan;
4724 s->nqs_per_uld = s->ofldqsets;
4727 for (i = 0; i < (s->max_ethqsets + EXTRA_VECS); ++i)
4728 adap->msix_info[i].vec = entries[i].vector;
4730 for (j = 0 ; i < allocated; ++i, j++) {
4731 adap->msix_info_ulds[j].vec = entries[i].vector;
4732 adap->msix_info_ulds[j].idx = i;
4734 adap->msix_bmap_ulds.mapsize = j;
4736 dev_info(adap->pdev_dev, "%d MSI-X vectors allocated, "
4737 "nic %d per uld %d\n",
4738 allocated, s->max_ethqsets, s->nqs_per_uld);
4746 static int init_rss(struct adapter *adap)
4751 err = t4_init_rss_mode(adap, adap->mbox);
4755 for_each_port(adap, i) {
4756 struct port_info *pi = adap2pinfo(adap, i);
4758 pi->rss = kcalloc(pi->rss_size, sizeof(u16), GFP_KERNEL);
4765 static int cxgb4_get_pcie_dev_link_caps(struct adapter *adap,
4766 enum pci_bus_speed *speed,
4767 enum pcie_link_width *width)
4769 u32 lnkcap1, lnkcap2;
4772 #define PCIE_MLW_CAP_SHIFT 4 /* start of MLW mask in link capabilities */
4774 *speed = PCI_SPEED_UNKNOWN;
4775 *width = PCIE_LNK_WIDTH_UNKNOWN;
4777 err1 = pcie_capability_read_dword(adap->pdev, PCI_EXP_LNKCAP,
4779 err2 = pcie_capability_read_dword(adap->pdev, PCI_EXP_LNKCAP2,
4781 if (!err2 && lnkcap2) { /* PCIe r3.0-compliant */
4782 if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_8_0GB)
4783 *speed = PCIE_SPEED_8_0GT;
4784 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_5_0GB)
4785 *speed = PCIE_SPEED_5_0GT;
4786 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_2_5GB)
4787 *speed = PCIE_SPEED_2_5GT;
4790 *width = (lnkcap1 & PCI_EXP_LNKCAP_MLW) >> PCIE_MLW_CAP_SHIFT;
4791 if (!lnkcap2) { /* pre-r3.0 */
4792 if (lnkcap1 & PCI_EXP_LNKCAP_SLS_5_0GB)
4793 *speed = PCIE_SPEED_5_0GT;
4794 else if (lnkcap1 & PCI_EXP_LNKCAP_SLS_2_5GB)
4795 *speed = PCIE_SPEED_2_5GT;
4799 if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN)
4800 return err1 ? err1 : err2 ? err2 : -EINVAL;
4804 static void cxgb4_check_pcie_caps(struct adapter *adap)
4806 enum pcie_link_width width, width_cap;
4807 enum pci_bus_speed speed, speed_cap;
4809 #define PCIE_SPEED_STR(speed) \
4810 (speed == PCIE_SPEED_8_0GT ? "8.0GT/s" : \
4811 speed == PCIE_SPEED_5_0GT ? "5.0GT/s" : \
4812 speed == PCIE_SPEED_2_5GT ? "2.5GT/s" : \
4815 if (cxgb4_get_pcie_dev_link_caps(adap, &speed_cap, &width_cap)) {
4816 dev_warn(adap->pdev_dev,
4817 "Unable to determine PCIe device BW capabilities\n");
4821 if (pcie_get_minimum_link(adap->pdev, &speed, &width) ||
4822 speed == PCI_SPEED_UNKNOWN || width == PCIE_LNK_WIDTH_UNKNOWN) {
4823 dev_warn(adap->pdev_dev,
4824 "Unable to determine PCI Express bandwidth.\n");
4828 dev_info(adap->pdev_dev, "PCIe link speed is %s, device supports %s\n",
4829 PCIE_SPEED_STR(speed), PCIE_SPEED_STR(speed_cap));
4830 dev_info(adap->pdev_dev, "PCIe link width is x%d, device supports x%d\n",
4832 if (speed < speed_cap || width < width_cap)
4833 dev_info(adap->pdev_dev,
4834 "A slot with more lanes and/or higher speed is "
4835 "suggested for optimal performance.\n");
4838 /* Dump basic information about the adapter */
