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>
70 #include "cxgb4_filter.h"
72 #include "t4_values.h"
75 #include "t4fw_version.h"
76 #include "cxgb4_dcb.h"
77 #include "cxgb4_debugfs.h"
81 #include "cxgb4_tc_u32.h"
83 char cxgb4_driver_name[] = KBUILD_MODNAME;
88 #define DRV_VERSION "2.0.0-ko"
89 const char cxgb4_driver_version[] = DRV_VERSION;
90 #define DRV_DESC "Chelsio T4/T5/T6 Network Driver"
92 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
93 NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
94 NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
96 /* Macros needed to support the PCI Device ID Table ...
98 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_BEGIN \
99 static const struct pci_device_id cxgb4_pci_tbl[] = {
100 #define CH_PCI_DEVICE_ID_FUNCTION 0x4
102 /* Include PCI Device IDs for both PF4 and PF0-3 so our PCI probe() routine is
105 #define CH_PCI_DEVICE_ID_FUNCTION2 0x0
107 #define CH_PCI_ID_TABLE_ENTRY(devid) \
108 {PCI_VDEVICE(CHELSIO, (devid)), 4}
110 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_END \
114 #include "t4_pci_id_tbl.h"
116 #define FW4_FNAME "cxgb4/t4fw.bin"
117 #define FW5_FNAME "cxgb4/t5fw.bin"
118 #define FW6_FNAME "cxgb4/t6fw.bin"
119 #define FW4_CFNAME "cxgb4/t4-config.txt"
120 #define FW5_CFNAME "cxgb4/t5-config.txt"
121 #define FW6_CFNAME "cxgb4/t6-config.txt"
122 #define PHY_AQ1202_FIRMWARE "cxgb4/aq1202_fw.cld"
123 #define PHY_BCM84834_FIRMWARE "cxgb4/bcm8483.bin"
124 #define PHY_AQ1202_DEVICEID 0x4409
125 #define PHY_BCM84834_DEVICEID 0x4486
127 MODULE_DESCRIPTION(DRV_DESC);
128 MODULE_AUTHOR("Chelsio Communications");
129 MODULE_LICENSE("Dual BSD/GPL");
130 MODULE_VERSION(DRV_VERSION);
131 MODULE_DEVICE_TABLE(pci, cxgb4_pci_tbl);
132 MODULE_FIRMWARE(FW4_FNAME);
133 MODULE_FIRMWARE(FW5_FNAME);
134 MODULE_FIRMWARE(FW6_FNAME);
137 * The driver uses the best interrupt scheme available on a platform in the
138 * order MSI-X, MSI, legacy INTx interrupts. This parameter determines which
139 * of these schemes the driver may consider as follows:
141 * msi = 2: choose from among all three options
142 * msi = 1: only consider MSI and INTx interrupts
143 * msi = 0: force INTx interrupts
147 module_param(msi, int, 0644);
148 MODULE_PARM_DESC(msi, "whether to use INTx (0), MSI (1) or MSI-X (2)");
151 * Normally we tell the chip to deliver Ingress Packets into our DMA buffers
152 * offset by 2 bytes in order to have the IP headers line up on 4-byte
153 * boundaries. This is a requirement for many architectures which will throw
154 * a machine check fault if an attempt is made to access one of the 4-byte IP
155 * header fields on a non-4-byte boundary. And it's a major performance issue
156 * even on some architectures which allow it like some implementations of the
157 * x86 ISA. However, some architectures don't mind this and for some very
158 * edge-case performance sensitive applications (like forwarding large volumes
159 * of small packets), setting this DMA offset to 0 will decrease the number of
160 * PCI-E Bus transfers enough to measurably affect performance.
162 static int rx_dma_offset = 2;
164 /* TX Queue select used to determine what algorithm to use for selecting TX
165 * queue. Select between the kernel provided function (select_queue=0) or user
166 * cxgb_select_queue function (select_queue=1)
168 * Default: select_queue=0
170 static int select_queue;
171 module_param(select_queue, int, 0644);
172 MODULE_PARM_DESC(select_queue,
173 "Select between kernel provided method of selecting or driver method of selecting TX queue. Default is kernel method.");
175 static struct dentry *cxgb4_debugfs_root;
177 LIST_HEAD(adapter_list);
178 DEFINE_MUTEX(uld_mutex);
180 static void link_report(struct net_device *dev)
182 if (!netif_carrier_ok(dev))
183 netdev_info(dev, "link down\n");
185 static const char *fc[] = { "no", "Rx", "Tx", "Tx/Rx" };
188 const struct port_info *p = netdev_priv(dev);
190 switch (p->link_cfg.speed) {
210 pr_info("%s: unsupported speed: %d\n",
211 dev->name, p->link_cfg.speed);
215 netdev_info(dev, "link up, %s, full-duplex, %s PAUSE\n", s,
220 #ifdef CONFIG_CHELSIO_T4_DCB
221 /* Set up/tear down Data Center Bridging Priority mapping for a net device. */
222 static void dcb_tx_queue_prio_enable(struct net_device *dev, int enable)
224 struct port_info *pi = netdev_priv(dev);
225 struct adapter *adap = pi->adapter;
226 struct sge_eth_txq *txq = &adap->sge.ethtxq[pi->first_qset];
229 /* We use a simple mapping of Port TX Queue Index to DCB
230 * Priority when we're enabling DCB.
232 for (i = 0; i < pi->nqsets; i++, txq++) {
236 name = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DMAQ) |
238 FW_PARAMS_PARAM_DMAQ_EQ_DCBPRIO_ETH) |
239 FW_PARAMS_PARAM_YZ_V(txq->q.cntxt_id));
240 value = enable ? i : 0xffffffff;
242 /* Since we can be called while atomic (from "interrupt
243 * level") we need to issue the Set Parameters Commannd
244 * without sleeping (timeout < 0).
246 err = t4_set_params_timeout(adap, adap->mbox, adap->pf, 0, 1,
248 -FW_CMD_MAX_TIMEOUT);
251 dev_err(adap->pdev_dev,
252 "Can't %s DCB Priority on port %d, TX Queue %d: err=%d\n",
253 enable ? "set" : "unset", pi->port_id, i, -err);
255 txq->dcb_prio = value;
259 static int cxgb4_dcb_enabled(const struct net_device *dev)
261 struct port_info *pi = netdev_priv(dev);
263 if (!pi->dcb.enabled)
266 return ((pi->dcb.state == CXGB4_DCB_STATE_FW_ALLSYNCED) ||
267 (pi->dcb.state == CXGB4_DCB_STATE_HOST));
269 #endif /* CONFIG_CHELSIO_T4_DCB */
271 void t4_os_link_changed(struct adapter *adapter, int port_id, int link_stat)
273 struct net_device *dev = adapter->port[port_id];
275 /* Skip changes from disabled ports. */
276 if (netif_running(dev) && link_stat != netif_carrier_ok(dev)) {
278 netif_carrier_on(dev);
280 #ifdef CONFIG_CHELSIO_T4_DCB
281 if (cxgb4_dcb_enabled(dev)) {
282 cxgb4_dcb_state_init(dev);
283 dcb_tx_queue_prio_enable(dev, false);
285 #endif /* CONFIG_CHELSIO_T4_DCB */
286 netif_carrier_off(dev);
293 void t4_os_portmod_changed(const struct adapter *adap, int port_id)
295 static const char *mod_str[] = {
296 NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
299 const struct net_device *dev = adap->port[port_id];
300 const struct port_info *pi = netdev_priv(dev);
302 if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
303 netdev_info(dev, "port module unplugged\n");
304 else if (pi->mod_type < ARRAY_SIZE(mod_str))
305 netdev_info(dev, "%s module inserted\n", mod_str[pi->mod_type]);
306 else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
307 netdev_info(dev, "%s: unsupported port module inserted\n",
309 else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
310 netdev_info(dev, "%s: unknown port module inserted\n",
312 else if (pi->mod_type == FW_PORT_MOD_TYPE_ERROR)
313 netdev_info(dev, "%s: transceiver module error\n", dev->name);
315 netdev_info(dev, "%s: unknown module type %d inserted\n",
316 dev->name, pi->mod_type);
319 int dbfifo_int_thresh = 10; /* 10 == 640 entry threshold */
320 module_param(dbfifo_int_thresh, int, 0644);
321 MODULE_PARM_DESC(dbfifo_int_thresh, "doorbell fifo interrupt threshold");
324 * usecs to sleep while draining the dbfifo
326 static int dbfifo_drain_delay = 1000;
327 module_param(dbfifo_drain_delay, int, 0644);
328 MODULE_PARM_DESC(dbfifo_drain_delay,
329 "usecs to sleep while draining the dbfifo");
331 static inline int cxgb4_set_addr_hash(struct port_info *pi)
333 struct adapter *adap = pi->adapter;
336 struct hash_mac_addr *entry;
338 /* Calculate the hash vector for the updated list and program it */
339 list_for_each_entry(entry, &adap->mac_hlist, list) {
340 ucast |= is_unicast_ether_addr(entry->addr);
341 vec |= (1ULL << hash_mac_addr(entry->addr));
343 return t4_set_addr_hash(adap, adap->mbox, pi->viid, ucast,
347 static int cxgb4_mac_sync(struct net_device *netdev, const u8 *mac_addr)
349 struct port_info *pi = netdev_priv(netdev);
350 struct adapter *adap = pi->adapter;
355 bool ucast = is_unicast_ether_addr(mac_addr);
356 const u8 *maclist[1] = {mac_addr};
357 struct hash_mac_addr *new_entry;
359 ret = t4_alloc_mac_filt(adap, adap->mbox, pi->viid, free, 1, maclist,
360 NULL, ucast ? &uhash : &mhash, false);
363 /* if hash != 0, then add the addr to hash addr list
364 * so on the end we will calculate the hash for the
365 * list and program it
367 if (uhash || mhash) {
368 new_entry = kzalloc(sizeof(*new_entry), GFP_ATOMIC);
371 ether_addr_copy(new_entry->addr, mac_addr);
372 list_add_tail(&new_entry->list, &adap->mac_hlist);
373 ret = cxgb4_set_addr_hash(pi);
376 return ret < 0 ? ret : 0;
379 static int cxgb4_mac_unsync(struct net_device *netdev, const u8 *mac_addr)
381 struct port_info *pi = netdev_priv(netdev);
382 struct adapter *adap = pi->adapter;
384 const u8 *maclist[1] = {mac_addr};
385 struct hash_mac_addr *entry, *tmp;
387 /* If the MAC address to be removed is in the hash addr
388 * list, delete it from the list and update hash vector
390 list_for_each_entry_safe(entry, tmp, &adap->mac_hlist, list) {
391 if (ether_addr_equal(entry->addr, mac_addr)) {
392 list_del(&entry->list);
394 return cxgb4_set_addr_hash(pi);
398 ret = t4_free_mac_filt(adap, adap->mbox, pi->viid, 1, maclist, false);
399 return ret < 0 ? -EINVAL : 0;
403 * Set Rx properties of a port, such as promiscruity, address filters, and MTU.
404 * If @mtu is -1 it is left unchanged.
406 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
408 struct port_info *pi = netdev_priv(dev);
409 struct adapter *adapter = pi->adapter;
411 __dev_uc_sync(dev, cxgb4_mac_sync, cxgb4_mac_unsync);
412 __dev_mc_sync(dev, cxgb4_mac_sync, cxgb4_mac_unsync);
414 return t4_set_rxmode(adapter, adapter->mbox, pi->viid, mtu,
415 (dev->flags & IFF_PROMISC) ? 1 : 0,
416 (dev->flags & IFF_ALLMULTI) ? 1 : 0, 1, -1,
421 * link_start - enable a port
422 * @dev: the port to enable
424 * Performs the MAC and PHY actions needed to enable a port.
426 static int link_start(struct net_device *dev)
429 struct port_info *pi = netdev_priv(dev);
430 unsigned int mb = pi->adapter->pf;
433 * We do not set address filters and promiscuity here, the stack does
434 * that step explicitly.
436 ret = t4_set_rxmode(pi->adapter, mb, pi->viid, dev->mtu, -1, -1, -1,
437 !!(dev->features & NETIF_F_HW_VLAN_CTAG_RX), true);
439 ret = t4_change_mac(pi->adapter, mb, pi->viid,
440 pi->xact_addr_filt, dev->dev_addr, true,
443 pi->xact_addr_filt = ret;
448 ret = t4_link_l1cfg(pi->adapter, mb, pi->tx_chan,
452 ret = t4_enable_vi_params(pi->adapter, mb, pi->viid, true,
453 true, CXGB4_DCB_ENABLED);
460 #ifdef CONFIG_CHELSIO_T4_DCB
461 /* Handle a Data Center Bridging update message from the firmware. */
462 static void dcb_rpl(struct adapter *adap, const struct fw_port_cmd *pcmd)
464 int port = FW_PORT_CMD_PORTID_G(ntohl(pcmd->op_to_portid));
465 struct net_device *dev = adap->port[adap->chan_map[port]];
466 int old_dcb_enabled = cxgb4_dcb_enabled(dev);
469 cxgb4_dcb_handle_fw_update(adap, pcmd);
470 new_dcb_enabled = cxgb4_dcb_enabled(dev);
472 /* If the DCB has become enabled or disabled on the port then we're
473 * going to need to set up/tear down DCB Priority parameters for the
474 * TX Queues associated with the port.
476 if (new_dcb_enabled != old_dcb_enabled)
477 dcb_tx_queue_prio_enable(dev, new_dcb_enabled);
479 #endif /* CONFIG_CHELSIO_T4_DCB */
481 /* Response queue handler for the FW event queue.
483 static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
484 const struct pkt_gl *gl)
486 u8 opcode = ((const struct rss_header *)rsp)->opcode;
488 rsp++; /* skip RSS header */
490 /* FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG.
492 if (unlikely(opcode == CPL_FW4_MSG &&
493 ((const struct cpl_fw4_msg *)rsp)->type == FW_TYPE_RSSCPL)) {
495 opcode = ((const struct rss_header *)rsp)->opcode;
497 if (opcode != CPL_SGE_EGR_UPDATE) {
498 dev_err(q->adap->pdev_dev, "unexpected FW4/CPL %#x on FW event queue\n"
504 if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
505 const struct cpl_sge_egr_update *p = (void *)rsp;
506 unsigned int qid = EGR_QID_G(ntohl(p->opcode_qid));
509 txq = q->adap->sge.egr_map[qid - q->adap->sge.egr_start];
511 if (txq->q_type == CXGB4_TXQ_ETH) {
512 struct sge_eth_txq *eq;
514 eq = container_of(txq, struct sge_eth_txq, q);
515 netif_tx_wake_queue(eq->txq);
517 struct sge_uld_txq *oq;
519 oq = container_of(txq, struct sge_uld_txq, q);
520 tasklet_schedule(&oq->qresume_tsk);
522 } else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
523 const struct cpl_fw6_msg *p = (void *)rsp;
525 #ifdef CONFIG_CHELSIO_T4_DCB
526 const struct fw_port_cmd *pcmd = (const void *)p->data;
527 unsigned int cmd = FW_CMD_OP_G(ntohl(pcmd->op_to_portid));
528 unsigned int action =
529 FW_PORT_CMD_ACTION_G(ntohl(pcmd->action_to_len16));
531 if (cmd == FW_PORT_CMD &&
532 action == FW_PORT_ACTION_GET_PORT_INFO) {
533 int port = FW_PORT_CMD_PORTID_G(
534 be32_to_cpu(pcmd->op_to_portid));
535 struct net_device *dev =
536 q->adap->port[q->adap->chan_map[port]];
537 int state_input = ((pcmd->u.info.dcbxdis_pkd &
538 FW_PORT_CMD_DCBXDIS_F)
539 ? CXGB4_DCB_INPUT_FW_DISABLED
540 : CXGB4_DCB_INPUT_FW_ENABLED);
542 cxgb4_dcb_state_fsm(dev, state_input);
545 if (cmd == FW_PORT_CMD &&
546 action == FW_PORT_ACTION_L2_DCB_CFG)
547 dcb_rpl(q->adap, pcmd);
551 t4_handle_fw_rpl(q->adap, p->data);
552 } else if (opcode == CPL_L2T_WRITE_RPL) {
553 const struct cpl_l2t_write_rpl *p = (void *)rsp;
555 do_l2t_write_rpl(q->adap, p);
556 } else if (opcode == CPL_SET_TCB_RPL) {
557 const struct cpl_set_tcb_rpl *p = (void *)rsp;
559 filter_rpl(q->adap, p);
561 dev_err(q->adap->pdev_dev,
562 "unexpected CPL %#x on FW event queue\n", opcode);
567 static void disable_msi(struct adapter *adapter)
569 if (adapter->flags & USING_MSIX) {
570 pci_disable_msix(adapter->pdev);
571 adapter->flags &= ~USING_MSIX;
572 } else if (adapter->flags & USING_MSI) {
573 pci_disable_msi(adapter->pdev);
574 adapter->flags &= ~USING_MSI;
579 * Interrupt handler for non-data events used with MSI-X.
581 static irqreturn_t t4_nondata_intr(int irq, void *cookie)
583 struct adapter *adap = cookie;
584 u32 v = t4_read_reg(adap, MYPF_REG(PL_PF_INT_CAUSE_A));
588 t4_write_reg(adap, MYPF_REG(PL_PF_INT_CAUSE_A), v);
590 if (adap->flags & MASTER_PF)
591 t4_slow_intr_handler(adap);
596 * Name the MSI-X interrupts.