4839 static void print_adapter_info(struct adapter *adapter)
4841 /* Hardware/Firmware/etc. Version/Revision IDs */
4842 t4_dump_version_info(adapter);
4844 /* Software/Hardware configuration */
4845 dev_info(adapter->pdev_dev, "Configuration: %sNIC %s, %s capable\n",
4846 is_offload(adapter) ? "R" : "",
4847 ((adapter->flags & USING_MSIX) ? "MSI-X" :
4848 (adapter->flags & USING_MSI) ? "MSI" : ""),
4849 is_offload(adapter) ? "Offload" : "non-Offload");
4852 static void print_port_info(const struct net_device *dev)
4856 const char *spd = "";
4857 const struct port_info *pi = netdev_priv(dev);
4858 const struct adapter *adap = pi->adapter;
4860 if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_2_5GB)
4862 else if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_5_0GB)
4864 else if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_8_0GB)
4867 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_100M)
4868 bufp += sprintf(bufp, "100M/");
4869 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_1G)
4870 bufp += sprintf(bufp, "1G/");
4871 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_10G)
4872 bufp += sprintf(bufp, "10G/");
4873 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_25G)
4874 bufp += sprintf(bufp, "25G/");
4875 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_40G)
4876 bufp += sprintf(bufp, "40G/");
4877 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_50G)
4878 bufp += sprintf(bufp, "50G/");
4879 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_100G)
4880 bufp += sprintf(bufp, "100G/");
4881 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_200G)
4882 bufp += sprintf(bufp, "200G/");
4883 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_400G)
4884 bufp += sprintf(bufp, "400G/");
4887 sprintf(bufp, "BASE-%s", t4_get_port_type_description(pi->port_type));
4889 netdev_info(dev, "%s: Chelsio %s (%s) %s\n",
4890 dev->name, adap->params.vpd.id, adap->name, buf);
4894 * Free the following resources:
4895 * - memory used for tables
4898 * - resources FW is holding for us
4900 static void free_some_resources(struct adapter *adapter)
4904 kvfree(adapter->smt);
4905 kvfree(adapter->l2t);
4906 t4_cleanup_sched(adapter);
4907 kvfree(adapter->tids.tid_tab);
4908 cxgb4_cleanup_tc_flower(adapter);
4909 cxgb4_cleanup_tc_u32(adapter);
4910 kfree(adapter->sge.egr_map);
4911 kfree(adapter->sge.ingr_map);
4912 kfree(adapter->sge.starving_fl);
4913 kfree(adapter->sge.txq_maperr);
4914 #ifdef CONFIG_DEBUG_FS
4915 kfree(adapter->sge.blocked_fl);
4917 disable_msi(adapter);
4919 for_each_port(adapter, i)
4920 if (adapter->port[i]) {
4921 struct port_info *pi = adap2pinfo(adapter, i);
4924 t4_free_vi(adapter, adapter->mbox, adapter->pf,
4926 kfree(adap2pinfo(adapter, i)->rss);
4927 free_netdev(adapter->port[i]);
4929 if (adapter->flags & FW_OK)
4930 t4_fw_bye(adapter, adapter->pf);
4933 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
4934 #define VLAN_FEAT (NETIF_F_SG | NETIF_F_IP_CSUM | TSO_FLAGS | \