598 static void name_msix_vecs(struct adapter *adap)
600 int i, j, msi_idx = 2, n = sizeof(adap->msix_info[0].desc);
602 /* non-data interrupts */
603 snprintf(adap->msix_info[0].desc, n, "%s", adap->port[0]->name);
606 snprintf(adap->msix_info[1].desc, n, "%s-FWeventq",
607 adap->port[0]->name);
609 /* Ethernet queues */
610 for_each_port(adap, j) {
611 struct net_device *d = adap->port[j];
612 const struct port_info *pi = netdev_priv(d);
614 for (i = 0; i < pi->nqsets; i++, msi_idx++)
615 snprintf(adap->msix_info[msi_idx].desc, n, "%s-Rx%d",
620 static int request_msix_queue_irqs(struct adapter *adap)
622 struct sge *s = &adap->sge;
626 err = request_irq(adap->msix_info[1].vec, t4_sge_intr_msix, 0,
627 adap->msix_info[1].desc, &s->fw_evtq);
631 for_each_ethrxq(s, ethqidx) {
632 err = request_irq(adap->msix_info[msi_index].vec,
634 adap->msix_info[msi_index].desc,
635 &s->ethrxq[ethqidx].rspq);
643 while (--ethqidx >= 0)
644 free_irq(adap->msix_info[--msi_index].vec,
645 &s->ethrxq[ethqidx].rspq);
646 free_irq(adap->msix_info[1].vec, &s->fw_evtq);
650 static void free_msix_queue_irqs(struct adapter *adap)
652 int i, msi_index = 2;
653 struct sge *s = &adap->sge;
655 free_irq(adap->msix_info[1].vec, &s->fw_evtq);
656 for_each_ethrxq(s, i)
657 free_irq(adap->msix_info[msi_index++].vec, &s->ethrxq[i].rspq);
661 * cxgb4_write_rss - write the RSS table for a given port
663 * @queues: array of queue indices for RSS
665 * Sets up the portion of the HW RSS table for the port's VI to distribute
666 * packets to the Rx queues in @queues.
667 * Should never be called before setting up sge eth rx queues
669 int cxgb4_write_rss(const struct port_info *pi, const u16 *queues)
673 struct adapter *adapter = pi->adapter;
674 const struct sge_eth_rxq *rxq;
676 rxq = &adapter->sge.ethrxq[pi->first_qset];
677 rss = kmalloc(pi->rss_size * sizeof(u16), GFP_KERNEL);
681 /* map the queue indices to queue ids */
682 for (i = 0; i < pi->rss_size; i++, queues++)
683 rss[i] = rxq[*queues].rspq.abs_id;
685 err = t4_config_rss_range(adapter, adapter->pf, pi->viid, 0,
686 pi->rss_size, rss, pi->rss_size);
687 /* If Tunnel All Lookup isn't specified in the global RSS
688 * Configuration, then we need to specify a default Ingress
689 * Queue for any ingress packets which aren't hashed. We'll
690 * use our first ingress queue ...
693 err = t4_config_vi_rss(adapter, adapter->mbox, pi->viid,
694 FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN_F |
695 FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN_F |
696 FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN_F |
697 FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN_F |
698 FW_RSS_VI_CONFIG_CMD_UDPEN_F,
705 * setup_rss - configure RSS
708 * Sets up RSS for each port.
710 static int setup_rss(struct adapter *adap)
714 for_each_port(adap, i) {
715 const struct port_info *pi = adap2pinfo(adap, i);
717 /* Fill default values with equal distribution */
718 for (j = 0; j < pi->rss_size; j++)
719 pi->rss[j] = j % pi->nqsets;
721 err = cxgb4_write_rss(pi, pi->rss);
729 * Return the channel of the ingress queue with the given qid.
731 static unsigned int rxq_to_chan(const struct sge *p, unsigned int qid)
733 qid -= p->ingr_start;
734 return netdev2pinfo(p->ingr_map[qid]->netdev)->tx_chan;
738 * Wait until all NAPI handlers are descheduled.
740 static void quiesce_rx(struct adapter *adap)
744 for (i = 0; i < adap->sge.ingr_sz; i++) {
745 struct sge_rspq *q = adap->sge.ingr_map[i];
748 napi_disable(&q->napi);
752 /* Disable interrupt and napi handler */
753 static void disable_interrupts(struct adapter *adap)
755 if (adap->flags & FULL_INIT_DONE) {
756 t4_intr_disable(adap);
757 if (adap->flags & USING_MSIX) {
758 free_msix_queue_irqs(adap);
759 free_irq(adap->msix_info[0].vec, adap);
761 free_irq(adap->pdev->irq, adap);
768 * Enable NAPI scheduling and interrupt generation for all Rx queues.
770 static void enable_rx(struct adapter *adap)
774 for (i = 0; i < adap->sge.ingr_sz; i++) {
775 struct sge_rspq *q = adap->sge.ingr_map[i];
780 napi_enable(&q->napi);
782 /* 0-increment GTS to start the timer and enable interrupts */
783 t4_write_reg(adap, MYPF_REG(SGE_PF_GTS_A),
784 SEINTARM_V(q->intr_params) |
785 INGRESSQID_V(q->cntxt_id));
790 static int setup_fw_sge_queues(struct adapter *adap)
792 struct sge *s = &adap->sge;
795 bitmap_zero(s->starving_fl, s->egr_sz);
796 bitmap_zero(s->txq_maperr, s->egr_sz);
798 if (adap->flags & USING_MSIX)
799 adap->msi_idx = 1; /* vector 0 is for non-queue interrupts */
801 err = t4_sge_alloc_rxq(adap, &s->intrq, false, adap->port[0], 0,
802 NULL, NULL, NULL, -1);
805 adap->msi_idx = -((int)s->intrq.abs_id + 1);
808 err = t4_sge_alloc_rxq(adap, &s->fw_evtq, true, adap->port[0],
809 adap->msi_idx, NULL, fwevtq_handler, NULL, -1);
811 t4_free_sge_resources(adap);
816 * setup_sge_queues - configure SGE Tx/Rx/response queues
819 * Determines how many sets of SGE queues to use and initializes them.
820 * We support multiple queue sets per port if we have MSI-X, otherwise
821 * just one queue set per port.
823 static int setup_sge_queues(struct adapter *adap)
826 struct sge *s = &adap->sge;
827 struct sge_uld_rxq_info *rxq_info = s->uld_rxq_info[CXGB4_ULD_RDMA];
828 unsigned int cmplqid = 0;
830 for_each_port(adap, i) {
831 struct net_device *dev = adap->port[i];
832 struct port_info *pi = netdev_priv(dev);
833 struct sge_eth_rxq *q = &s->ethrxq[pi->first_qset];
834 struct sge_eth_txq *t = &s->ethtxq[pi->first_qset];
836 for (j = 0; j < pi->nqsets; j++, q++) {
837 if (adap->msi_idx > 0)
839 err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev,
840 adap->msi_idx, &q->fl,
843 t4_get_mps_bg_map(adap,
848 memset(&q->stats, 0, sizeof(q->stats));
850 for (j = 0; j < pi->nqsets; j++, t++) {
851 err = t4_sge_alloc_eth_txq(adap, t, dev,
852 netdev_get_tx_queue(dev, j),
853 s->fw_evtq.cntxt_id);
859 for_each_port(adap, i) {
860 /* Note that cmplqid below is 0 if we don't
861 * have RDMA queues, and that's the right value.
864 cmplqid = rxq_info->uldrxq[i].rspq.cntxt_id;
866 err = t4_sge_alloc_ctrl_txq(adap, &s->ctrlq[i], adap->port[i],
867 s->fw_evtq.cntxt_id, cmplqid);
872 t4_write_reg(adap, is_t4(adap->params.chip) ?
873 MPS_TRC_RSS_CONTROL_A :
874 MPS_T5_TRC_RSS_CONTROL_A,
875 RSSCONTROL_V(netdev2pinfo(adap->port[0])->tx_chan) |
876 QUEUENUMBER_V(s->ethrxq[0].rspq.abs_id));
879 t4_free_sge_resources(adap);
883 static u16 cxgb_select_queue(struct net_device *dev, struct sk_buff *skb,
884 void *accel_priv, select_queue_fallback_t fallback)
888 #ifdef CONFIG_CHELSIO_T4_DCB
889 /* If a Data Center Bridging has been successfully negotiated on this
890 * link then we'll use the skb's priority to map it to a TX Queue.
891 * The skb's priority is determined via the VLAN Tag Priority Code
894 if (cxgb4_dcb_enabled(dev)) {
898 err = vlan_get_tag(skb, &vlan_tci);
902 "TX Packet without VLAN Tag on DCB Link\n");
905 txq = (vlan_tci & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
906 #ifdef CONFIG_CHELSIO_T4_FCOE
907 if (skb->protocol == htons(ETH_P_FCOE))
908 txq = skb->priority & 0x7;
909 #endif /* CONFIG_CHELSIO_T4_FCOE */
913 #endif /* CONFIG_CHELSIO_T4_DCB */
916 txq = (skb_rx_queue_recorded(skb)
917 ? skb_get_rx_queue(skb)
918 : smp_processor_id());
920 while (unlikely(txq >= dev->real_num_tx_queues))
921 txq -= dev->real_num_tx_queues;
926 return fallback(dev, skb) % dev->real_num_tx_queues;
929 static int closest_timer(const struct sge *s, int time)
931 int i, delta, match = 0, min_delta = INT_MAX;
933 for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
934 delta = time - s->timer_val[i];
937 if (delta < min_delta) {
945 static int closest_thres(const struct sge *s, int thres)
947 int i, delta, match = 0, min_delta = INT_MAX;
949 for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
950 delta = thres - s->counter_val[i];
953 if (delta < min_delta) {
962 * cxgb4_set_rspq_intr_params - set a queue's interrupt holdoff parameters
964 * @us: the hold-off time in us, or 0 to disable timer
965 * @cnt: the hold-off packet count, or 0 to disable counter
967 * Sets an Rx queue's interrupt hold-off time and packet count. At least
968 * one of the two needs to be enabled for the queue to generate interrupts.
970 int cxgb4_set_rspq_intr_params(struct sge_rspq *q,
971 unsigned int us, unsigned int cnt)
973 struct adapter *adap = q->adap;
982 new_idx = closest_thres(&adap->sge, cnt);
983 if (q->desc && q->pktcnt_idx != new_idx) {
984 /* the queue has already been created, update it */
985 v = FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DMAQ) |
987 FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
988 FW_PARAMS_PARAM_YZ_V(q->cntxt_id);
989 err = t4_set_params(adap, adap->mbox, adap->pf, 0, 1,
994 q->pktcnt_idx = new_idx;
997 us = us == 0 ? 6 : closest_timer(&adap->sge, us);
998 q->intr_params = QINTR_TIMER_IDX_V(us) | QINTR_CNT_EN_V(cnt > 0);
1002 static int cxgb_set_features(struct net_device *dev, netdev_features_t features)
1004 const struct port_info *pi = netdev_priv(dev);
1005 netdev_features_t changed = dev->features ^ features;
1008 if (!(changed & NETIF_F_HW_VLAN_CTAG_RX))
1011 err = t4_set_rxmode(pi->adapter, pi->adapter->pf, pi->viid, -1,
1013 !!(features & NETIF_F_HW_VLAN_CTAG_RX), true);
1015 dev->features = features ^ NETIF_F_HW_VLAN_CTAG_RX;
1019 static int setup_debugfs(struct adapter *adap)
1021 if (IS_ERR_OR_NULL(adap->debugfs_root))
1024 #ifdef CONFIG_DEBUG_FS
1025 t4_setup_debugfs(adap);
1031 * upper-layer driver support
1035 * Allocate an active-open TID and set it to the supplied value.
1037 int cxgb4_alloc_atid(struct tid_info *t, void *data)
1041 spin_lock_bh(&t->atid_lock);
1043 union aopen_entry *p = t->afree;
1045 atid = (p - t->atid_tab) + t->atid_base;
1050 spin_unlock_bh(&t->atid_lock);
1053 EXPORT_SYMBOL(cxgb4_alloc_atid);
1056 * Release an active-open TID.
1058 void cxgb4_free_atid(struct tid_info *t, unsigned int atid)
1060 union aopen_entry *p = &t->atid_tab[atid - t->atid_base];
1062 spin_lock_bh(&t->atid_lock);
1066 spin_unlock_bh(&t->atid_lock);
1068 EXPORT_SYMBOL(cxgb4_free_atid);
1071 * Allocate a server TID and set it to the supplied value.
1073 int cxgb4_alloc_stid(struct tid_info *t, int family, void *data)
1077 spin_lock_bh(&t->stid_lock);
1078 if (family == PF_INET) {
1079 stid = find_first_zero_bit(t->stid_bmap, t->nstids);
1080 if (stid < t->nstids)
1081 __set_bit(stid, t->stid_bmap);
1085 stid = bitmap_find_free_region(t->stid_bmap, t->nstids, 1);
1090 t->stid_tab[stid].data = data;
1091 stid += t->stid_base;
1092 /* IPv6 requires max of 520 bits or 16 cells in TCAM
1093 * This is equivalent to 4 TIDs. With CLIP enabled it
1096 if (family == PF_INET)
1099 t->stids_in_use += 2;
1101 spin_unlock_bh(&t->stid_lock);
1104 EXPORT_SYMBOL(cxgb4_alloc_stid);
1106 /* Allocate a server filter TID and set it to the supplied value.
1108 int cxgb4_alloc_sftid(struct tid_info *t, int family, void *data)
1112 spin_lock_bh(&t->stid_lock);
1113 if (family == PF_INET) {
1114 stid = find_next_zero_bit(t->stid_bmap,
1115 t->nstids + t->nsftids, t->nstids);
1116 if (stid < (t->nstids + t->nsftids))
1117 __set_bit(stid, t->stid_bmap);
1124 t->stid_tab[stid].data = data;
1126 stid += t->sftid_base;
1129 spin_unlock_bh(&t->stid_lock);
1132 EXPORT_SYMBOL(cxgb4_alloc_sftid);
1134 /* Release a server TID.
1136 void cxgb4_free_stid(struct tid_info *t, unsigned int stid, int family)
1138 /* Is it a server filter TID? */
1139 if (t->nsftids && (stid >= t->sftid_base)) {
1140 stid -= t->sftid_base;
1143 stid -= t->stid_base;
1146 spin_lock_bh(&t->stid_lock);
1147 if (family == PF_INET)
1148 __clear_bit(stid, t->stid_bmap);
1150 bitmap_release_region(t->stid_bmap, stid, 1);
1151 t->stid_tab[stid].data = NULL;
1152 if (stid < t->nstids) {
1153 if (family == PF_INET)
1156 t->stids_in_use -= 2;
1160 spin_unlock_bh(&t->stid_lock);
1162 EXPORT_SYMBOL(cxgb4_free_stid);
1165 * Populate a TID_RELEASE WR. Caller must properly size the skb.
1167 static void mk_tid_release(struct sk_buff *skb, unsigned int chan,
1170 struct cpl_tid_release *req;
1172 set_wr_txq(skb, CPL_PRIORITY_SETUP, chan);
1173 req = (struct cpl_tid_release *)__skb_put(skb, sizeof(*req));
1174 INIT_TP_WR(req, tid);
1175 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, tid));
1179 * Queue a TID release request and if necessary schedule a work queue to
1182 static void cxgb4_queue_tid_release(struct tid_info *t, unsigned int chan,
1185 void **p = &t->tid_tab[tid];
1186 struct adapter *adap = container_of(t, struct adapter, tids);
1188 spin_lock_bh(&adap->tid_release_lock);
1189 *p = adap->tid_release_head;
1190 /* Low 2 bits encode the Tx channel number */
1191 adap->tid_release_head = (void **)((uintptr_t)p | chan);
1192 if (!adap->tid_release_task_busy) {
1193 adap->tid_release_task_busy = true;
1194 queue_work(adap->workq, &adap->tid_release_task);
1196 spin_unlock_bh(&adap->tid_release_lock);
1200 * Process the list of pending TID release requests.
1202 static void process_tid_release_list(struct work_struct *work)
1204 struct sk_buff *skb;
1205 struct adapter *adap;
1207 adap = container_of(work, struct adapter, tid_release_task);
1209 spin_lock_bh(&adap->tid_release_lock);
1210 while (adap->tid_release_head) {
1211 void **p = adap->tid_release_head;
1212 unsigned int chan = (uintptr_t)p & 3;
1213 p = (void *)p - chan;
1215 adap->tid_release_head = *p;
1217 spin_unlock_bh(&adap->tid_release_lock);
1219 while (!(skb = alloc_skb(sizeof(struct cpl_tid_release),
1221 schedule_timeout_uninterruptible(1);
1223 mk_tid_release(skb, chan, p - adap->tids.tid_tab);
1224 t4_ofld_send(adap, skb);
1225 spin_lock_bh(&adap->tid_release_lock);
1227 adap->tid_release_task_busy = false;
1228 spin_unlock_bh(&adap->tid_release_lock);
1232 * Release a TID and inform HW. If we are unable to allocate the release
1233 * message we defer to a work queue.
1235 void cxgb4_remove_tid(struct tid_info *t, unsigned int chan, unsigned int tid)
1237 struct sk_buff *skb;
1238 struct adapter *adap = container_of(t, struct adapter, tids);
1240 WARN_ON(tid >= t->ntids);
1242 if (t->tid_tab[tid]) {
1243 t->tid_tab[tid] = NULL;
1244 if (t->hash_base && (tid >= t->hash_base))
1245 atomic_dec(&t->hash_tids_in_use);
1247 atomic_dec(&t->tids_in_use);
1250 skb = alloc_skb(sizeof(struct cpl_tid_release), GFP_ATOMIC);
1252 mk_tid_release(skb, chan, tid);
1253 t4_ofld_send(adap, skb);
1255 cxgb4_queue_tid_release(t, chan, tid);
1257 EXPORT_SYMBOL(cxgb4_remove_tid);
1260 * Allocate and initialize the TID tables. Returns 0 on success.
1262 static int tid_init(struct tid_info *t)
1264 struct adapter *adap = container_of(t, struct adapter, tids);
1265 unsigned int max_ftids = t->nftids + t->nsftids;
1266 unsigned int natids = t->natids;
1267 unsigned int stid_bmap_size;
1268 unsigned int ftid_bmap_size;
1271 stid_bmap_size = BITS_TO_LONGS(t->nstids + t->nsftids);
1272 ftid_bmap_size = BITS_TO_LONGS(t->nftids);
1273 size = t->ntids * sizeof(*t->tid_tab) +
1274 natids * sizeof(*t->atid_tab) +
1275 t->nstids * sizeof(*t->stid_tab) +
1276 t->nsftids * sizeof(*t->stid_tab) +
1277 stid_bmap_size * sizeof(long) +
1278 max_ftids * sizeof(*t->ftid_tab) +
1279 ftid_bmap_size * sizeof(long);
1281 t->tid_tab = kvzalloc(size, GFP_KERNEL);
1285 t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
1286 t->stid_tab = (struct serv_entry *)&t->atid_tab[natids];
1287 t->stid_bmap = (unsigned long *)&t->stid_tab[t->nstids + t->nsftids];
1288 t->ftid_tab = (struct filter_entry *)&t->stid_bmap[stid_bmap_size];
1289 t->ftid_bmap = (unsigned long *)&t->ftid_tab[max_ftids];
1290 spin_lock_init(&t->stid_lock);
1291 spin_lock_init(&t->atid_lock);
1292 spin_lock_init(&t->ftid_lock);
1294 t->stids_in_use = 0;
1295 t->sftids_in_use = 0;
1297 t->atids_in_use = 0;
1298 atomic_set(&t->tids_in_use, 0);
1299 atomic_set(&t->hash_tids_in_use, 0);
1301 /* Setup the free list for atid_tab and clear the stid bitmap. */
1304 t->atid_tab[natids - 1].next = &t->atid_tab[natids];
1305 t->afree = t->atid_tab;
1308 if (is_offload(adap)) {
1309 bitmap_zero(t->stid_bmap, t->nstids + t->nsftids);
1310 /* Reserve stid 0 for T4/T5 adapters */
1311 if (!t->stid_base &&
1312 CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5)
1313 __set_bit(0, t->stid_bmap);
1316 bitmap_zero(t->ftid_bmap, t->nftids);
1321 * cxgb4_create_server - create an IP server
1323 * @stid: the server TID
1324 * @sip: local IP address to bind server to
1325 * @sport: the server's TCP port
1326 * @queue: queue to direct messages from this server to
1328 * Create an IP server for the given port and address.
1329 * Returns <0 on error and one of the %NET_XMIT_* values on success.
1331 int cxgb4_create_server(const struct net_device *dev, unsigned int stid,
1332 __be32 sip, __be16 sport, __be16 vlan,
1336 struct sk_buff *skb;
1337 struct adapter *adap;
1338 struct cpl_pass_open_req *req;
1341 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
1345 adap = netdev2adap(dev);
1346 req = (struct cpl_pass_open_req *)__skb_put(skb, sizeof(*req));
1348 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, stid));
1349 req->local_port = sport;
1350 req->peer_port = htons(0);
1351 req->local_ip = sip;
1352 req->peer_ip = htonl(0);
1353 chan = rxq_to_chan(&adap->sge, queue);
1354 req->opt0 = cpu_to_be64(TX_CHAN_V(chan));
1355 req->opt1 = cpu_to_be64(CONN_POLICY_V(CPL_CONN_POLICY_ASK) |
1356 SYN_RSS_ENABLE_F | SYN_RSS_QUEUE_V(queue));
1357 ret = t4_mgmt_tx(adap, skb);
1358 return net_xmit_eval(ret);
1360 EXPORT_SYMBOL(cxgb4_create_server);
1362 /* cxgb4_create_server6 - create an IPv6 server
1364 * @stid: the server TID
1365 * @sip: local IPv6 address to bind server to
1366 * @sport: the server's TCP port
1367 * @queue: queue to direct messages from this server to
1369 * Create an IPv6 server for the given port and address.
1370 * Returns <0 on error and one of the %NET_XMIT_* values on success.
1372 int cxgb4_create_server6(const struct net_device *dev, unsigned int stid,
1373 const struct in6_addr *sip, __be16 sport,
1377 struct sk_buff *skb;
1378 struct adapter *adap;
1379 struct cpl_pass_open_req6 *req;
1382 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
1386 adap = netdev2adap(dev);
1387 req = (struct cpl_pass_open_req6 *)__skb_put(skb, sizeof(*req));
1389 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ6, stid));
1390 req->local_port = sport;
1391 req->peer_port = htons(0);
1392 req->local_ip_hi = *(__be64 *)(sip->s6_addr);
1393 req->local_ip_lo = *(__be64 *)(sip->s6_addr + 8);
1394 req->peer_ip_hi = cpu_to_be64(0);
1395 req->peer_ip_lo = cpu_to_be64(0);
1396 chan = rxq_to_chan(&adap->sge, queue);
1397 req->opt0 = cpu_to_be64(TX_CHAN_V(chan));
1398 req->opt1 = cpu_to_be64(CONN_POLICY_V(CPL_CONN_POLICY_ASK) |
1399 SYN_RSS_ENABLE_F | SYN_RSS_QUEUE_V(queue));
1400 ret = t4_mgmt_tx(adap, skb);
1401 return net_xmit_eval(ret);
1403 EXPORT_SYMBOL(cxgb4_create_server6);
1405 int cxgb4_remove_server(const struct net_device *dev, unsigned int stid,
1406 unsigned int queue, bool ipv6)
1408 struct sk_buff *skb;
1409 struct adapter *adap;
1410 struct cpl_close_listsvr_req *req;
1413 adap = netdev2adap(dev);
1415 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
1419 req = (struct cpl_close_listsvr_req *)__skb_put(skb, sizeof(*req));
1421 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, stid));
1422 req->reply_ctrl = htons(NO_REPLY_V(0) | (ipv6 ? LISTSVR_IPV6_V(1) :
1423 LISTSVR_IPV6_V(0)) | QUEUENO_V(queue));
1424 ret = t4_mgmt_tx(adap, skb);
1425 return net_xmit_eval(ret);
1427 EXPORT_SYMBOL(cxgb4_remove_server);
1430 * cxgb4_best_mtu - find the entry in the MTU table closest to an MTU
1431 * @mtus: the HW MTU table
1432 * @mtu: the target MTU
1433 * @idx: index of selected entry in the MTU table
1435 * Returns the index and the value in the HW MTU table that is closest to
1436 * but does not exceed @mtu, unless @mtu is smaller than any value in the
1437 * table, in which case that smallest available value is selected.
1439 unsigned int cxgb4_best_mtu(const unsigned short *mtus, unsigned short mtu,
1444 while (i < NMTUS - 1 && mtus[i + 1] <= mtu)
1450 EXPORT_SYMBOL(cxgb4_best_mtu);
1453 * cxgb4_best_aligned_mtu - find best MTU, [hopefully] data size aligned
1454 * @mtus: the HW MTU table
1455 * @header_size: Header Size
1456 * @data_size_max: maximum Data Segment Size
1457 * @data_size_align: desired Data Segment Size Alignment (2^N)
1458 * @mtu_idxp: HW MTU Table Index return value pointer (possibly NULL)
1460 * Similar to cxgb4_best_mtu() but instead of searching the Hardware
1461 * MTU Table based solely on a Maximum MTU parameter, we break that
1462 * parameter up into a Header Size and Maximum Data Segment Size, and
1463 * provide a desired Data Segment Size Alignment. If we find an MTU in
1464 * the Hardware MTU Table which will result in a Data Segment Size with
1465 * the requested alignment _and_ that MTU isn't "too far" from the
1466 * closest MTU, then we'll return that rather than the closest MTU.
1468 unsigned int cxgb4_best_aligned_mtu(const unsigned short *mtus,
1469 unsigned short header_size,
1470 unsigned short data_size_max,
1471 unsigned short data_size_align,
1472 unsigned int *mtu_idxp)
1474 unsigned short max_mtu = header_size + data_size_max;
1475 unsigned short data_size_align_mask = data_size_align - 1;
1476 int mtu_idx, aligned_mtu_idx;
1478 /* Scan the MTU Table till we find an MTU which is larger than our
1479 * Maximum MTU or we reach the end of the table. Along the way,
1480 * record the last MTU found, if any, which will result in a Data
1481 * Segment Length matching the requested alignment.
1483 for (mtu_idx = 0, aligned_mtu_idx = -1; mtu_idx < NMTUS; mtu_idx++) {
1484 unsigned short data_size = mtus[mtu_idx] - header_size;
1486 /* If this MTU minus the Header Size would result in a
1487 * Data Segment Size of the desired alignment, remember it.
1489 if ((data_size & data_size_align_mask) == 0)
1490 aligned_mtu_idx = mtu_idx;
1492 /* If we're not at the end of the Hardware MTU Table and the
1493 * next element is larger than our Maximum MTU, drop out of
1496 if (mtu_idx+1 < NMTUS && mtus[mtu_idx+1] > max_mtu)
1500 /* If we fell out of the loop because we ran to the end of the table,
1501 * then we just have to use the last [largest] entry.
1503 if (mtu_idx == NMTUS)
1506 /* If we found an MTU which resulted in the requested Data Segment
1507 * Length alignment and that's "not far" from the largest MTU which is
1508 * less than or equal to the maximum MTU, then use that.
1510 if (aligned_mtu_idx >= 0 &&
1511 mtu_idx - aligned_mtu_idx <= 1)
1512 mtu_idx = aligned_mtu_idx;
1514 /* If the caller has passed in an MTU Index pointer, pass the
1515 * MTU Index back. Return the MTU value.
1518 *mtu_idxp = mtu_idx;
1519 return mtus[mtu_idx];
1521 EXPORT_SYMBOL(cxgb4_best_aligned_mtu);
1524 * cxgb4_tp_smt_idx - Get the Source Mac Table index for this VI
1526 * @viid: VI id of the given port
1528 * Return the SMT index for this VI.
1530 unsigned int cxgb4_tp_smt_idx(enum chip_type chip, unsigned int viid)
1532 /* In T4/T5, SMT contains 256 SMAC entries organized in
1533 * 128 rows of 2 entries each.
1534 * In T6, SMT contains 256 SMAC entries in 256 rows.
1535 * TODO: The below code needs to be updated when we add support
1538 if (CHELSIO_CHIP_VERSION(chip) <= CHELSIO_T5)
1539 return ((viid & 0x7f) << 1);
1541 return (viid & 0x7f);
1543 EXPORT_SYMBOL(cxgb4_tp_smt_idx);
1546 * cxgb4_port_chan - get the HW channel of a port
1547 * @dev: the net device for the port
1549 * Return the HW Tx channel of the given port.
1551 unsigned int cxgb4_port_chan(const struct net_device *dev)
1553 return netdev2pinfo(dev)->tx_chan;
1555 EXPORT_SYMBOL(cxgb4_port_chan);
1557 unsigned int cxgb4_dbfifo_count(const struct net_device *dev, int lpfifo)
1559 struct adapter *adap = netdev2adap(dev);
1560 u32 v1, v2, lp_count, hp_count;
1562 v1 = t4_read_reg(adap, SGE_DBFIFO_STATUS_A);
1563 v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2_A);
1564 if (is_t4(adap->params.chip)) {
1565 lp_count = LP_COUNT_G(v1);
1566 hp_count = HP_COUNT_G(v1);
1568 lp_count = LP_COUNT_T5_G(v1);
1569 hp_count = HP_COUNT_T5_G(v2);
1571 return lpfifo ? lp_count : hp_count;
1573 EXPORT_SYMBOL(cxgb4_dbfifo_count);
1576 * cxgb4_port_viid - get the VI id of a port
1577 * @dev: the net device for the port
1579 * Return the VI id of the given port.
1581 unsigned int cxgb4_port_viid(const struct net_device *dev)
1583 return netdev2pinfo(dev)->viid;
1585 EXPORT_SYMBOL(cxgb4_port_viid);
1588 * cxgb4_port_idx - get the index of a port
1589 * @dev: the net device for the port
1591 * Return the index of the given port.
1593 unsigned int cxgb4_port_idx(const struct net_device *dev)
1595 return netdev2pinfo(dev)->port_id;
1597 EXPORT_SYMBOL(cxgb4_port_idx);
1599 void cxgb4_get_tcp_stats(struct pci_dev *pdev, struct tp_tcp_stats *v4,
1600 struct tp_tcp_stats *v6)
1602 struct adapter *adap = pci_get_drvdata(pdev);
1604 spin_lock(&adap->stats_lock);
1605 t4_tp_get_tcp_stats(adap, v4, v6);
1606 spin_unlock(&adap->stats_lock);
1608 EXPORT_SYMBOL(cxgb4_get_tcp_stats);
1610 void cxgb4_iscsi_init(struct net_device *dev, unsigned int tag_mask,
1611 const unsigned int *pgsz_order)
1613 struct adapter *adap = netdev2adap(dev);
1615 t4_write_reg(adap, ULP_RX_ISCSI_TAGMASK_A, tag_mask);
1616 t4_write_reg(adap, ULP_RX_ISCSI_PSZ_A, HPZ0_V(pgsz_order[0]) |
1617 HPZ1_V(pgsz_order[1]) | HPZ2_V(pgsz_order[2]) |
1618 HPZ3_V(pgsz_order[3]));
1620 EXPORT_SYMBOL(cxgb4_iscsi_init);
1622 int cxgb4_flush_eq_cache(struct net_device *dev)
1624 struct adapter *adap = netdev2adap(dev);
1626 return t4_sge_ctxt_flush(adap, adap->mbox);
1628 EXPORT_SYMBOL(cxgb4_flush_eq_cache);
1630 static int read_eq_indices(struct adapter *adap, u16 qid, u16 *pidx, u16 *cidx)
1632 u32 addr = t4_read_reg(adap, SGE_DBQ_CTXT_BADDR_A) + 24 * qid + 8;
1636 spin_lock(&adap->win0_lock);
1637 ret = t4_memory_rw(adap, 0, MEM_EDC0, addr,
1638 sizeof(indices), (__be32 *)&indices,
1640 spin_unlock(&adap->win0_lock);
1642 *cidx = (be64_to_cpu(indices) >> 25) & 0xffff;
1643 *pidx = (be64_to_cpu(indices) >> 9) & 0xffff;
1648 int cxgb4_sync_txq_pidx(struct net_device *dev, u16 qid, u16 pidx,
1651 struct adapter *adap = netdev2adap(dev);
1652 u16 hw_pidx, hw_cidx;
1655 ret = read_eq_indices(adap, qid, &hw_pidx, &hw_cidx);
1659 if (pidx != hw_pidx) {
1663 if (pidx >= hw_pidx)
1664 delta = pidx - hw_pidx;
1666 delta = size - hw_pidx + pidx;
1668 if (is_t4(adap->params.chip))
1669 val = PIDX_V(delta);
1671 val = PIDX_T5_V(delta);
1673 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
1679 EXPORT_SYMBOL(cxgb4_sync_txq_pidx);
1681 int cxgb4_read_tpte(struct net_device *dev, u32 stag, __be32 *tpte)
1683 struct adapter *adap;
1684 u32 offset, memtype, memaddr;
1685 u32 edc0_size, edc1_size, mc0_size, mc1_size, size;
1686 u32 edc0_end, edc1_end, mc0_end, mc1_end;
1689 adap = netdev2adap(dev);
1691 offset = ((stag >> 8) * 32) + adap->vres.stag.start;
1693 /* Figure out where the offset lands in the Memory Type/Address scheme.
1694 * This code assumes that the memory is laid out starting at offset 0
1695 * with no breaks as: EDC0, EDC1, MC0, MC1. All cards have both EDC0
1696 * and EDC1. Some cards will have neither MC0 nor MC1, most cards have
1697 * MC0, and some have both MC0 and MC1.