4935 NETIF_F_IPV6_CSUM | NETIF_F_HIGHDMA)
4936 #define SEGMENT_SIZE 128
4938 static int get_chip_type(struct pci_dev *pdev, u32 pl_rev)
4942 /* Retrieve adapter's device ID */
4943 pci_read_config_word(pdev, PCI_DEVICE_ID, &device_id);
4945 switch (device_id >> 12) {
4947 return CHELSIO_CHIP_CODE(CHELSIO_T4, pl_rev);
4949 return CHELSIO_CHIP_CODE(CHELSIO_T5, pl_rev);
4951 return CHELSIO_CHIP_CODE(CHELSIO_T6, pl_rev);
4953 dev_err(&pdev->dev, "Device %d is not supported\n",
4959 #ifdef CONFIG_PCI_IOV
4960 static void cxgb4_mgmt_setup(struct net_device *dev)
4962 dev->type = ARPHRD_NONE;
4964 dev->hard_header_len = 0;
4966 dev->tx_queue_len = 0;
4967 dev->flags |= IFF_NOARP;
4968 dev->priv_flags |= IFF_NO_QUEUE;
4970 /* Initialize the device structure. */
4971 dev->netdev_ops = &cxgb4_mgmt_netdev_ops;
4972 dev->ethtool_ops = &cxgb4_mgmt_ethtool_ops;
4975 static int cxgb4_iov_configure(struct pci_dev *pdev, int num_vfs)
4977 struct adapter *adap = pci_get_drvdata(pdev);
4979 int current_vfs = pci_num_vf(pdev);
4982 pcie_fw = readl(adap->regs + PCIE_FW_A);
4983 /* Check if cxgb4 is the MASTER and fw is initialized */
4985 (!(pcie_fw & PCIE_FW_INIT_F) ||
4986 !(pcie_fw & PCIE_FW_MASTER_VLD_F) ||
4987 PCIE_FW_MASTER_G(pcie_fw) != CXGB4_UNIFIED_PF)) {
4988 dev_warn(&pdev->dev,
4989 "cxgb4 driver needs to be MASTER to support SRIOV\n");
4993 /* If any of the VF's is already assigned to Guest OS, then
4994 * SRIOV for the same cannot be modified
4996 if (current_vfs && pci_vfs_assigned(pdev)) {
4998 "Cannot modify SR-IOV while VFs are assigned\n");
5001 /* Note that the upper-level code ensures that we're never called with
5002 * a non-zero "num_vfs" when we already have VFs instantiated. But
5003 * it never hurts to code defensively.
5005 if (num_vfs != 0 && current_vfs != 0)
5008 /* Nothing to do for no change. */
5009 if (num_vfs == current_vfs)
5012 /* Disable SRIOV when zero is passed. */
5014 pci_disable_sriov(pdev);
5015 /* free VF Management Interface */
5016 unregister_netdev(adap->port[0]);
5017 free_netdev(adap->port[0]);
5018 adap->port[0] = NULL;
5020 /* free VF resources */
5022 kfree(adap->vfinfo);
5023 adap->vfinfo = NULL;
5028 struct fw_pfvf_cmd port_cmd, port_rpl;
5029 struct net_device *netdev;
5030 unsigned int pmask, port;
5031 struct pci_dev *pbridge;
5032 struct port_info *pi;
5033 char name[IFNAMSIZ];
5038 /* If we want to instantiate Virtual Functions, then our
5039 * parent bridge's PCI-E needs to support Alternative Routing
5040 * ID (ARI) because our VFs will show up at function offset 8
5043 pbridge = pdev->bus->self;
5044 pos = pci_find_capability(pbridge, PCI_CAP_ID_EXP);
5045 pci_read_config_word(pbridge, pos + PCI_EXP_FLAGS, &flags);
5046 pci_read_config_dword(pbridge, pos + PCI_EXP_DEVCAP2, &devcap2);
5048 if ((flags & PCI_EXP_FLAGS_VERS) < 2 ||
5049 !(devcap2 & PCI_EXP_DEVCAP2_ARI)) {
5050 /* Our parent bridge does not support ARI so issue a
5051 * warning and skip instantiating the VFs. They
5052 * won't be reachable.