1699 size = t4_read_reg(adap, MA_EDRAM0_BAR_A);
1700 edc0_size = EDRAM0_SIZE_G(size) << 20;
1701 size = t4_read_reg(adap, MA_EDRAM1_BAR_A);
1702 edc1_size = EDRAM1_SIZE_G(size) << 20;
1703 size = t4_read_reg(adap, MA_EXT_MEMORY0_BAR_A);
1704 mc0_size = EXT_MEM0_SIZE_G(size) << 20;
1706 edc0_end = edc0_size;
1707 edc1_end = edc0_end + edc1_size;
1708 mc0_end = edc1_end + mc0_size;
1710 if (offset < edc0_end) {
1713 } else if (offset < edc1_end) {
1715 memaddr = offset - edc0_end;
1717 if (offset < mc0_end) {
1719 memaddr = offset - edc1_end;
1720 } else if (is_t5(adap->params.chip)) {
1721 size = t4_read_reg(adap, MA_EXT_MEMORY1_BAR_A);
1722 mc1_size = EXT_MEM1_SIZE_G(size) << 20;
1723 mc1_end = mc0_end + mc1_size;
1724 if (offset < mc1_end) {
1726 memaddr = offset - mc0_end;
1728 /* offset beyond the end of any memory */
1732 /* T4/T6 only has a single memory channel */
1737 spin_lock(&adap->win0_lock);
1738 ret = t4_memory_rw(adap, 0, memtype, memaddr, 32, tpte, T4_MEMORY_READ);
1739 spin_unlock(&adap->win0_lock);
1743 dev_err(adap->pdev_dev, "stag %#x, offset %#x out of range\n",
1747 EXPORT_SYMBOL(cxgb4_read_tpte);
1749 u64 cxgb4_read_sge_timestamp(struct net_device *dev)
1752 struct adapter *adap;
1754 adap = netdev2adap(dev);
1755 lo = t4_read_reg(adap, SGE_TIMESTAMP_LO_A);
1756 hi = TSVAL_G(t4_read_reg(adap, SGE_TIMESTAMP_HI_A));
1758 return ((u64)hi << 32) | (u64)lo;
1760 EXPORT_SYMBOL(cxgb4_read_sge_timestamp);
1762 int cxgb4_bar2_sge_qregs(struct net_device *dev,
1764 enum cxgb4_bar2_qtype qtype,
1767 unsigned int *pbar2_qid)
1769 return t4_bar2_sge_qregs(netdev2adap(dev),
1771 (qtype == CXGB4_BAR2_QTYPE_EGRESS
1772 ? T4_BAR2_QTYPE_EGRESS
1773 : T4_BAR2_QTYPE_INGRESS),
1778 EXPORT_SYMBOL(cxgb4_bar2_sge_qregs);
1780 static struct pci_driver cxgb4_driver;
1782 static void check_neigh_update(struct neighbour *neigh)
1784 const struct device *parent;
1785 const struct net_device *netdev = neigh->dev;
1787 if (is_vlan_dev(netdev))
1788 netdev = vlan_dev_real_dev(netdev);
1789 parent = netdev->dev.parent;
1790 if (parent && parent->driver == &cxgb4_driver.driver)
1791 t4_l2t_update(dev_get_drvdata(parent), neigh);
1794 static int netevent_cb(struct notifier_block *nb, unsigned long event,
1798 case NETEVENT_NEIGH_UPDATE:
1799 check_neigh_update(data);
1801 case NETEVENT_REDIRECT:
1808 static bool netevent_registered;
1809 static struct notifier_block cxgb4_netevent_nb = {
1810 .notifier_call = netevent_cb
1813 static void drain_db_fifo(struct adapter *adap, int usecs)
1815 u32 v1, v2, lp_count, hp_count;
1818 v1 = t4_read_reg(adap, SGE_DBFIFO_STATUS_A);
1819 v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2_A);
1820 if (is_t4(adap->params.chip)) {
1821 lp_count = LP_COUNT_G(v1);
1822 hp_count = HP_COUNT_G(v1);
1824 lp_count = LP_COUNT_T5_G(v1);
1825 hp_count = HP_COUNT_T5_G(v2);
1828 if (lp_count == 0 && hp_count == 0)
1830 set_current_state(TASK_UNINTERRUPTIBLE);
1831 schedule_timeout(usecs_to_jiffies(usecs));
1835 static void disable_txq_db(struct sge_txq *q)
1837 unsigned long flags;
1839 spin_lock_irqsave(&q->db_lock, flags);
1841 spin_unlock_irqrestore(&q->db_lock, flags);
1844 static void enable_txq_db(struct adapter *adap, struct sge_txq *q)
1846 spin_lock_irq(&q->db_lock);
1847 if (q->db_pidx_inc) {
1848 /* Make sure that all writes to the TX descriptors
1849 * are committed before we tell HW about them.
1852 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
1853 QID_V(q->cntxt_id) | PIDX_V(q->db_pidx_inc));
1857 spin_unlock_irq(&q->db_lock);
1860 static void disable_dbs(struct adapter *adap)
1864 for_each_ethrxq(&adap->sge, i)
1865 disable_txq_db(&adap->sge.ethtxq[i].q);
1866 if (is_offload(adap)) {
1867 struct sge_uld_txq_info *txq_info =
1868 adap->sge.uld_txq_info[CXGB4_TX_OFLD];
1871 for_each_ofldtxq(&adap->sge, i) {
1872 struct sge_uld_txq *txq = &txq_info->uldtxq[i];
1874 disable_txq_db(&txq->q);
1878 for_each_port(adap, i)
1879 disable_txq_db(&adap->sge.ctrlq[i].q);
1882 static void enable_dbs(struct adapter *adap)
1886 for_each_ethrxq(&adap->sge, i)
1887 enable_txq_db(adap, &adap->sge.ethtxq[i].q);
1888 if (is_offload(adap)) {
1889 struct sge_uld_txq_info *txq_info =
1890 adap->sge.uld_txq_info[CXGB4_TX_OFLD];
1893 for_each_ofldtxq(&adap->sge, i) {
1894 struct sge_uld_txq *txq = &txq_info->uldtxq[i];
1896 enable_txq_db(adap, &txq->q);
1900 for_each_port(adap, i)
1901 enable_txq_db(adap, &adap->sge.ctrlq[i].q);
1904 static void notify_rdma_uld(struct adapter *adap, enum cxgb4_control cmd)
1906 enum cxgb4_uld type = CXGB4_ULD_RDMA;
1908 if (adap->uld && adap->uld[type].handle)
1909 adap->uld[type].control(adap->uld[type].handle, cmd);
1912 static void process_db_full(struct work_struct *work)
1914 struct adapter *adap;
1916 adap = container_of(work, struct adapter, db_full_task);
1918 drain_db_fifo(adap, dbfifo_drain_delay);
1920 notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
1921 if (CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5)
1922 t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
1923 DBFIFO_HP_INT_F | DBFIFO_LP_INT_F,
1924 DBFIFO_HP_INT_F | DBFIFO_LP_INT_F);
1926 t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
1927 DBFIFO_LP_INT_F, DBFIFO_LP_INT_F);
1930 static void sync_txq_pidx(struct adapter *adap, struct sge_txq *q)
1932 u16 hw_pidx, hw_cidx;
1935 spin_lock_irq(&q->db_lock);
1936 ret = read_eq_indices(adap, (u16)q->cntxt_id, &hw_pidx, &hw_cidx);
1939 if (q->db_pidx != hw_pidx) {
1943 if (q->db_pidx >= hw_pidx)
1944 delta = q->db_pidx - hw_pidx;
1946 delta = q->size - hw_pidx + q->db_pidx;
1948 if (is_t4(adap->params.chip))
1949 val = PIDX_V(delta);
1951 val = PIDX_T5_V(delta);
1953 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
1954 QID_V(q->cntxt_id) | val);
1959 spin_unlock_irq(&q->db_lock);
1961 CH_WARN(adap, "DB drop recovery failed.\n");
1964 static void recover_all_queues(struct adapter *adap)
1968 for_each_ethrxq(&adap->sge, i)
1969 sync_txq_pidx(adap, &adap->sge.ethtxq[i].q);
1970 if (is_offload(adap)) {
1971 struct sge_uld_txq_info *txq_info =
1972 adap->sge.uld_txq_info[CXGB4_TX_OFLD];
1974 for_each_ofldtxq(&adap->sge, i) {
1975 struct sge_uld_txq *txq = &txq_info->uldtxq[i];
1977 sync_txq_pidx(adap, &txq->q);
1981 for_each_port(adap, i)
1982 sync_txq_pidx(adap, &adap->sge.ctrlq[i].q);
1985 static void process_db_drop(struct work_struct *work)
1987 struct adapter *adap;
1989 adap = container_of(work, struct adapter, db_drop_task);
1991 if (is_t4(adap->params.chip)) {
1992 drain_db_fifo(adap, dbfifo_drain_delay);
1993 notify_rdma_uld(adap, CXGB4_CONTROL_DB_DROP);
1994 drain_db_fifo(adap, dbfifo_drain_delay);
1995 recover_all_queues(adap);
1996 drain_db_fifo(adap, dbfifo_drain_delay);
1998 notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
1999 } else if (is_t5(adap->params.chip)) {
2000 u32 dropped_db = t4_read_reg(adap, 0x010ac);
2001 u16 qid = (dropped_db >> 15) & 0x1ffff;
2002 u16 pidx_inc = dropped_db & 0x1fff;
2004 unsigned int bar2_qid;
2007 ret = t4_bar2_sge_qregs(adap, qid, T4_BAR2_QTYPE_EGRESS,
2008 0, &bar2_qoffset, &bar2_qid);
2010 dev_err(adap->pdev_dev, "doorbell drop recovery: "
2011 "qid=%d, pidx_inc=%d\n", qid, pidx_inc);
2013 writel(PIDX_T5_V(pidx_inc) | QID_V(bar2_qid),
2014 adap->bar2 + bar2_qoffset + SGE_UDB_KDOORBELL);
2016 /* Re-enable BAR2 WC */
2017 t4_set_reg_field(adap, 0x10b0, 1<<15, 1<<15);
2020 if (CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5)
2021 t4_set_reg_field(adap, SGE_DOORBELL_CONTROL_A, DROPPED_DB_F, 0);
2024 void t4_db_full(struct adapter *adap)
2026 if (is_t4(adap->params.chip)) {
2028 notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
2029 t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
2030 DBFIFO_HP_INT_F | DBFIFO_LP_INT_F, 0);
2031 queue_work(adap->workq, &adap->db_full_task);
2035 void t4_db_dropped(struct adapter *adap)
2037 if (is_t4(adap->params.chip)) {
2039 notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
2041 queue_work(adap->workq, &adap->db_drop_task);
2044 void t4_register_netevent_notifier(void)
2046 if (!netevent_registered) {
2047 register_netevent_notifier(&cxgb4_netevent_nb);
2048 netevent_registered = true;
2052 static void detach_ulds(struct adapter *adap)
2056 mutex_lock(&uld_mutex);
2057 list_del(&adap->list_node);
2058 for (i = 0; i < CXGB4_ULD_MAX; i++)
2059 if (adap->uld && adap->uld[i].handle) {
2060 adap->uld[i].state_change(adap->uld[i].handle,
2061 CXGB4_STATE_DETACH);
2062 adap->uld[i].handle = NULL;
2064 if (netevent_registered && list_empty(&adapter_list)) {
2065 unregister_netevent_notifier(&cxgb4_netevent_nb);
2066 netevent_registered = false;
2068 mutex_unlock(&uld_mutex);
2071 static void notify_ulds(struct adapter *adap, enum cxgb4_state new_state)
2075 mutex_lock(&uld_mutex);
2076 for (i = 0; i < CXGB4_ULD_MAX; i++)
2077 if (adap->uld && adap->uld[i].handle)
2078 adap->uld[i].state_change(adap->uld[i].handle,
2080 mutex_unlock(&uld_mutex);
2083 #if IS_ENABLED(CONFIG_IPV6)
2084 static int cxgb4_inet6addr_handler(struct notifier_block *this,
2085 unsigned long event, void *data)
2087 struct inet6_ifaddr *ifa = data;
2088 struct net_device *event_dev = ifa->idev->dev;
2089 const struct device *parent = NULL;
2090 #if IS_ENABLED(CONFIG_BONDING)
2091 struct adapter *adap;
2093 if (is_vlan_dev(event_dev))
2094 event_dev = vlan_dev_real_dev(event_dev);
2095 #if IS_ENABLED(CONFIG_BONDING)
2096 if (event_dev->flags & IFF_MASTER) {
2097 list_for_each_entry(adap, &adapter_list, list_node) {
2100 cxgb4_clip_get(adap->port[0],
2101 (const u32 *)ifa, 1);
2104 cxgb4_clip_release(adap->port[0],
2105 (const u32 *)ifa, 1);
2116 parent = event_dev->dev.parent;
2118 if (parent && parent->driver == &cxgb4_driver.driver) {
2121 cxgb4_clip_get(event_dev, (const u32 *)ifa, 1);
2124 cxgb4_clip_release(event_dev, (const u32 *)ifa, 1);
2133 static bool inet6addr_registered;
2134 static struct notifier_block cxgb4_inet6addr_notifier = {
2135 .notifier_call = cxgb4_inet6addr_handler
2138 static void update_clip(const struct adapter *adap)
2141 struct net_device *dev;
2146 for (i = 0; i < MAX_NPORTS; i++) {
2147 dev = adap->port[i];
2151 ret = cxgb4_update_root_dev_clip(dev);
2158 #endif /* IS_ENABLED(CONFIG_IPV6) */
2161 * cxgb_up - enable the adapter
2162 * @adap: adapter being enabled
2164 * Called when the first port is enabled, this function performs the
2165 * actions necessary to make an adapter operational, such as completing
2166 * the initialization of HW modules, and enabling interrupts.
2168 * Must be called with the rtnl lock held.
2170 static int cxgb_up(struct adapter *adap)
2174 err = setup_sge_queues(adap);
2177 err = setup_rss(adap);
2181 if (adap->flags & USING_MSIX) {
2182 name_msix_vecs(adap);
2183 err = request_irq(adap->msix_info[0].vec, t4_nondata_intr, 0,
2184 adap->msix_info[0].desc, adap);
2187 err = request_msix_queue_irqs(adap);
2189 free_irq(adap->msix_info[0].vec, adap);
2193 err = request_irq(adap->pdev->irq, t4_intr_handler(adap),
2194 (adap->flags & USING_MSI) ? 0 : IRQF_SHARED,
2195 adap->port[0]->name, adap);
2200 mutex_lock(&uld_mutex);
2203 t4_intr_enable(adap);
2204 adap->flags |= FULL_INIT_DONE;
2205 mutex_unlock(&uld_mutex);
2207 notify_ulds(adap, CXGB4_STATE_UP);
2208 #if IS_ENABLED(CONFIG_IPV6)
2211 /* Initialize hash mac addr list*/
2212 INIT_LIST_HEAD(&adap->mac_hlist);
2216 dev_err(adap->pdev_dev, "request_irq failed, err %d\n", err);
2218 t4_free_sge_resources(adap);
2222 static void cxgb_down(struct adapter *adapter)
2224 cancel_work_sync(&adapter->tid_release_task);
2225 cancel_work_sync(&adapter->db_full_task);
2226 cancel_work_sync(&adapter->db_drop_task);
2227 adapter->tid_release_task_busy = false;
2228 adapter->tid_release_head = NULL;
2230 t4_sge_stop(adapter);
2231 t4_free_sge_resources(adapter);
2232 adapter->flags &= ~FULL_INIT_DONE;
2236 * net_device operations
2238 static int cxgb_open(struct net_device *dev)
2241 struct port_info *pi = netdev_priv(dev);
2242 struct adapter *adapter = pi->adapter;
2244 netif_carrier_off(dev);
2246 if (!(adapter->flags & FULL_INIT_DONE)) {
2247 err = cxgb_up(adapter);
2252 err = link_start(dev);
2254 netif_tx_start_all_queues(dev);
2258 static int cxgb_close(struct net_device *dev)
2260 struct port_info *pi = netdev_priv(dev);
2261 struct adapter *adapter = pi->adapter;
2263 netif_tx_stop_all_queues(dev);
2264 netif_carrier_off(dev);
2265 return t4_enable_vi(adapter, adapter->pf, pi->viid, false, false);
2268 int cxgb4_create_server_filter(const struct net_device *dev, unsigned int stid,
2269 __be32 sip, __be16 sport, __be16 vlan,
2270 unsigned int queue, unsigned char port, unsigned char mask)
2273 struct filter_entry *f;
2274 struct adapter *adap;
2278 adap = netdev2adap(dev);
2280 /* Adjust stid to correct filter index */
2281 stid -= adap->tids.sftid_base;
2282 stid += adap->tids.nftids;
2284 /* Check to make sure the filter requested is writable ...
2286 f = &adap->tids.ftid_tab[stid];
2287 ret = writable_filter(f);
2291 /* Clear out any old resources being used by the filter before
2292 * we start constructing the new filter.
2295 clear_filter(adap, f);
2297 /* Clear out filter specifications */
2298 memset(&f->fs, 0, sizeof(struct ch_filter_specification));
2299 f->fs.val.lport = cpu_to_be16(sport);
2300 f->fs.mask.lport = ~0;
2302 if ((val[0] | val[1] | val[2] | val[3]) != 0) {
2303 for (i = 0; i < 4; i++) {
2304 f->fs.val.lip[i] = val[i];
2305 f->fs.mask.lip[i] = ~0;
2307 if (adap->params.tp.vlan_pri_map & PORT_F) {
2308 f->fs.val.iport = port;
2309 f->fs.mask.iport = mask;
2313 if (adap->params.tp.vlan_pri_map & PROTOCOL_F) {
2314 f->fs.val.proto = IPPROTO_TCP;
2315 f->fs.mask.proto = ~0;
2320 /* Mark filter as locked */
2324 /* Save the actual tid. We need this to get the corresponding
2325 * filter entry structure in filter_rpl.
2327 f->tid = stid + adap->tids.ftid_base;
2328 ret = set_filter_wr(adap, stid);
2330 clear_filter(adap, f);
2336 EXPORT_SYMBOL(cxgb4_create_server_filter);
2338 int cxgb4_remove_server_filter(const struct net_device *dev, unsigned int stid,
2339 unsigned int queue, bool ipv6)
2341 struct filter_entry *f;
2342 struct adapter *adap;
2344 adap = netdev2adap(dev);
2346 /* Adjust stid to correct filter index */
2347 stid -= adap->tids.sftid_base;
2348 stid += adap->tids.nftids;
2350 f = &adap->tids.ftid_tab[stid];
2351 /* Unlock the filter */
2354 return delete_filter(adap, stid);
2356 EXPORT_SYMBOL(cxgb4_remove_server_filter);
2358 static void cxgb_get_stats(struct net_device *dev,
2359 struct rtnl_link_stats64 *ns)
2361 struct port_stats stats;
2362 struct port_info *p = netdev_priv(dev);
2363 struct adapter *adapter = p->adapter;
2365 /* Block retrieving statistics during EEH error
2366 * recovery. Otherwise, the recovery might fail
2367 * and the PCI device will be removed permanently
2369 spin_lock(&adapter->stats_lock);
2370 if (!netif_device_present(dev)) {
2371 spin_unlock(&adapter->stats_lock);
2374 t4_get_port_stats_offset(adapter, p->tx_chan, &stats,
2376 spin_unlock(&adapter->stats_lock);
2378 ns->tx_bytes = stats.tx_octets;
2379 ns->tx_packets = stats.tx_frames;
2380 ns->rx_bytes = stats.rx_octets;
2381 ns->rx_packets = stats.rx_frames;
2382 ns->multicast = stats.rx_mcast_frames;
2384 /* detailed rx_errors */
2385 ns->rx_length_errors = stats.rx_jabber + stats.rx_too_long +
2387 ns->rx_over_errors = 0;
2388 ns->rx_crc_errors = stats.rx_fcs_err;
2389 ns->rx_frame_errors = stats.rx_symbol_err;
2390 ns->rx_dropped = stats.rx_ovflow0 + stats.rx_ovflow1 +
2391 stats.rx_ovflow2 + stats.rx_ovflow3 +
2392 stats.rx_trunc0 + stats.rx_trunc1 +
2393 stats.rx_trunc2 + stats.rx_trunc3;
2394 ns->rx_missed_errors = 0;
2396 /* detailed tx_errors */
2397 ns->tx_aborted_errors = 0;
2398 ns->tx_carrier_errors = 0;
2399 ns->tx_fifo_errors = 0;
2400 ns->tx_heartbeat_errors = 0;
2401 ns->tx_window_errors = 0;
2403 ns->tx_errors = stats.tx_error_frames;
2404 ns->rx_errors = stats.rx_symbol_err + stats.rx_fcs_err +
2405 ns->rx_length_errors + stats.rx_len_err + ns->rx_fifo_errors;
2408 static int cxgb_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
2411 int ret = 0, prtad, devad;
2412 struct port_info *pi = netdev_priv(dev);
2413 struct mii_ioctl_data *data = (struct mii_ioctl_data *)&req->ifr_data;
2417 if (pi->mdio_addr < 0)
2419 data->phy_id = pi->mdio_addr;
2423 if (mdio_phy_id_is_c45(data->phy_id)) {
2424 prtad = mdio_phy_id_prtad(data->phy_id);
2425 devad = mdio_phy_id_devad(data->phy_id);
2426 } else if (data->phy_id < 32) {
2427 prtad = data->phy_id;
2429 data->reg_num &= 0x1f;
2433 mbox = pi->adapter->pf;
2434 if (cmd == SIOCGMIIREG)
2435 ret = t4_mdio_rd(pi->adapter, mbox, prtad, devad,
2436 data->reg_num, &data->val_out);
2438 ret = t4_mdio_wr(pi->adapter, mbox, prtad, devad,
2439 data->reg_num, data->val_in);
2442 return copy_to_user(req->ifr_data, &pi->tstamp_config,
2443 sizeof(pi->tstamp_config)) ?
2446 if (copy_from_user(&pi->tstamp_config, req->ifr_data,
2447 sizeof(pi->tstamp_config)))
2450 switch (pi->tstamp_config.rx_filter) {
2451 case HWTSTAMP_FILTER_NONE:
2452 pi->rxtstamp = false;
2454 case HWTSTAMP_FILTER_ALL:
2455 pi->rxtstamp = true;
2458 pi->tstamp_config.rx_filter = HWTSTAMP_FILTER_NONE;
2462 return copy_to_user(req->ifr_data, &pi->tstamp_config,
2463 sizeof(pi->tstamp_config)) ?
2471 static void cxgb_set_rxmode(struct net_device *dev)
2473 /* unfortunately we can't return errors to the stack */
2474 set_rxmode(dev, -1, false);
2477 static int cxgb_change_mtu(struct net_device *dev, int new_mtu)
2480 struct port_info *pi = netdev_priv(dev);
2482 ret = t4_set_rxmode(pi->adapter, pi->adapter->pf, pi->viid, new_mtu, -1,
2489 #ifdef CONFIG_PCI_IOV
2490 static int dummy_open(struct net_device *dev)
2492 /* Turn carrier off since we don't have to transmit anything on this
2495 netif_carrier_off(dev);
2499 /* Fill MAC address that will be assigned by the FW */
2500 static void fill_vf_station_mac_addr(struct adapter *adap)
2503 u8 hw_addr[ETH_ALEN], macaddr[ETH_ALEN];
2508 err = t4_get_raw_vpd_params(adap, &adap->params.vpd);
2510 na = adap->params.vpd.na;
2511 for (i = 0; i < ETH_ALEN; i++)
2512 hw_addr[i] = (hex2val(na[2 * i + 0]) * 16 +
2513 hex2val(na[2 * i + 1]));
2514 a = (hw_addr[0] << 8) | hw_addr[1];
2515 b = (hw_addr[1] << 8) | hw_addr[2];
2517 a |= 0x0200; /* locally assigned Ethernet MAC address */
2518 a &= ~0x0100; /* not a multicast Ethernet MAC address */
2519 macaddr[0] = a >> 8;
2520 macaddr[1] = a & 0xff;
2522 for (i = 2; i < 5; i++)
2523 macaddr[i] = hw_addr[i + 1];
2525 for (i = 0; i < adap->num_vfs; i++) {
2526 macaddr[5] = adap->pf * 16 + i;
2527 ether_addr_copy(adap->vfinfo[i].vf_mac_addr, macaddr);
2532 static int cxgb_set_vf_mac(struct net_device *dev, int vf, u8 *mac)
2534 struct port_info *pi = netdev_priv(dev);
2535 struct adapter *adap = pi->adapter;
2538 /* verify MAC addr is valid */
2539 if (!is_valid_ether_addr(mac)) {
2540 dev_err(pi->adapter->pdev_dev,
2541 "Invalid Ethernet address %pM for VF %d\n",
2546 dev_info(pi->adapter->pdev_dev,
2547 "Setting MAC %pM on VF %d\n", mac, vf);
2548 ret = t4_set_vf_mac_acl(adap, vf + 1, 1, mac);
2550 ether_addr_copy(adap->vfinfo[vf].vf_mac_addr, mac);
2554 static int cxgb_get_vf_config(struct net_device *dev,
2555 int vf, struct ifla_vf_info *ivi)
2557 struct port_info *pi = netdev_priv(dev);
2558 struct adapter *adap = pi->adapter;
2560 if (vf >= adap->num_vfs)
2563 ether_addr_copy(ivi->mac, adap->vfinfo[vf].vf_mac_addr);
2567 static int cxgb_get_phys_port_id(struct net_device *dev,
2568 struct netdev_phys_item_id *ppid)
2570 struct port_info *pi = netdev_priv(dev);
2571 unsigned int phy_port_id;
2573 phy_port_id = pi->adapter->adap_idx * 10 + pi->port_id;
2574 ppid->id_len = sizeof(phy_port_id);
2575 memcpy(ppid->id, &phy_port_id, ppid->id_len);
2581 static int cxgb_set_mac_addr(struct net_device *dev, void *p)
2584 struct sockaddr *addr = p;
2585 struct port_info *pi = netdev_priv(dev);
2587 if (!is_valid_ether_addr(addr->sa_data))
2588 return -EADDRNOTAVAIL;
2590 ret = t4_change_mac(pi->adapter, pi->adapter->pf, pi->viid,
2591 pi->xact_addr_filt, addr->sa_data, true, true);
2595 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
2596 pi->xact_addr_filt = ret;
2600 #ifdef CONFIG_NET_POLL_CONTROLLER
2601 static void cxgb_netpoll(struct net_device *dev)
2603 struct port_info *pi = netdev_priv(dev);
2604 struct adapter *adap = pi->adapter;
2606 if (adap->flags & USING_MSIX) {
2608 struct sge_eth_rxq *rx = &adap->sge.ethrxq[pi->first_qset];
2610 for (i = pi->nqsets; i; i--, rx++)
2611 t4_sge_intr_msix(0, &rx->rspq);
2613 t4_intr_handler(adap)(0, adap);
2617 static int cxgb_set_tx_maxrate(struct net_device *dev, int index, u32 rate)
2619 struct port_info *pi = netdev_priv(dev);
2620 struct adapter *adap = pi->adapter;
2621 struct sched_class *e;
2622 struct ch_sched_params p;
2623 struct ch_sched_queue qe;
2627 if (!can_sched(dev))
2630 if (index < 0 || index > pi->nqsets - 1)
2633 if (!(adap->flags & FULL_INIT_DONE)) {
2634 dev_err(adap->pdev_dev,
2635 "Failed to rate limit on queue %d. Link Down?\n",
2640 /* Convert from Mbps to Kbps */
2641 req_rate = rate << 10;
2643 /* Max rate is 10 Gbps */
2644 if (req_rate >= SCHED_MAX_RATE_KBPS) {
2645 dev_err(adap->pdev_dev,
2646 "Invalid rate %u Mbps, Max rate is %u Gbps\n",
2647 rate, SCHED_MAX_RATE_KBPS);
2651 /* First unbind the queue from any existing class */
2652 memset(&qe, 0, sizeof(qe));
2654 qe.class = SCHED_CLS_NONE;
2656 err = cxgb4_sched_class_unbind(dev, (void *)(&qe), SCHED_QUEUE);
2658 dev_err(adap->pdev_dev,
2659 "Unbinding Queue %d on port %d fail. Err: %d\n",
2660 index, pi->port_id, err);
2664 /* Queue already unbound */
2668 /* Fetch any available unused or matching scheduling class */
2669 memset(&p, 0, sizeof(p));
2670 p.type = SCHED_CLASS_TYPE_PACKET;
2671 p.u.params.level = SCHED_CLASS_LEVEL_CL_RL;
2672 p.u.params.mode = SCHED_CLASS_MODE_CLASS;
2673 p.u.params.rateunit = SCHED_CLASS_RATEUNIT_BITS;
2674 p.u.params.ratemode = SCHED_CLASS_RATEMODE_ABS;
2675 p.u.params.channel = pi->tx_chan;
2676 p.u.params.class = SCHED_CLS_NONE;
2677 p.u.params.minrate = 0;
2678 p.u.params.maxrate = req_rate;
2679 p.u.params.weight = 0;
2680 p.u.params.pktsize = dev->mtu;
2682 e = cxgb4_sched_class_alloc(dev, &p);
2686 /* Bind the queue to a scheduling class */
2687 memset(&qe, 0, sizeof(qe));
2691 err = cxgb4_sched_class_bind(dev, (void *)(&qe), SCHED_QUEUE);
2693 dev_err(adap->pdev_dev,
2694 "Queue rate limiting failed. Err: %d\n", err);
2698 static int cxgb_setup_tc(struct net_device *dev, u32 handle, __be16 proto,
2699 struct tc_to_netdev *tc)
2701 struct port_info *pi = netdev2pinfo(dev);
2702 struct adapter *adap = netdev2adap(dev);
2704 if (!(adap->flags & FULL_INIT_DONE)) {
2705 dev_err(adap->pdev_dev,
2706 "Failed to setup tc on port %d. Link Down?\n",
2711 if (TC_H_MAJ(handle) == TC_H_MAJ(TC_H_INGRESS) &&
2712 tc->type == TC_SETUP_CLSU32) {
2713 switch (tc->cls_u32->command) {
2714 case TC_CLSU32_NEW_KNODE:
2715 case TC_CLSU32_REPLACE_KNODE:
2716 return cxgb4_config_knode(dev, proto, tc->cls_u32);
2717 case TC_CLSU32_DELETE_KNODE:
2718 return cxgb4_delete_knode(dev, proto, tc->cls_u32);
2727 static const struct net_device_ops cxgb4_netdev_ops = {
2728 .ndo_open = cxgb_open,
2729 .ndo_stop = cxgb_close,
2730 .ndo_start_xmit = t4_eth_xmit,
2731 .ndo_select_queue = cxgb_select_queue,
2732 .ndo_get_stats64 = cxgb_get_stats,
2733 .ndo_set_rx_mode = cxgb_set_rxmode,
2734 .ndo_set_mac_address = cxgb_set_mac_addr,
2735 .ndo_set_features = cxgb_set_features,
2736 .ndo_validate_addr = eth_validate_addr,
2737 .ndo_do_ioctl = cxgb_ioctl,
2738 .ndo_change_mtu = cxgb_change_mtu,
2739 #ifdef CONFIG_NET_POLL_CONTROLLER
2740 .ndo_poll_controller = cxgb_netpoll,
2742 #ifdef CONFIG_CHELSIO_T4_FCOE
2743 .ndo_fcoe_enable = cxgb_fcoe_enable,
2744 .ndo_fcoe_disable = cxgb_fcoe_disable,
2745 #endif /* CONFIG_CHELSIO_T4_FCOE */
2746 .ndo_set_tx_maxrate = cxgb_set_tx_maxrate,
2747 .ndo_setup_tc = cxgb_setup_tc,
2750 #ifdef CONFIG_PCI_IOV
2751 static const struct net_device_ops cxgb4_mgmt_netdev_ops = {
2752 .ndo_open = dummy_open,
2753 .ndo_set_vf_mac = cxgb_set_vf_mac,
2754 .ndo_get_vf_config = cxgb_get_vf_config,
2755 .ndo_get_phys_port_id = cxgb_get_phys_port_id,
2759 static void get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2761 struct adapter *adapter = netdev2adap(dev);
2763 strlcpy(info->driver, cxgb4_driver_name, sizeof(info->driver));
2764 strlcpy(info->version, cxgb4_driver_version,
2765 sizeof(info->version));
2766 strlcpy(info->bus_info, pci_name(adapter->pdev),
2767 sizeof(info->bus_info));
2770 static const struct ethtool_ops cxgb4_mgmt_ethtool_ops = {
2771 .get_drvinfo = get_drvinfo,
2774 void t4_fatal_err(struct adapter *adap)
2778 if (pci_channel_offline(adap->pdev))
2781 /* Disable the SGE since ULDs are going to free resources that
2782 * could be exposed to the adapter. RDMA MWs for example...
2784 t4_shutdown_adapter(adap);
2785 for_each_port(adap, port) {
2786 struct net_device *dev = adap->port[port];
2788 /* If we get here in very early initialization the network
2789 * devices may not have been set up yet.
2794 netif_tx_stop_all_queues(dev);
2795 netif_carrier_off(dev);
2797 dev_alert(adap->pdev_dev, "encountered fatal error, adapter stopped\n");
2800 static void setup_memwin(struct adapter *adap)
2802 u32 nic_win_base = t4_get_util_window(adap);
2804 t4_setup_memwin(adap, nic_win_base, MEMWIN_NIC);
2807 static void setup_memwin_rdma(struct adapter *adap)
2809 if (adap->vres.ocq.size) {
2813 start = t4_read_pcie_cfg4(adap, PCI_BASE_ADDRESS_2);
2814 start &= PCI_BASE_ADDRESS_MEM_MASK;
2815 start += OCQ_WIN_OFFSET(adap->pdev, &adap->vres);
2816 sz_kb = roundup_pow_of_two(adap->vres.ocq.size) >> 10;
2818 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN_A, 3),
2819 start | BIR_V(1) | WINDOW_V(ilog2(sz_kb)));
2821 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET_A, 3),
2822 adap->vres.ocq.start);
2824 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET_A, 3));
2828 static int adap_init1(struct adapter *adap, struct fw_caps_config_cmd *c)
2833 /* get device capabilities */
2834 memset(c, 0, sizeof(*c));
2835 c->op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
2836 FW_CMD_REQUEST_F | FW_CMD_READ_F);
2837 c->cfvalid_to_len16 = htonl(FW_LEN16(*c));
2838 ret = t4_wr_mbox(adap, adap->mbox, c, sizeof(*c), c);
2842 c->op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
2843 FW_CMD_REQUEST_F | FW_CMD_WRITE_F);
2844 ret = t4_wr_mbox(adap, adap->mbox, c, sizeof(*c), NULL);
2848 ret = t4_config_glbl_rss(adap, adap->pf,
2849 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL,
2850 FW_RSS_GLB_CONFIG_CMD_TNLMAPEN_F |
2851 FW_RSS_GLB_CONFIG_CMD_TNLALLLKP_F);
2855 ret = t4_cfg_pfvf(adap, adap->mbox, adap->pf, 0, adap->sge.egr_sz, 64,
2856 MAX_INGQ, 0, 0, 4, 0xf, 0xf, 16, FW_CMD_CAP_PF,
2863 /* tweak some settings */
2864 t4_write_reg(adap, TP_SHIFT_CNT_A, 0x64f8849);
2865 t4_write_reg(adap, ULP_RX_TDDP_PSZ_A, HPZ0_V(PAGE_SHIFT - 12));
2866 t4_write_reg(adap, TP_PIO_ADDR_A, TP_INGRESS_CONFIG_A);
2867 v = t4_read_reg(adap, TP_PIO_DATA_A);
2868 t4_write_reg(adap, TP_PIO_DATA_A, v & ~CSUM_HAS_PSEUDO_HDR_F);
2870 /* first 4 Tx modulation queues point to consecutive Tx channels */
2871 adap->params.tp.tx_modq_map = 0xE4;
2872 t4_write_reg(adap, TP_TX_MOD_QUEUE_REQ_MAP_A,
2873 TX_MOD_QUEUE_REQ_MAP_V(adap->params.tp.tx_modq_map));
2875 /* associate each Tx modulation queue with consecutive Tx channels */
2877 t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
2878 &v, 1, TP_TX_SCHED_HDR_A);
2879 t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
2880 &v, 1, TP_TX_SCHED_FIFO_A);
2881 t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
2882 &v, 1, TP_TX_SCHED_PCMD_A);
2884 #define T4_TX_MODQ_10G_WEIGHT_DEFAULT 16 /* in KB units */
2885 if (is_offload(adap)) {
2886 t4_write_reg(adap, TP_TX_MOD_QUEUE_WEIGHT0_A,
2887 TX_MODQ_WEIGHT0_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
2888 TX_MODQ_WEIGHT1_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
2889 TX_MODQ_WEIGHT2_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
2890 TX_MODQ_WEIGHT3_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
2891 t4_write_reg(adap, TP_TX_MOD_CHANNEL_WEIGHT_A,
2892 TX_MODQ_WEIGHT0_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
2893 TX_MODQ_WEIGHT1_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
2894 TX_MODQ_WEIGHT2_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
2895 TX_MODQ_WEIGHT3_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
2898 /* get basic stuff going */
2899 return t4_early_init(adap, adap->pf);
2903 * Max # of ATIDs. The absolute HW max is 16K but we keep it lower.