5054 dev_warn(&pdev->dev, "Parent bridge %02x:%02x.%x doesn't support ARI; can't instantiate Virtual Functions\n",
5055 pbridge->bus->number, PCI_SLOT(pbridge->devfn),
5056 PCI_FUNC(pbridge->devfn));
5059 memset(&port_cmd, 0, sizeof(port_cmd));
5060 port_cmd.op_to_vfn = cpu_to_be32(FW_CMD_OP_V(FW_PFVF_CMD) |
5063 FW_PFVF_CMD_PFN_V(adap->pf) |
5064 FW_PFVF_CMD_VFN_V(0));
5065 port_cmd.retval_len16 = cpu_to_be32(FW_LEN16(port_cmd));
5066 err = t4_wr_mbox(adap, adap->mbox, &port_cmd, sizeof(port_cmd),
5070 pmask = FW_PFVF_CMD_PMASK_G(be32_to_cpu(port_rpl.type_to_neq));
5071 port = ffs(pmask) - 1;
5072 /* Allocate VF Management Interface. */
5073 snprintf(name, IFNAMSIZ, "mgmtpf%d,%d", adap->adap_idx,
5075 netdev = alloc_netdev(sizeof(struct port_info),
5076 name, NET_NAME_UNKNOWN, cxgb4_mgmt_setup);
5080 pi = netdev_priv(netdev);
5084 SET_NETDEV_DEV(netdev, &pdev->dev);
5086 adap->port[0] = netdev;
5089 err = register_netdev(adap->port[0]);
5091 pr_info("Unable to register VF mgmt netdev %s\n", name);
5092 free_netdev(adap->port[0]);
5093 adap->port[0] = NULL;
5096 /* Allocate and set up VF Information. */
5097 adap->vfinfo = kcalloc(pci_sriov_get_totalvfs(pdev),
5098 sizeof(struct vf_info), GFP_KERNEL);
5099 if (!adap->vfinfo) {
5100 unregister_netdev(adap->port[0]);
5101 free_netdev(adap->port[0]);
5102 adap->port[0] = NULL;
5105 cxgb4_mgmt_fill_vf_station_mac_addr(adap);
5107 /* Instantiate the requested number of VFs. */
5108 err = pci_enable_sriov(pdev, num_vfs);
5110 pr_info("Unable to instantiate %d VFs\n", num_vfs);
5112 unregister_netdev(adap->port[0]);
5113 free_netdev(adap->port[0]);
5114 adap->port[0] = NULL;
5115 kfree(adap->vfinfo);
5116 adap->vfinfo = NULL;
5121 adap->num_vfs = num_vfs;
5124 #endif /* CONFIG_PCI_IOV */
5126 static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
5128 int func, i, err, s_qpp, qpp, num_seg;
5129 struct port_info *pi;
5130 bool highdma = false;
5131 struct adapter *adapter = NULL;
5132 struct net_device *netdev;
5135 enum chip_type chip;
5136 static int adap_idx = 1;
5138 printk_once(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
5140 err = pci_request_regions(pdev, KBUILD_MODNAME);
5142 /* Just info, some other driver may have claimed the device. */
5143 dev_info(&pdev->dev, "cannot obtain PCI resources\n");
5147 err = pci_enable_device(pdev);
5149 dev_err(&pdev->dev, "cannot enable PCI device\n");
5150 goto out_release_regions;
5153 regs = pci_ioremap_bar(pdev, 0);
5155 dev_err(&pdev->dev, "cannot map device registers\n");
5157 goto out_disable_device;
5160 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
5163 goto out_unmap_bar0;
5166 adapter->regs = regs;
5167 err = t4_wait_dev_ready(regs);
5169 goto out_free_adapter;
5171 /* We control everything through one PF */
5172 whoami = readl(regs + PL_WHOAMI_A);
5173 pl_rev = REV_G(readl(regs + PL_REV_A));
5174 chip = get_chip_type(pdev, pl_rev);
5175 func = CHELSIO_CHIP_VERSION(chip) <= CHELSIO_T5 ?