2905 #define MAX_ATIDS 8192U
2908 * Phase 0 of initialization: contact FW, obtain config, perform basic init.
2910 * If the firmware we're dealing with has Configuration File support, then
2911 * we use that to perform all configuration
2915 * Tweak configuration based on module parameters, etc. Most of these have
2916 * defaults assigned to them by Firmware Configuration Files (if we're using
2917 * them) but need to be explicitly set if we're using hard-coded
2918 * initialization. But even in the case of using Firmware Configuration
2919 * Files, we'd like to expose the ability to change these via module
2920 * parameters so these are essentially common tweaks/settings for
2921 * Configuration Files and hard-coded initialization ...
2923 static int adap_init0_tweaks(struct adapter *adapter)
2926 * Fix up various Host-Dependent Parameters like Page Size, Cache
2927 * Line Size, etc. The firmware default is for a 4KB Page Size and
2928 * 64B Cache Line Size ...
2930 t4_fixup_host_params(adapter, PAGE_SIZE, L1_CACHE_BYTES);
2933 * Process module parameters which affect early initialization.
2935 if (rx_dma_offset != 2 && rx_dma_offset != 0) {
2936 dev_err(&adapter->pdev->dev,
2937 "Ignoring illegal rx_dma_offset=%d, using 2\n",
2941 t4_set_reg_field(adapter, SGE_CONTROL_A,
2942 PKTSHIFT_V(PKTSHIFT_M),
2943 PKTSHIFT_V(rx_dma_offset));
2946 * Don't include the "IP Pseudo Header" in CPL_RX_PKT checksums: Linux
2947 * adds the pseudo header itself.
2949 t4_tp_wr_bits_indirect(adapter, TP_INGRESS_CONFIG_A,
2950 CSUM_HAS_PSEUDO_HDR_F, 0);
2955 /* 10Gb/s-BT PHY Support. chip-external 10Gb/s-BT PHYs are complex chips
2956 * unto themselves and they contain their own firmware to perform their
2959 static int phy_aq1202_version(const u8 *phy_fw_data,
2964 /* At offset 0x8 you're looking for the primary image's
2965 * starting offset which is 3 Bytes wide
2967 * At offset 0xa of the primary image, you look for the offset
2968 * of the DRAM segment which is 3 Bytes wide.
2970 * The FW version is at offset 0x27e of the DRAM and is 2 Bytes
2973 #define be16(__p) (((__p)[0] << 8) | (__p)[1])
2974 #define le16(__p) ((__p)[0] | ((__p)[1] << 8))
2975 #define le24(__p) (le16(__p) | ((__p)[2] << 16))
2977 offset = le24(phy_fw_data + 0x8) << 12;
2978 offset = le24(phy_fw_data + offset + 0xa);
2979 return be16(phy_fw_data + offset + 0x27e);
2986 static struct info_10gbt_phy_fw {
2987 unsigned int phy_fw_id; /* PCI Device ID */
2988 char *phy_fw_file; /* /lib/firmware/ PHY Firmware file */
2989 int (*phy_fw_version)(const u8 *phy_fw_data, size_t phy_fw_size);
2990 int phy_flash; /* Has FLASH for PHY Firmware */
2991 } phy_info_array[] = {
2993 PHY_AQ1202_DEVICEID,
2994 PHY_AQ1202_FIRMWARE,
2999 PHY_BCM84834_DEVICEID,
3000 PHY_BCM84834_FIRMWARE,
3007 static struct info_10gbt_phy_fw *find_phy_info(int devid)
3011 for (i = 0; i < ARRAY_SIZE(phy_info_array); i++) {
3012 if (phy_info_array[i].phy_fw_id == devid)
3013 return &phy_info_array[i];
3018 /* Handle updating of chip-external 10Gb/s-BT PHY firmware. This needs to
3019 * happen after the FW_RESET_CMD but before the FW_INITIALIZE_CMD. On error
3020 * we return a negative error number. If we transfer new firmware we return 1
3021 * (from t4_load_phy_fw()). If we don't do anything we return 0.
3023 static int adap_init0_phy(struct adapter *adap)
3025 const struct firmware *phyf;
3027 struct info_10gbt_phy_fw *phy_info;
3029 /* Use the device ID to determine which PHY file to flash.
3031 phy_info = find_phy_info(adap->pdev->device);
3033 dev_warn(adap->pdev_dev,
3034 "No PHY Firmware file found for this PHY\n");
3038 /* If we have a T4 PHY firmware file under /lib/firmware/cxgb4/, then
3039 * use that. The adapter firmware provides us with a memory buffer
3040 * where we can load a PHY firmware file from the host if we want to
3041 * override the PHY firmware File in flash.
3043 ret = request_firmware_direct(&phyf, phy_info->phy_fw_file,
3046 /* For adapters without FLASH attached to PHY for their
3047 * firmware, it's obviously a fatal error if we can't get the
3048 * firmware to the adapter. For adapters with PHY firmware
3049 * FLASH storage, it's worth a warning if we can't find the
3050 * PHY Firmware but we'll neuter the error ...
3052 dev_err(adap->pdev_dev, "unable to find PHY Firmware image "
3053 "/lib/firmware/%s, error %d\n",
3054 phy_info->phy_fw_file, -ret);
3055 if (phy_info->phy_flash) {
3056 int cur_phy_fw_ver = 0;
3058 t4_phy_fw_ver(adap, &cur_phy_fw_ver);
3059 dev_warn(adap->pdev_dev, "continuing with, on-adapter "
3060 "FLASH copy, version %#x\n", cur_phy_fw_ver);
3067 /* Load PHY Firmware onto adapter.
3069 ret = t4_load_phy_fw(adap, MEMWIN_NIC, &adap->win0_lock,
3070 phy_info->phy_fw_version,
3071 (u8 *)phyf->data, phyf->size);
3073 dev_err(adap->pdev_dev, "PHY Firmware transfer error %d\n",
3076 int new_phy_fw_ver = 0;
3078 if (phy_info->phy_fw_version)
3079 new_phy_fw_ver = phy_info->phy_fw_version(phyf->data,
3081 dev_info(adap->pdev_dev, "Successfully transferred PHY "
3082 "Firmware /lib/firmware/%s, version %#x\n",
3083 phy_info->phy_fw_file, new_phy_fw_ver);
3086 release_firmware(phyf);
3092 * Attempt to initialize the adapter via a Firmware Configuration File.
3094 static int adap_init0_config(struct adapter *adapter, int reset)
3096 struct fw_caps_config_cmd caps_cmd;
3097 const struct firmware *cf;
3098 unsigned long mtype = 0, maddr = 0;
3099 u32 finiver, finicsum, cfcsum;
3101 int config_issued = 0;
3102 char *fw_config_file, fw_config_file_path[256];
3103 char *config_name = NULL;
3106 * Reset device if necessary.
3109 ret = t4_fw_reset(adapter, adapter->mbox,
3110 PIORSTMODE_F | PIORST_F);
3115 /* If this is a 10Gb/s-BT adapter make sure the chip-external
3116 * 10Gb/s-BT PHYs have up-to-date firmware. Note that this step needs
3117 * to be performed after any global adapter RESET above since some
3118 * PHYs only have local RAM copies of the PHY firmware.
3120 if (is_10gbt_device(adapter->pdev->device)) {
3121 ret = adap_init0_phy(adapter);
3126 * If we have a T4 configuration file under /lib/firmware/cxgb4/,
3127 * then use that. Otherwise, use the configuration file stored
3128 * in the adapter flash ...
3130 switch (CHELSIO_CHIP_VERSION(adapter->params.chip)) {
3132 fw_config_file = FW4_CFNAME;
3135 fw_config_file = FW5_CFNAME;
3138 fw_config_file = FW6_CFNAME;
3141 dev_err(adapter->pdev_dev, "Device %d is not supported\n",
3142 adapter->pdev->device);
3147 ret = request_firmware(&cf, fw_config_file, adapter->pdev_dev);
3149 config_name = "On FLASH";
3150 mtype = FW_MEMTYPE_CF_FLASH;
3151 maddr = t4_flash_cfg_addr(adapter);
3153 u32 params[7], val[7];
3155 sprintf(fw_config_file_path,
3156 "/lib/firmware/%s", fw_config_file);
3157 config_name = fw_config_file_path;
3159 if (cf->size >= FLASH_CFG_MAX_SIZE)
3162 params[0] = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
3163 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_CF));
3164 ret = t4_query_params(adapter, adapter->mbox,
3165 adapter->pf, 0, 1, params, val);
3168 * For t4_memory_rw() below addresses and
3169 * sizes have to be in terms of multiples of 4
3170 * bytes. So, if the Configuration File isn't
3171 * a multiple of 4 bytes in length we'll have
3172 * to write that out separately since we can't
3173 * guarantee that the bytes following the
3174 * residual byte in the buffer returned by
3175 * request_firmware() are zeroed out ...
3177 size_t resid = cf->size & 0x3;
3178 size_t size = cf->size & ~0x3;
3179 __be32 *data = (__be32 *)cf->data;
3181 mtype = FW_PARAMS_PARAM_Y_G(val[0]);
3182 maddr = FW_PARAMS_PARAM_Z_G(val[0]) << 16;
3184 spin_lock(&adapter->win0_lock);
3185 ret = t4_memory_rw(adapter, 0, mtype, maddr,
3186 size, data, T4_MEMORY_WRITE);
3187 if (ret == 0 && resid != 0) {
3194 last.word = data[size >> 2];
3195 for (i = resid; i < 4; i++)
3197 ret = t4_memory_rw(adapter, 0, mtype,
3202 spin_unlock(&adapter->win0_lock);
3206 release_firmware(cf);
3212 * Issue a Capability Configuration command to the firmware to get it
3213 * to parse the Configuration File. We don't use t4_fw_config_file()
3214 * because we want the ability to modify various features after we've
3215 * processed the configuration file ...
3217 memset(&caps_cmd, 0, sizeof(caps_cmd));
3218 caps_cmd.op_to_write =
3219 htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3222 caps_cmd.cfvalid_to_len16 =
3223 htonl(FW_CAPS_CONFIG_CMD_CFVALID_F |
3224 FW_CAPS_CONFIG_CMD_MEMTYPE_CF_V(mtype) |
3225 FW_CAPS_CONFIG_CMD_MEMADDR64K_CF_V(maddr >> 16) |
3226 FW_LEN16(caps_cmd));
3227 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
3230 /* If the CAPS_CONFIG failed with an ENOENT (for a Firmware
3231 * Configuration File in FLASH), our last gasp effort is to use the
3232 * Firmware Configuration File which is embedded in the firmware. A
3233 * very few early versions of the firmware didn't have one embedded
3234 * but we can ignore those.
3236 if (ret == -ENOENT) {
3237 memset(&caps_cmd, 0, sizeof(caps_cmd));
3238 caps_cmd.op_to_write =
3239 htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3242 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
3243 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd,
3244 sizeof(caps_cmd), &caps_cmd);
3245 config_name = "Firmware Default";
3252 finiver = ntohl(caps_cmd.finiver);
3253 finicsum = ntohl(caps_cmd.finicsum);
3254 cfcsum = ntohl(caps_cmd.cfcsum);
3255 if (finicsum != cfcsum)
3256 dev_warn(adapter->pdev_dev, "Configuration File checksum "\
3257 "mismatch: [fini] csum=%#x, computed csum=%#x\n",
3261 * And now tell the firmware to use the configuration we just loaded.
3263 caps_cmd.op_to_write =
3264 htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3267 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
3268 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
3274 * Tweak configuration based on system architecture, module
3277 ret = adap_init0_tweaks(adapter);
3282 * And finally tell the firmware to initialize itself using the
3283 * parameters from the Configuration File.
3285 ret = t4_fw_initialize(adapter, adapter->mbox);
3289 /* Emit Firmware Configuration File information and return
3292 dev_info(adapter->pdev_dev, "Successfully configured using Firmware "\
3293 "Configuration File \"%s\", version %#x, computed checksum %#x\n",
3294 config_name, finiver, cfcsum);
3298 * Something bad happened. Return the error ... (If the "error"
3299 * is that there's no Configuration File on the adapter we don't
3300 * want to issue a warning since this is fairly common.)
3303 if (config_issued && ret != -ENOENT)
3304 dev_warn(adapter->pdev_dev, "\"%s\" configuration file error %d\n",
3309 static struct fw_info fw_info_array[] = {
3312 .fs_name = FW4_CFNAME,
3313 .fw_mod_name = FW4_FNAME,
3315 .chip = FW_HDR_CHIP_T4,
3316 .fw_ver = __cpu_to_be32(FW_VERSION(T4)),
3317 .intfver_nic = FW_INTFVER(T4, NIC),
3318 .intfver_vnic = FW_INTFVER(T4, VNIC),
3319 .intfver_ri = FW_INTFVER(T4, RI),
3320 .intfver_iscsi = FW_INTFVER(T4, ISCSI),
3321 .intfver_fcoe = FW_INTFVER(T4, FCOE),
3325 .fs_name = FW5_CFNAME,
3326 .fw_mod_name = FW5_FNAME,
3328 .chip = FW_HDR_CHIP_T5,
3329 .fw_ver = __cpu_to_be32(FW_VERSION(T5)),
3330 .intfver_nic = FW_INTFVER(T5, NIC),
3331 .intfver_vnic = FW_INTFVER(T5, VNIC),
3332 .intfver_ri = FW_INTFVER(T5, RI),
3333 .intfver_iscsi = FW_INTFVER(T5, ISCSI),
3334 .intfver_fcoe = FW_INTFVER(T5, FCOE),
3338 .fs_name = FW6_CFNAME,
3339 .fw_mod_name = FW6_FNAME,
3341 .chip = FW_HDR_CHIP_T6,
3342 .fw_ver = __cpu_to_be32(FW_VERSION(T6)),
3343 .intfver_nic = FW_INTFVER(T6, NIC),
3344 .intfver_vnic = FW_INTFVER(T6, VNIC),
3345 .intfver_ofld = FW_INTFVER(T6, OFLD),
3346 .intfver_ri = FW_INTFVER(T6, RI),
3347 .intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
3348 .intfver_iscsi = FW_INTFVER(T6, ISCSI),
3349 .intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
3350 .intfver_fcoe = FW_INTFVER(T6, FCOE),
3356 static struct fw_info *find_fw_info(int chip)
3360 for (i = 0; i < ARRAY_SIZE(fw_info_array); i++) {
3361 if (fw_info_array[i].chip == chip)
3362 return &fw_info_array[i];
3368 * Phase 0 of initialization: contact FW, obtain config, perform basic init.
3370 static int adap_init0(struct adapter *adap)
3374 enum dev_state state;
3375 u32 params[7], val[7];
3376 struct fw_caps_config_cmd caps_cmd;
3379 /* Grab Firmware Device Log parameters as early as possible so we have
3380 * access to it for debugging, etc.
3382 ret = t4_init_devlog_params(adap);
3386 /* Contact FW, advertising Master capability */
3387 ret = t4_fw_hello(adap, adap->mbox, adap->mbox,
3388 is_kdump_kernel() ? MASTER_MUST : MASTER_MAY, &state);
3390 dev_err(adap->pdev_dev, "could not connect to FW, error %d\n",
3394 if (ret == adap->mbox)
3395 adap->flags |= MASTER_PF;
3398 * If we're the Master PF Driver and the device is uninitialized,
3399 * then let's consider upgrading the firmware ... (We always want
3400 * to check the firmware version number in order to A. get it for
3401 * later reporting and B. to warn if the currently loaded firmware
3402 * is excessively mismatched relative to the driver.)
3404 t4_get_fw_version(adap, &adap->params.fw_vers);
3405 t4_get_bs_version(adap, &adap->params.bs_vers);
3406 t4_get_tp_version(adap, &adap->params.tp_vers);
3407 t4_get_exprom_version(adap, &adap->params.er_vers);
3409 ret = t4_check_fw_version(adap);
3410 /* If firmware is too old (not supported by driver) force an update. */
3412 state = DEV_STATE_UNINIT;
3413 if ((adap->flags & MASTER_PF) && state != DEV_STATE_INIT) {
3414 struct fw_info *fw_info;
3415 struct fw_hdr *card_fw;
3416 const struct firmware *fw;
3417 const u8 *fw_data = NULL;
3418 unsigned int fw_size = 0;
3420 /* This is the firmware whose headers the driver was compiled
3423 fw_info = find_fw_info(CHELSIO_CHIP_VERSION(adap->params.chip));
3424 if (fw_info == NULL) {
3425 dev_err(adap->pdev_dev,
3426 "unable to get firmware info for chip %d.\n",
3427 CHELSIO_CHIP_VERSION(adap->params.chip));
3431 /* allocate memory to read the header of the firmware on the
3434 card_fw = kvzalloc(sizeof(*card_fw), GFP_KERNEL);
3436 /* Get FW from from /lib/firmware/ */
3437 ret = request_firmware(&fw, fw_info->fw_mod_name,
3440 dev_err(adap->pdev_dev,
3441 "unable to load firmware image %s, error %d\n",
3442 fw_info->fw_mod_name, ret);
3448 /* upgrade FW logic */
3449 ret = t4_prep_fw(adap, fw_info, fw_data, fw_size, card_fw,
3453 release_firmware(fw);
3461 * Grab VPD parameters. This should be done after we establish a
3462 * connection to the firmware since some of the VPD parameters
3463 * (notably the Core Clock frequency) are retrieved via requests to
3464 * the firmware. On the other hand, we need these fairly early on
3465 * so we do this right after getting ahold of the firmware.
3467 ret = t4_get_vpd_params(adap, &adap->params.vpd);
3472 * Find out what ports are available to us. Note that we need to do
3473 * this before calling adap_init0_no_config() since it needs nports
3477 FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
3478 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_PORTVEC);
3479 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, &v, &port_vec);
3483 adap->params.nports = hweight32(port_vec);
3484 adap->params.portvec = port_vec;
3486 /* If the firmware is initialized already, emit a simply note to that
3487 * effect. Otherwise, it's time to try initializing the adapter.
3489 if (state == DEV_STATE_INIT) {
3490 dev_info(adap->pdev_dev, "Coming up as %s: "\
3491 "Adapter already initialized\n",
3492 adap->flags & MASTER_PF ? "MASTER" : "SLAVE");
3494 dev_info(adap->pdev_dev, "Coming up as MASTER: "\
3495 "Initializing adapter\n");
3497 /* Find out whether we're dealing with a version of the
3498 * firmware which has configuration file support.
3500 params[0] = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
3501 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_CF));
3502 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1,
3505 /* If the firmware doesn't support Configuration Files,
3509 dev_err(adap->pdev_dev, "firmware doesn't support "
3510 "Firmware Configuration Files\n");
3514 /* The firmware provides us with a memory buffer where we can
3515 * load a Configuration File from the host if we want to
3516 * override the Configuration File in flash.
3518 ret = adap_init0_config(adap, reset);
3519 if (ret == -ENOENT) {
3520 dev_err(adap->pdev_dev, "no Configuration File "
3521 "present on adapter.\n");
3525 dev_err(adap->pdev_dev, "could not initialize "
3526 "adapter, error %d\n", -ret);
3531 /* Give the SGE code a chance to pull in anything that it needs ...
3532 * Note that this must be called after we retrieve our VPD parameters
3533 * in order to know how to convert core ticks to seconds, etc.
3535 ret = t4_sge_init(adap);
3539 if (is_bypass_device(adap->pdev->device))
3540 adap->params.bypass = 1;
3543 * Grab some of our basic fundamental operating parameters.
3545 #define FW_PARAM_DEV(param) \
3546 (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) | \
3547 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_##param))
3549 #define FW_PARAM_PFVF(param) \
3550 FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) | \
3551 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_##param)| \
3552 FW_PARAMS_PARAM_Y_V(0) | \
3553 FW_PARAMS_PARAM_Z_V(0)
3555 params[0] = FW_PARAM_PFVF(EQ_START);
3556 params[1] = FW_PARAM_PFVF(L2T_START);
3557 params[2] = FW_PARAM_PFVF(L2T_END);
3558 params[3] = FW_PARAM_PFVF(FILTER_START);
3559 params[4] = FW_PARAM_PFVF(FILTER_END);
3560 params[5] = FW_PARAM_PFVF(IQFLINT_START);
3561 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6, params, val);
3564 adap->sge.egr_start = val[0];
3565 adap->l2t_start = val[1];
3566 adap->l2t_end = val[2];
3567 adap->tids.ftid_base = val[3];
3568 adap->tids.nftids = val[4] - val[3] + 1;
3569 adap->sge.ingr_start = val[5];
3571 /* qids (ingress/egress) returned from firmware can be anywhere
3572 * in the range from EQ(IQFLINT)_START to EQ(IQFLINT)_END.
3573 * Hence driver needs to allocate memory for this range to
3574 * store the queue info. Get the highest IQFLINT/EQ index returned
3575 * in FW_EQ_*_CMD.alloc command.
3577 params[0] = FW_PARAM_PFVF(EQ_END);
3578 params[1] = FW_PARAM_PFVF(IQFLINT_END);
3579 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
3582 adap->sge.egr_sz = val[0] - adap->sge.egr_start + 1;
3583 adap->sge.ingr_sz = val[1] - adap->sge.ingr_start + 1;
3585 adap->sge.egr_map = kcalloc(adap->sge.egr_sz,
3586 sizeof(*adap->sge.egr_map), GFP_KERNEL);
3587 if (!adap->sge.egr_map) {
3592 adap->sge.ingr_map = kcalloc(adap->sge.ingr_sz,
3593 sizeof(*adap->sge.ingr_map), GFP_KERNEL);
3594 if (!adap->sge.ingr_map) {
3599 /* Allocate the memory for the vaious egress queue bitmaps
3600 * ie starving_fl, txq_maperr and blocked_fl.
3602 adap->sge.starving_fl = kcalloc(BITS_TO_LONGS(adap->sge.egr_sz),
3603 sizeof(long), GFP_KERNEL);
3604 if (!adap->sge.starving_fl) {
3609 adap->sge.txq_maperr = kcalloc(BITS_TO_LONGS(adap->sge.egr_sz),
3610 sizeof(long), GFP_KERNEL);
3611 if (!adap->sge.txq_maperr) {
3616 #ifdef CONFIG_DEBUG_FS
3617 adap->sge.blocked_fl = kcalloc(BITS_TO_LONGS(adap->sge.egr_sz),
3618 sizeof(long), GFP_KERNEL);
3619 if (!adap->sge.blocked_fl) {
3625 params[0] = FW_PARAM_PFVF(CLIP_START);
3626 params[1] = FW_PARAM_PFVF(CLIP_END);
3627 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
3630 adap->clipt_start = val[0];
3631 adap->clipt_end = val[1];
3633 /* We don't yet have a PARAMs calls to retrieve the number of Traffic
3634 * Classes supported by the hardware/firmware so we hard code it here
3637 adap->params.nsched_cls = is_t4(adap->params.chip) ? 15 : 16;
3639 /* query params related to active filter region */
3640 params[0] = FW_PARAM_PFVF(ACTIVE_FILTER_START);
3641 params[1] = FW_PARAM_PFVF(ACTIVE_FILTER_END);
3642 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
3643 /* If Active filter size is set we enable establishing
3644 * offload connection through firmware work request
3646 if ((val[0] != val[1]) && (ret >= 0)) {
3647 adap->flags |= FW_OFLD_CONN;
3648 adap->tids.aftid_base = val[0];
3649 adap->tids.aftid_end = val[1];
3652 /* If we're running on newer firmware, let it know that we're
3653 * prepared to deal with encapsulated CPL messages. Older
3654 * firmware won't understand this and we'll just get
3655 * unencapsulated messages ...
3657 params[0] = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
3659 (void)t4_set_params(adap, adap->mbox, adap->pf, 0, 1, params, val);
3662 * Find out whether we're allowed to use the T5+ ULPTX MEMWRITE DSGL
3663 * capability. Earlier versions of the firmware didn't have the
3664 * ULPTX_MEMWRITE_DSGL so we'll interpret a query failure as no
3665 * permission to use ULPTX MEMWRITE DSGL.
3667 if (is_t4(adap->params.chip)) {
3668 adap->params.ulptx_memwrite_dsgl = false;
3670 params[0] = FW_PARAM_DEV(ULPTX_MEMWRITE_DSGL);
3671 ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
3673 adap->params.ulptx_memwrite_dsgl = (ret == 0 && val[0] != 0);
3676 /* See if FW supports FW_RI_FR_NSMR_TPTE_WR work request */
3677 params[0] = FW_PARAM_DEV(RI_FR_NSMR_TPTE_WR);
3678 ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
3680 adap->params.fr_nsmr_tpte_wr_support = (ret == 0 && val[0] != 0);
3683 * Get device capabilities so we can determine what resources we need
3686 memset(&caps_cmd, 0, sizeof(caps_cmd));
3687 caps_cmd.op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3688 FW_CMD_REQUEST_F | FW_CMD_READ_F);
3689 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
3690 ret = t4_wr_mbox(adap, adap->mbox, &caps_cmd, sizeof(caps_cmd),
3695 if (caps_cmd.ofldcaps) {
3696 /* query offload-related parameters */
3697 params[0] = FW_PARAM_DEV(NTID);
3698 params[1] = FW_PARAM_PFVF(SERVER_START);
3699 params[2] = FW_PARAM_PFVF(SERVER_END);
3700 params[3] = FW_PARAM_PFVF(TDDP_START);
3701 params[4] = FW_PARAM_PFVF(TDDP_END);
3702 params[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3703 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6,
3707 adap->tids.ntids = val[0];
3708 adap->tids.natids = min(adap->tids.ntids / 2, MAX_ATIDS);
3709 adap->tids.stid_base = val[1];
3710 adap->tids.nstids = val[2] - val[1] + 1;
3712 * Setup server filter region. Divide the available filter
3713 * region into two parts. Regular filters get 1/3rd and server
3714 * filters get 2/3rd part. This is only enabled if workarond
3716 * 1. For regular filters.
3717 * 2. Server filter: This are special filters which are used
3718 * to redirect SYN packets to offload queue.
3720 if (adap->flags & FW_OFLD_CONN && !is_bypass(adap)) {
3721 adap->tids.sftid_base = adap->tids.ftid_base +
3722 DIV_ROUND_UP(adap->tids.nftids, 3);
3723 adap->tids.nsftids = adap->tids.nftids -
3724 DIV_ROUND_UP(adap->tids.nftids, 3);
3725 adap->tids.nftids = adap->tids.sftid_base -
3726 adap->tids.ftid_base;
3728 adap->vres.ddp.start = val[3];
3729 adap->vres.ddp.size = val[4] - val[3] + 1;
3730 adap->params.ofldq_wr_cred = val[5];
3732 adap->params.offload = 1;
3733 adap->num_ofld_uld += 1;
3735 if (caps_cmd.rdmacaps) {
3736 params[0] = FW_PARAM_PFVF(STAG_START);
3737 params[1] = FW_PARAM_PFVF(STAG_END);
3738 params[2] = FW_PARAM_PFVF(RQ_START);
3739 params[3] = FW_PARAM_PFVF(RQ_END);
3740 params[4] = FW_PARAM_PFVF(PBL_START);
3741 params[5] = FW_PARAM_PFVF(PBL_END);
3742 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6,
3746 adap->vres.stag.start = val[0];
3747 adap->vres.stag.size = val[1] - val[0] + 1;
3748 adap->vres.rq.start = val[2];
3749 adap->vres.rq.size = val[3] - val[2] + 1;
3750 adap->vres.pbl.start = val[4];
3751 adap->vres.pbl.size = val[5] - val[4] + 1;
3753 params[0] = FW_PARAM_PFVF(SQRQ_START);
3754 params[1] = FW_PARAM_PFVF(SQRQ_END);
3755 params[2] = FW_PARAM_PFVF(CQ_START);
3756 params[3] = FW_PARAM_PFVF(CQ_END);
3757 params[4] = FW_PARAM_PFVF(OCQ_START);
3758 params[5] = FW_PARAM_PFVF(OCQ_END);
3759 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6, params,
3763 adap->vres.qp.start = val[0];
3764 adap->vres.qp.size = val[1] - val[0] + 1;
3765 adap->vres.cq.start = val[2];
3766 adap->vres.cq.size = val[3] - val[2] + 1;
3767 adap->vres.ocq.start = val[4];
3768 adap->vres.ocq.size = val[5] - val[4] + 1;
3770 params[0] = FW_PARAM_DEV(MAXORDIRD_QP);
3771 params[1] = FW_PARAM_DEV(MAXIRD_ADAPTER);
3772 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params,
3775 adap->params.max_ordird_qp = 8;
3776 adap->params.max_ird_adapter = 32 * adap->tids.ntids;
3779 adap->params.max_ordird_qp = val[0];
3780 adap->params.max_ird_adapter = val[1];
3782 dev_info(adap->pdev_dev,
3783 "max_ordird_qp %d max_ird_adapter %d\n",
3784 adap->params.max_ordird_qp,
3785 adap->params.max_ird_adapter);
3786 adap->num_ofld_uld += 2;
3788 if (caps_cmd.iscsicaps) {
3789 params[0] = FW_PARAM_PFVF(ISCSI_START);
3790 params[1] = FW_PARAM_PFVF(ISCSI_END);
3791 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
3795 adap->vres.iscsi.start = val[0];
3796 adap->vres.iscsi.size = val[1] - val[0] + 1;
3797 /* LIO target and cxgb4i initiaitor */
3798 adap->num_ofld_uld += 2;
3800 if (caps_cmd.cryptocaps) {
3801 /* Should query params here...TODO */
3802 params[0] = FW_PARAM_PFVF(NCRYPTO_LOOKASIDE);
3803 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
3809 adap->vres.ncrypto_fc = val[0];
3811 adap->params.crypto |= ULP_CRYPTO_LOOKASIDE;
3814 #undef FW_PARAM_PFVF
3817 /* The MTU/MSS Table is initialized by now, so load their values. If
3818 * we're initializing the adapter, then we'll make any modifications
3819 * we want to the MTU/MSS Table and also initialize the congestion
3822 t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
3823 if (state != DEV_STATE_INIT) {
3826 /* The default MTU Table contains values 1492 and 1500.
3827 * However, for TCP, it's better to have two values which are
3828 * a multiple of 8 +/- 4 bytes apart near this popular MTU.
3829 * This allows us to have a TCP Data Payload which is a
3830 * multiple of 8 regardless of what combination of TCP Options
3831 * are in use (always a multiple of 4 bytes) which is
3832 * important for performance reasons. For instance, if no
3833 * options are in use, then we have a 20-byte IP header and a
3834 * 20-byte TCP header. In this case, a 1500-byte MSS would
3835 * result in a TCP Data Payload of 1500 - 40 == 1460 bytes
3836 * which is not a multiple of 8. So using an MSS of 1488 in
3837 * this case results in a TCP Data Payload of 1448 bytes which
3838 * is a multiple of 8. On the other hand, if 12-byte TCP Time
3839 * Stamps have been negotiated, then an MTU of 1500 bytes
3840 * results in a TCP Data Payload of 1448 bytes which, as
3841 * above, is a multiple of 8 bytes ...
3843 for (i = 0; i < NMTUS; i++)
3844 if (adap->params.mtus[i] == 1492) {
3845 adap->params.mtus[i] = 1488;
3849 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
3850 adap->params.b_wnd);
3852 t4_init_sge_params(adap);
3853 adap->flags |= FW_OK;
3854 t4_init_tp_params(adap);
3858 * Something bad happened. If a command timed out or failed with EIO
3859 * FW does not operate within its spec or something catastrophic
3860 * happened to HW/FW, stop issuing commands.
3863 kfree(adap->sge.egr_map);
3864 kfree(adap->sge.ingr_map);
3865 kfree(adap->sge.starving_fl);
3866 kfree(adap->sge.txq_maperr);
3867 #ifdef CONFIG_DEBUG_FS
3868 kfree(adap->sge.blocked_fl);
3870 if (ret != -ETIMEDOUT && ret != -EIO)
3871 t4_fw_bye(adap, adap->mbox);
3877 static pci_ers_result_t eeh_err_detected(struct pci_dev *pdev,
3878 pci_channel_state_t state)
3881 struct adapter *adap = pci_get_drvdata(pdev);
3887 adap->flags &= ~FW_OK;
3888 notify_ulds(adap, CXGB4_STATE_START_RECOVERY);
3889 spin_lock(&adap->stats_lock);
3890 for_each_port(adap, i) {
3891 struct net_device *dev = adap->port[i];
3893 netif_device_detach(dev);
3894 netif_carrier_off(dev);
3897 spin_unlock(&adap->stats_lock);
3898 disable_interrupts(adap);
3899 if (adap->flags & FULL_INIT_DONE)
3902 if ((adap->flags & DEV_ENABLED)) {
3903 pci_disable_device(pdev);
3904 adap->flags &= ~DEV_ENABLED;
3906 out: return state == pci_channel_io_perm_failure ?
3907 PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
3910 static pci_ers_result_t eeh_slot_reset(struct pci_dev *pdev)
3913 struct fw_caps_config_cmd c;
3914 struct adapter *adap = pci_get_drvdata(pdev);
3917 pci_restore_state(pdev);
3918 pci_save_state(pdev);
3919 return PCI_ERS_RESULT_RECOVERED;
3922 if (!(adap->flags & DEV_ENABLED)) {
3923 if (pci_enable_device(pdev)) {
3924 dev_err(&pdev->dev, "Cannot reenable PCI "
3925 "device after reset\n");
3926 return PCI_ERS_RESULT_DISCONNECT;
3928 adap->flags |= DEV_ENABLED;
3931 pci_set_master(pdev);
3932 pci_restore_state(pdev);
3933 pci_save_state(pdev);
3934 pci_cleanup_aer_uncorrect_error_status(pdev);
3936 if (t4_wait_dev_ready(adap->regs) < 0)
3937 return PCI_ERS_RESULT_DISCONNECT;
3938 if (t4_fw_hello(adap, adap->mbox, adap->pf, MASTER_MUST, NULL) < 0)
3939 return PCI_ERS_RESULT_DISCONNECT;
3940 adap->flags |= FW_OK;
3941 if (adap_init1(adap, &c))
3942 return PCI_ERS_RESULT_DISCONNECT;
3944 for_each_port(adap, i) {
3945 struct port_info *p = adap2pinfo(adap, i);
3947 ret = t4_alloc_vi(adap, adap->mbox, p->tx_chan, adap->pf, 0, 1,
3950 return PCI_ERS_RESULT_DISCONNECT;
3952 p->xact_addr_filt = -1;
3955 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
3956 adap->params.b_wnd);
3959 return PCI_ERS_RESULT_DISCONNECT;
3960 return PCI_ERS_RESULT_RECOVERED;
3963 static void eeh_resume(struct pci_dev *pdev)
3966 struct adapter *adap = pci_get_drvdata(pdev);
3972 for_each_port(adap, i) {
3973 struct net_device *dev = adap->port[i];
3975 if (netif_running(dev)) {
3977 cxgb_set_rxmode(dev);
3979 netif_device_attach(dev);
3985 static const struct pci_error_handlers cxgb4_eeh = {
3986 .error_detected = eeh_err_detected,
3987 .slot_reset = eeh_slot_reset,
3988 .resume = eeh_resume,
3991 /* Return true if the Link Configuration supports "High Speeds" (those greater
3994 static inline bool is_x_10g_port(const struct link_config *lc)
3996 unsigned int speeds, high_speeds;
3998 speeds = FW_PORT_CAP_SPEED_V(FW_PORT_CAP_SPEED_G(lc->supported));
3999 high_speeds = speeds & ~(FW_PORT_CAP_SPEED_100M | FW_PORT_CAP_SPEED_1G);
4001 return high_speeds != 0;
4005 * Perform default configuration of DMA queues depending on the number and type
4006 * of ports we found and the number of available CPUs. Most settings can be
4007 * modified by the admin prior to actual use.