5176 SOURCEPF_G(whoami) : T6_SOURCEPF_G(whoami);
5178 adapter->pdev = pdev;
5179 adapter->pdev_dev = &pdev->dev;
5180 adapter->name = pci_name(pdev);
5181 adapter->mbox = func;
5183 adapter->params.chip = chip;
5184 adapter->adap_idx = adap_idx;
5185 adapter->msg_enable = DFLT_MSG_ENABLE;
5186 adapter->mbox_log = kzalloc(sizeof(*adapter->mbox_log) +
5187 (sizeof(struct mbox_cmd) *
5188 T4_OS_LOG_MBOX_CMDS),
5190 if (!adapter->mbox_log) {
5192 goto out_free_adapter;
5194 spin_lock_init(&adapter->mbox_lock);
5195 INIT_LIST_HEAD(&adapter->mlist.list);
5196 pci_set_drvdata(pdev, adapter);
5198 if (func != ent->driver_data) {
5199 pci_disable_device(pdev);
5200 pci_save_state(pdev); /* to restore SR-IOV later */
5204 if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
5206 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
5208 dev_err(&pdev->dev, "unable to obtain 64-bit DMA for "
5209 "coherent allocations\n");
5210 goto out_free_adapter;
5213 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
5215 dev_err(&pdev->dev, "no usable DMA configuration\n");
5216 goto out_free_adapter;
5220 pci_enable_pcie_error_reporting(pdev);
5221 pci_set_master(pdev);
5222 pci_save_state(pdev);
5224 adapter->workq = create_singlethread_workqueue("cxgb4");
5225 if (!adapter->workq) {
5227 goto out_free_adapter;
5230 adapter->mbox_log->size = T4_OS_LOG_MBOX_CMDS;
5232 /* PCI device has been enabled */
5233 adapter->flags |= DEV_ENABLED;
5234 memset(adapter->chan_map, 0xff, sizeof(adapter->chan_map));
5236 /* If possible, we use PCIe Relaxed Ordering Attribute to deliver
5237 * Ingress Packet Data to Free List Buffers in order to allow for
5238 * chipset performance optimizations between the Root Complex and
5239 * Memory Controllers. (Messages to the associated Ingress Queue
5240 * notifying new Packet Placement in the Free Lists Buffers will be
5241 * send without the Relaxed Ordering Attribute thus guaranteeing that
5242 * all preceding PCIe Transaction Layer Packets will be processed
5243 * first.) But some Root Complexes have various issues with Upstream
5244 * Transaction Layer Packets with the Relaxed Ordering Attribute set.
5245 * The PCIe devices which under the Root Complexes will be cleared the
5246 * Relaxed Ordering bit in the configuration space, So we check our
5247 * PCIe configuration space to see if it's flagged with advice against
5248 * using Relaxed Ordering.
5250 if (!pcie_relaxed_ordering_enabled(pdev))
5251 adapter->flags |= ROOT_NO_RELAXED_ORDERING;
5253 spin_lock_init(&adapter->stats_lock);
5254 spin_lock_init(&adapter->tid_release_lock);
5255 spin_lock_init(&adapter->win0_lock);
5257 INIT_WORK(&adapter->tid_release_task, process_tid_release_list);
5258 INIT_WORK(&adapter->db_full_task, process_db_full);
5259 INIT_WORK(&adapter->db_drop_task, process_db_drop);
5261 err = t4_prep_adapter(adapter);
5263 goto out_free_adapter;
5266 if (!is_t4(adapter->params.chip)) {
5267 s_qpp = (QUEUESPERPAGEPF0_S +
5268 (QUEUESPERPAGEPF1_S - QUEUESPERPAGEPF0_S) *
5270 qpp = 1 << QUEUESPERPAGEPF0_G(t4_read_reg(adapter,
5271 SGE_EGRESS_QUEUES_PER_PAGE_PF_A) >> s_qpp);
5272 num_seg = PAGE_SIZE / SEGMENT_SIZE;
5274 /* Each segment size is 128B. Write coalescing is enabled only
5275 * when SGE_EGRESS_QUEUES_PER_PAGE_PF reg value for the
5276 * queue is less no of segments that can be accommodated in
5279 if (qpp > num_seg) {
5281 "Incorrect number of egress queues per page\n");
5283 goto out_free_adapter;
5285 adapter->bar2 = ioremap_wc(pci_resource_start(pdev, 2),
5286 pci_resource_len(pdev, 2));
5287 if (!adapter->bar2) {
5288 dev_err(&pdev->dev, "cannot map device bar2 region\n");
5290 goto out_free_adapter;
5294 setup_memwin(adapter);
5295 err = adap_init0(adapter);
5296 #ifdef CONFIG_DEBUG_FS
5297 bitmap_zero(adapter->sge.blocked_fl, adapter->sge.egr_sz);
5299 setup_memwin_rdma(adapter);
5303 /* configure SGE_STAT_CFG_A to read WC stats */
5304 if (!is_t4(adapter->params.chip))
5305 t4_write_reg(adapter, SGE_STAT_CFG_A, STATSOURCE_T5_V(7) |
5306 (is_t5(adapter->params.chip) ? STATMODE_V(0) :
5309 for_each_port(adapter, i) {
5310 netdev = alloc_etherdev_mq(sizeof(struct port_info),
5317 SET_NETDEV_DEV(netdev, &pdev->dev);
5319 adapter->port[i] = netdev;
5320 pi = netdev_priv(netdev);
5321 pi->adapter = adapter;
5322 pi->xact_addr_filt = -1;
5324 netdev->irq = pdev->irq;
5326 netdev->hw_features = NETIF_F_SG | TSO_FLAGS |
5327 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
5328 NETIF_F_RXCSUM | NETIF_F_RXHASH |
5329 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
5332 if (CHELSIO_CHIP_VERSION(chip) > CHELSIO_T5)
5333 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
5336 netdev->hw_features |= NETIF_F_HIGHDMA;
5337 netdev->features |= netdev->hw_features;
5338 netdev->vlan_features = netdev->features & VLAN_FEAT;
5340 netdev->priv_flags |= IFF_UNICAST_FLT;
5342 /* MTU range: 81 - 9600 */
5343 netdev->min_mtu = 81; /* accommodate SACK */
5344 netdev->max_mtu = MAX_MTU;
5346 netdev->netdev_ops = &cxgb4_netdev_ops;
5347 #ifdef CONFIG_CHELSIO_T4_DCB
5348 netdev->dcbnl_ops = &cxgb4_dcb_ops;
5349 cxgb4_dcb_state_init(netdev);
5351 cxgb4_set_ethtool_ops(netdev);
5354 cxgb4_init_ethtool_dump(adapter);
5356 pci_set_drvdata(pdev, adapter);
5358 if (adapter->flags & FW_OK) {
5359 err = t4_port_init(adapter, func, func, 0);
5362 } else if (adapter->params.nports == 1) {
5363 /* If we don't have a connection to the firmware -- possibly
5364 * because of an error -- grab the raw VPD parameters so we
5365 * can set the proper MAC Address on the debug network
5366 * interface that we've created.
5368 u8 hw_addr[ETH_ALEN];
5369 u8 *na = adapter->params.vpd.na;
5371 err = t4_get_raw_vpd_params(adapter, &adapter->params.vpd);
5373 for (i = 0; i < ETH_ALEN; i++)
5374 hw_addr[i] = (hex2val(na[2 * i + 0]) * 16 +
5375 hex2val(na[2 * i + 1]));
5376 t4_set_hw_addr(adapter, 0, hw_addr);
5380 /* Configure queues and allocate tables now, they can be needed as
5381 * soon as the first register_netdev completes.
5383 cfg_queues(adapter);
5385 adapter->smt = t4_init_smt();
5386 if (!adapter->smt) {
5387 /* We tolerate a lack of SMT, giving up some functionality */
5388 dev_warn(&pdev->dev, "could not allocate SMT, continuing\n");
5391 adapter->l2t = t4_init_l2t(adapter->l2t_start, adapter->l2t_end);
5392 if (!adapter->l2t) {
5393 /* We tolerate a lack of L2T, giving up some functionality */
5394 dev_warn(&pdev->dev, "could not allocate L2T, continuing\n");
5395 adapter->params.offload = 0;
5398 #if IS_ENABLED(CONFIG_IPV6)
5399 if ((CHELSIO_CHIP_VERSION(adapter->params.chip) <= CHELSIO_T5) &&
5400 (!(t4_read_reg(adapter, LE_DB_CONFIG_A) & ASLIPCOMPEN_F))) {
5401 /* CLIP functionality is not present in hardware,
5402 * hence disable all offload features
5404 dev_warn(&pdev->dev,
5405 "CLIP not enabled in hardware, continuing\n");
5406 adapter->params.offload = 0;
5408 adapter->clipt = t4_init_clip_tbl(adapter->clipt_start,
5409 adapter->clipt_end);
5410 if (!adapter->clipt) {
5411 /* We tolerate a lack of clip_table, giving up
5412 * some functionality
5414 dev_warn(&pdev->dev,
5415 "could not allocate Clip table, continuing\n");
5416 adapter->params.offload = 0;
5421 for_each_port(adapter, i) {
5422 pi = adap2pinfo(adapter, i);
5423 pi->sched_tbl = t4_init_sched(adapter->params.nsched_cls);
5425 dev_warn(&pdev->dev,
5426 "could not activate scheduling on port %d\n",
5430 if (tid_init(&adapter->tids) < 0) {
5431 dev_warn(&pdev->dev, "could not allocate TID table, "
5433 adapter->params.offload = 0;
5435 adapter->tc_u32 = cxgb4_init_tc_u32(adapter);
5436 if (!adapter->tc_u32)
5437 dev_warn(&pdev->dev,
5438 "could not offload tc u32, continuing\n");
5440 if (cxgb4_init_tc_flower(adapter))
5441 dev_warn(&pdev->dev,
5442 "could not offload tc flower, continuing\n");
5445 if (is_offload(adapter) || is_hashfilter(adapter)) {
5446 if (t4_read_reg(adapter, LE_DB_CONFIG_A) & HASHEN_F) {
5447 u32 hash_base, hash_reg;
5449 if (chip <= CHELSIO_T5) {
5450 hash_reg = LE_DB_TID_HASHBASE_A;
5451 hash_base = t4_read_reg(adapter, hash_reg);
5452 adapter->tids.hash_base = hash_base / 4;
5454 hash_reg = T6_LE_DB_HASH_TID_BASE_A;
5455 hash_base = t4_read_reg(adapter, hash_reg);
5456 adapter->tids.hash_base = hash_base;
5461 /* See what interrupts we'll be using */
5462 if (msi > 1 && enable_msix(adapter) == 0)
5463 adapter->flags |= USING_MSIX;
5464 else if (msi > 0 && pci_enable_msi(pdev) == 0) {
5465 adapter->flags |= USING_MSI;
5467 free_msix_info(adapter);
5470 /* check for PCI Express bandwidth capabiltites */
5471 cxgb4_check_pcie_caps(adapter);
5473 err = init_rss(adapter);
5478 * The card is now ready to go. If any errors occur during device
5479 * registration we do not fail the whole card but rather proceed only
5480 * with the ports we manage to register successfully. However we must
5481 * register at least one net device.
5483 for_each_port(adapter, i) {
5484 pi = adap2pinfo(adapter, i);
5485 adapter->port[i]->dev_port = pi->lport;
5486 netif_set_real_num_tx_queues(adapter->port[i], pi->nqsets);
5487 netif_set_real_num_rx_queues(adapter->port[i], pi->nqsets);
5489 netif_carrier_off(adapter->port[i]);
5491 err = register_netdev(adapter->port[i]);
5494 adapter->chan_map[pi->tx_chan] = i;
5495 print_port_info(adapter->port[i]);
5498 dev_err(&pdev->dev, "could not register any net devices\n");
5502 dev_warn(&pdev->dev, "only %d net devices registered\n", i);
5506 if (cxgb4_debugfs_root) {
5507 adapter->debugfs_root = debugfs_create_dir(pci_name(pdev),
5508 cxgb4_debugfs_root);
5509 setup_debugfs(adapter);
5512 /* PCIe EEH recovery on powerpc platforms needs fundamental reset */
5513 pdev->needs_freset = 1;
5515 if (is_uld(adapter)) {
5516 mutex_lock(&uld_mutex);
5517 list_add_tail(&adapter->list_node, &adapter_list);
5518 mutex_unlock(&uld_mutex);
5521 if (!is_t4(adapter->params.chip))
5522 cxgb4_ptp_init(adapter);
5524 print_adapter_info(adapter);
5525 setup_fw_sge_queues(adapter);
5529 free_some_resources(adapter);
5530 if (adapter->flags & USING_MSIX)
5531 free_msix_info(adapter);
5532 if (adapter->num_uld || adapter->num_ofld_uld)
5533 t4_uld_mem_free(adapter);
5535 if (!is_t4(adapter->params.chip))
5536 iounmap(adapter->bar2);
5539 destroy_workqueue(adapter->workq);
5541 kfree(adapter->mbox_log);
5546 pci_disable_pcie_error_reporting(pdev);
5547 pci_disable_device(pdev);
5548 out_release_regions:
5549 pci_release_regions(pdev);
5553 static void remove_one(struct pci_dev *pdev)
5555 struct adapter *adapter = pci_get_drvdata(pdev);
5558 pci_release_regions(pdev);
5562 adapter->flags |= SHUTTING_DOWN;
5564 if (adapter->pf == 4) {
5567 /* Tear down per-adapter Work Queue first since it can contain
5568 * references to our adapter data structure.
5570 destroy_workqueue(adapter->workq);
5572 if (is_uld(adapter)) {
5573 detach_ulds(adapter);
5574 t4_uld_clean_up(adapter);
5577 disable_interrupts(adapter);
5579 for_each_port(adapter, i)
5580 if (adapter->port[i]->reg_state == NETREG_REGISTERED)
5581 unregister_netdev(adapter->port[i]);
5583 debugfs_remove_recursive(adapter->debugfs_root);
5585 if (!is_t4(adapter->params.chip))
5586 cxgb4_ptp_stop(adapter);
5588 /* If we allocated filters, free up state associated with any
5591 clear_all_filters(adapter);
5593 if (adapter->flags & FULL_INIT_DONE)
5596 if (adapter->flags & USING_MSIX)
5597 free_msix_info(adapter);
5598 if (adapter->num_uld || adapter->num_ofld_uld)
5599 t4_uld_mem_free(adapter);
5600 free_some_resources(adapter);
5601 #if IS_ENABLED(CONFIG_IPV6)
5602 t4_cleanup_clip_tbl(adapter);
5604 if (!is_t4(adapter->params.chip))
5605 iounmap(adapter->bar2);
5607 #ifdef CONFIG_PCI_IOV
5609 cxgb4_iov_configure(adapter->pdev, 0);
5612 iounmap(adapter->regs);
5613 pci_disable_pcie_error_reporting(pdev);
5614 if ((adapter->flags & DEV_ENABLED)) {
5615 pci_disable_device(pdev);
5616 adapter->flags &= ~DEV_ENABLED;
5618 pci_release_regions(pdev);
5619 kfree(adapter->mbox_log);
5624 /* "Shutdown" quiesces the device, stopping Ingress Packet and Interrupt
5625 * delivery. This is essentially a stripped down version of the PCI remove()
5626 * function where we do the minimal amount of work necessary to shutdown any
5629 static void shutdown_one(struct pci_dev *pdev)
5631 struct adapter *adapter = pci_get_drvdata(pdev);
5633 /* As with remove_one() above (see extended comment), we only want do
5634 * do cleanup on PCI Devices which went all the way through init_one()
5638 pci_release_regions(pdev);
5642 adapter->flags |= SHUTTING_DOWN;
5644 if (adapter->pf == 4) {
5647 for_each_port(adapter, i)
5648 if (adapter->port[i]->reg_state == NETREG_REGISTERED)
5649 cxgb_close(adapter->port[i]);
5651 if (is_uld(adapter)) {
5652 detach_ulds(adapter);
5653 t4_uld_clean_up(adapter);
5656 disable_interrupts(adapter);
5657 disable_msi(adapter);
5659 t4_sge_stop(adapter);
5660 if (adapter->flags & FW_OK)
5661 t4_fw_bye(adapter, adapter->mbox);
5665 static struct pci_driver cxgb4_driver = {
5666 .name = KBUILD_MODNAME,
5667 .id_table = cxgb4_pci_tbl,
5669 .remove = remove_one,
5670 .shutdown = shutdown_one,
5671 #ifdef CONFIG_PCI_IOV
5672 .sriov_configure = cxgb4_iov_configure,
5674 .err_handler = &cxgb4_eeh,
5677 static int __init cxgb4_init_module(void)
5681 /* Debugfs support is optional, just warn if this fails */
5682 cxgb4_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
5683 if (!cxgb4_debugfs_root)
5684 pr_warn("could not create debugfs entry, continuing\n");
5686 ret = pci_register_driver(&cxgb4_driver);
5688 debugfs_remove(cxgb4_debugfs_root);
5690 #if IS_ENABLED(CONFIG_IPV6)
5691 if (!inet6addr_registered) {
5692 register_inet6addr_notifier(&cxgb4_inet6addr_notifier);
5693 inet6addr_registered = true;
5700 static void __exit cxgb4_cleanup_module(void)
5702 #if IS_ENABLED(CONFIG_IPV6)
5703 if (inet6addr_registered) {
5704 unregister_inet6addr_notifier(&cxgb4_inet6addr_notifier);
5705 inet6addr_registered = false;
5708 pci_unregister_driver(&cxgb4_driver);
5709 debugfs_remove(cxgb4_debugfs_root); /* NULL ok */
5712 module_init(cxgb4_init_module);
5713 module_exit(cxgb4_cleanup_module);