4009 static void cfg_queues(struct adapter *adap)
4011 struct sge *s = &adap->sge;
4012 int i = 0, n10g = 0, qidx = 0;
4013 #ifndef CONFIG_CHELSIO_T4_DCB
4017 /* Reduce memory usage in kdump environment, disable all offload.
4019 if (is_kdump_kernel()) {
4020 adap->params.offload = 0;
4021 adap->params.crypto = 0;
4022 } else if (is_uld(adap) && t4_uld_mem_alloc(adap)) {
4023 adap->params.offload = 0;
4024 adap->params.crypto = 0;
4027 n10g += is_x_10g_port(&adap2pinfo(adap, i)->link_cfg);
4028 #ifdef CONFIG_CHELSIO_T4_DCB
4029 /* For Data Center Bridging support we need to be able to support up
4030 * to 8 Traffic Priorities; each of which will be assigned to its
4031 * own TX Queue in order to prevent Head-Of-Line Blocking.
4033 if (adap->params.nports * 8 > MAX_ETH_QSETS) {
4034 dev_err(adap->pdev_dev, "MAX_ETH_QSETS=%d < %d!\n",
4035 MAX_ETH_QSETS, adap->params.nports * 8);
4039 for_each_port(adap, i) {
4040 struct port_info *pi = adap2pinfo(adap, i);
4042 pi->first_qset = qidx;
4046 #else /* !CONFIG_CHELSIO_T4_DCB */
4048 * We default to 1 queue per non-10G port and up to # of cores queues
4052 q10g = (MAX_ETH_QSETS - (adap->params.nports - n10g)) / n10g;
4053 if (q10g > netif_get_num_default_rss_queues())
4054 q10g = netif_get_num_default_rss_queues();
4056 for_each_port(adap, i) {
4057 struct port_info *pi = adap2pinfo(adap, i);
4059 pi->first_qset = qidx;
4060 pi->nqsets = is_x_10g_port(&pi->link_cfg) ? q10g : 1;
4063 #endif /* !CONFIG_CHELSIO_T4_DCB */
4066 s->max_ethqsets = qidx; /* MSI-X may lower it later */
4070 * For offload we use 1 queue/channel if all ports are up to 1G,
4071 * otherwise we divide all available queues amongst the channels
4072 * capped by the number of available cores.
4075 i = min_t(int, MAX_OFLD_QSETS, num_online_cpus());
4076 s->ofldqsets = roundup(i, adap->params.nports);
4078 s->ofldqsets = adap->params.nports;
4082 for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
4083 struct sge_eth_rxq *r = &s->ethrxq[i];
4085 init_rspq(adap, &r->rspq, 5, 10, 1024, 64);
4089 for (i = 0; i < ARRAY_SIZE(s->ethtxq); i++)
4090 s->ethtxq[i].q.size = 1024;
4092 for (i = 0; i < ARRAY_SIZE(s->ctrlq); i++)
4093 s->ctrlq[i].q.size = 512;
4095 init_rspq(adap, &s->fw_evtq, 0, 1, 1024, 64);
4096 init_rspq(adap, &s->intrq, 0, 1, 512, 64);
4100 * Reduce the number of Ethernet queues across all ports to at most n.
4101 * n provides at least one queue per port.
4103 static void reduce_ethqs(struct adapter *adap, int n)
4106 struct port_info *pi;
4108 while (n < adap->sge.ethqsets)
4109 for_each_port(adap, i) {
4110 pi = adap2pinfo(adap, i);
4111 if (pi->nqsets > 1) {
4113 adap->sge.ethqsets--;
4114 if (adap->sge.ethqsets <= n)
4120 for_each_port(adap, i) {
4121 pi = adap2pinfo(adap, i);
4127 static int get_msix_info(struct adapter *adap)
4129 struct uld_msix_info *msix_info;
4130 unsigned int max_ingq = 0;
4132 if (is_offload(adap))
4133 max_ingq += MAX_OFLD_QSETS * adap->num_ofld_uld;
4134 if (is_pci_uld(adap))
4135 max_ingq += MAX_OFLD_QSETS * adap->num_uld;
4140 msix_info = kcalloc(max_ingq, sizeof(*msix_info), GFP_KERNEL);
4144 adap->msix_bmap_ulds.msix_bmap = kcalloc(BITS_TO_LONGS(max_ingq),
4145 sizeof(long), GFP_KERNEL);
4146 if (!adap->msix_bmap_ulds.msix_bmap) {
4150 spin_lock_init(&adap->msix_bmap_ulds.lock);
4151 adap->msix_info_ulds = msix_info;
4156 static void free_msix_info(struct adapter *adap)
4158 if (!(adap->num_uld && adap->num_ofld_uld))
4161 kfree(adap->msix_info_ulds);
4162 kfree(adap->msix_bmap_ulds.msix_bmap);
4165 /* 2 MSI-X vectors needed for the FW queue and non-data interrupts */
4166 #define EXTRA_VECS 2
4168 static int enable_msix(struct adapter *adap)
4170 int ofld_need = 0, uld_need = 0;
4171 int i, j, want, need, allocated;
4172 struct sge *s = &adap->sge;
4173 unsigned int nchan = adap->params.nports;
4174 struct msix_entry *entries;
4175 int max_ingq = MAX_INGQ;
4177 if (is_pci_uld(adap))
4178 max_ingq += (MAX_OFLD_QSETS * adap->num_uld);
4179 if (is_offload(adap))
4180 max_ingq += (MAX_OFLD_QSETS * adap->num_ofld_uld);
4181 entries = kmalloc(sizeof(*entries) * (max_ingq + 1),
4187 if (get_msix_info(adap)) {
4188 adap->params.offload = 0;
4189 adap->params.crypto = 0;
4192 for (i = 0; i < max_ingq + 1; ++i)
4193 entries[i].entry = i;
4195 want = s->max_ethqsets + EXTRA_VECS;
4196 if (is_offload(adap)) {
4197 want += adap->num_ofld_uld * s->ofldqsets;
4198 ofld_need = adap->num_ofld_uld * nchan;
4200 if (is_pci_uld(adap)) {
4201 want += adap->num_uld * s->ofldqsets;
4202 uld_need = adap->num_uld * nchan;
4204 #ifdef CONFIG_CHELSIO_T4_DCB
4205 /* For Data Center Bridging we need 8 Ethernet TX Priority Queues for
4208 need = 8 * adap->params.nports + EXTRA_VECS + ofld_need + uld_need;
4210 need = adap->params.nports + EXTRA_VECS + ofld_need + uld_need;
4212 allocated = pci_enable_msix_range(adap->pdev, entries, need, want);
4213 if (allocated < 0) {
4214 dev_info(adap->pdev_dev, "not enough MSI-X vectors left,"
4215 " not using MSI-X\n");
4220 /* Distribute available vectors to the various queue groups.
4221 * Every group gets its minimum requirement and NIC gets top
4222 * priority for leftovers.
4224 i = allocated - EXTRA_VECS - ofld_need - uld_need;
4225 if (i < s->max_ethqsets) {
4226 s->max_ethqsets = i;
4227 if (i < s->ethqsets)
4228 reduce_ethqs(adap, i);
4231 if (allocated < want)
4232 s->nqs_per_uld = nchan;
4234 s->nqs_per_uld = s->ofldqsets;
4237 for (i = 0; i < (s->max_ethqsets + EXTRA_VECS); ++i)
4238 adap->msix_info[i].vec = entries[i].vector;
4240 for (j = 0 ; i < allocated; ++i, j++) {
4241 adap->msix_info_ulds[j].vec = entries[i].vector;
4242 adap->msix_info_ulds[j].idx = i;
4244 adap->msix_bmap_ulds.mapsize = j;
4246 dev_info(adap->pdev_dev, "%d MSI-X vectors allocated, "
4247 "nic %d per uld %d\n",
4248 allocated, s->max_ethqsets, s->nqs_per_uld);
4256 static int init_rss(struct adapter *adap)
4261 err = t4_init_rss_mode(adap, adap->mbox);
4265 for_each_port(adap, i) {
4266 struct port_info *pi = adap2pinfo(adap, i);
4268 pi->rss = kcalloc(pi->rss_size, sizeof(u16), GFP_KERNEL);
4275 static int cxgb4_get_pcie_dev_link_caps(struct adapter *adap,
4276 enum pci_bus_speed *speed,
4277 enum pcie_link_width *width)
4279 u32 lnkcap1, lnkcap2;
4282 #define PCIE_MLW_CAP_SHIFT 4 /* start of MLW mask in link capabilities */
4284 *speed = PCI_SPEED_UNKNOWN;
4285 *width = PCIE_LNK_WIDTH_UNKNOWN;
4287 err1 = pcie_capability_read_dword(adap->pdev, PCI_EXP_LNKCAP,
4289 err2 = pcie_capability_read_dword(adap->pdev, PCI_EXP_LNKCAP2,
4291 if (!err2 && lnkcap2) { /* PCIe r3.0-compliant */
4292 if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_8_0GB)
4293 *speed = PCIE_SPEED_8_0GT;
4294 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_5_0GB)
4295 *speed = PCIE_SPEED_5_0GT;
4296 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_2_5GB)
4297 *speed = PCIE_SPEED_2_5GT;
4300 *width = (lnkcap1 & PCI_EXP_LNKCAP_MLW) >> PCIE_MLW_CAP_SHIFT;
4301 if (!lnkcap2) { /* pre-r3.0 */
4302 if (lnkcap1 & PCI_EXP_LNKCAP_SLS_5_0GB)
4303 *speed = PCIE_SPEED_5_0GT;
4304 else if (lnkcap1 & PCI_EXP_LNKCAP_SLS_2_5GB)
4305 *speed = PCIE_SPEED_2_5GT;
4309 if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN)
4310 return err1 ? err1 : err2 ? err2 : -EINVAL;
4314 static void cxgb4_check_pcie_caps(struct adapter *adap)
4316 enum pcie_link_width width, width_cap;
4317 enum pci_bus_speed speed, speed_cap;
4319 #define PCIE_SPEED_STR(speed) \
4320 (speed == PCIE_SPEED_8_0GT ? "8.0GT/s" : \
4321 speed == PCIE_SPEED_5_0GT ? "5.0GT/s" : \
4322 speed == PCIE_SPEED_2_5GT ? "2.5GT/s" : \
4325 if (cxgb4_get_pcie_dev_link_caps(adap, &speed_cap, &width_cap)) {
4326 dev_warn(adap->pdev_dev,
4327 "Unable to determine PCIe device BW capabilities\n");
4331 if (pcie_get_minimum_link(adap->pdev, &speed, &width) ||
4332 speed == PCI_SPEED_UNKNOWN || width == PCIE_LNK_WIDTH_UNKNOWN) {
4333 dev_warn(adap->pdev_dev,
4334 "Unable to determine PCI Express bandwidth.\n");
4338 dev_info(adap->pdev_dev, "PCIe link speed is %s, device supports %s\n",
4339 PCIE_SPEED_STR(speed), PCIE_SPEED_STR(speed_cap));
4340 dev_info(adap->pdev_dev, "PCIe link width is x%d, device supports x%d\n",
4342 if (speed < speed_cap || width < width_cap)
4343 dev_info(adap->pdev_dev,
4344 "A slot with more lanes and/or higher speed is "
4345 "suggested for optimal performance.\n");
4348 /* Dump basic information about the adapter */
4349 static void print_adapter_info(struct adapter *adapter)
4351 /* Device information */
4352 dev_info(adapter->pdev_dev, "Chelsio %s rev %d\n",
4353 adapter->params.vpd.id,
4354 CHELSIO_CHIP_RELEASE(adapter->params.chip));
4355 dev_info(adapter->pdev_dev, "S/N: %s, P/N: %s\n",
4356 adapter->params.vpd.sn, adapter->params.vpd.pn);
4358 /* Firmware Version */
4359 if (!adapter->params.fw_vers)
4360 dev_warn(adapter->pdev_dev, "No firmware loaded\n");
4362 dev_info(adapter->pdev_dev, "Firmware version: %u.%u.%u.%u\n",
4363 FW_HDR_FW_VER_MAJOR_G(adapter->params.fw_vers),
4364 FW_HDR_FW_VER_MINOR_G(adapter->params.fw_vers),
4365 FW_HDR_FW_VER_MICRO_G(adapter->params.fw_vers),
4366 FW_HDR_FW_VER_BUILD_G(adapter->params.fw_vers));
4368 /* Bootstrap Firmware Version. (Some adapters don't have Bootstrap
4369 * Firmware, so dev_info() is more appropriate here.)
4371 if (!adapter->params.bs_vers)
4372 dev_info(adapter->pdev_dev, "No bootstrap loaded\n");
4374 dev_info(adapter->pdev_dev, "Bootstrap version: %u.%u.%u.%u\n",
4375 FW_HDR_FW_VER_MAJOR_G(adapter->params.bs_vers),
4376 FW_HDR_FW_VER_MINOR_G(adapter->params.bs_vers),
4377 FW_HDR_FW_VER_MICRO_G(adapter->params.bs_vers),
4378 FW_HDR_FW_VER_BUILD_G(adapter->params.bs_vers));
4380 /* TP Microcode Version */
4381 if (!adapter->params.tp_vers)
4382 dev_warn(adapter->pdev_dev, "No TP Microcode loaded\n");
4384 dev_info(adapter->pdev_dev,
4385 "TP Microcode version: %u.%u.%u.%u\n",
4386 FW_HDR_FW_VER_MAJOR_G(adapter->params.tp_vers),
4387 FW_HDR_FW_VER_MINOR_G(adapter->params.tp_vers),
4388 FW_HDR_FW_VER_MICRO_G(adapter->params.tp_vers),
4389 FW_HDR_FW_VER_BUILD_G(adapter->params.tp_vers));
4391 /* Expansion ROM version */
4392 if (!adapter->params.er_vers)
4393 dev_info(adapter->pdev_dev, "No Expansion ROM loaded\n");
4395 dev_info(adapter->pdev_dev,
4396 "Expansion ROM version: %u.%u.%u.%u\n",
4397 FW_HDR_FW_VER_MAJOR_G(adapter->params.er_vers),
4398 FW_HDR_FW_VER_MINOR_G(adapter->params.er_vers),
4399 FW_HDR_FW_VER_MICRO_G(adapter->params.er_vers),
4400 FW_HDR_FW_VER_BUILD_G(adapter->params.er_vers));
4402 /* Software/Hardware configuration */
4403 dev_info(adapter->pdev_dev, "Configuration: %sNIC %s, %s capable\n",
4404 is_offload(adapter) ? "R" : "",
4405 ((adapter->flags & USING_MSIX) ? "MSI-X" :
4406 (adapter->flags & USING_MSI) ? "MSI" : ""),
4407 is_offload(adapter) ? "Offload" : "non-Offload");
4410 static void print_port_info(const struct net_device *dev)
4414 const char *spd = "";
4415 const struct port_info *pi = netdev_priv(dev);
4416 const struct adapter *adap = pi->adapter;
4418 if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_2_5GB)
4420 else if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_5_0GB)
4422 else if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_8_0GB)
4425 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_100M)
4426 bufp += sprintf(bufp, "100M/");
4427 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_1G)
4428 bufp += sprintf(bufp, "1G/");
4429 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_10G)
4430 bufp += sprintf(bufp, "10G/");
4431 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_25G)
4432 bufp += sprintf(bufp, "25G/");
4433 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_40G)
4434 bufp += sprintf(bufp, "40G/");
4435 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_100G)
4436 bufp += sprintf(bufp, "100G/");
4439 sprintf(bufp, "BASE-%s", t4_get_port_type_description(pi->port_type));
4441 netdev_info(dev, "%s: Chelsio %s (%s) %s\n",
4442 dev->name, adap->params.vpd.id, adap->name, buf);
4445 static void enable_pcie_relaxed_ordering(struct pci_dev *dev)
4447 pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_RELAX_EN);
4451 * Free the following resources:
4452 * - memory used for tables
4455 * - resources FW is holding for us
4457 static void free_some_resources(struct adapter *adapter)
4461 kvfree(adapter->l2t);
4462 t4_cleanup_sched(adapter);
4463 kvfree(adapter->tids.tid_tab);
4464 cxgb4_cleanup_tc_u32(adapter);
4465 kfree(adapter->sge.egr_map);
4466 kfree(adapter->sge.ingr_map);
4467 kfree(adapter->sge.starving_fl);
4468 kfree(adapter->sge.txq_maperr);
4469 #ifdef CONFIG_DEBUG_FS
4470 kfree(adapter->sge.blocked_fl);
4472 disable_msi(adapter);
4474 for_each_port(adapter, i)
4475 if (adapter->port[i]) {
4476 struct port_info *pi = adap2pinfo(adapter, i);
4479 t4_free_vi(adapter, adapter->mbox, adapter->pf,
4481 kfree(adap2pinfo(adapter, i)->rss);
4482 free_netdev(adapter->port[i]);
4484 if (adapter->flags & FW_OK)
4485 t4_fw_bye(adapter, adapter->pf);
4488 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
4489 #define VLAN_FEAT (NETIF_F_SG | NETIF_F_IP_CSUM | TSO_FLAGS | \
4490 NETIF_F_IPV6_CSUM | NETIF_F_HIGHDMA)
4491 #define SEGMENT_SIZE 128
4493 static int get_chip_type(struct pci_dev *pdev, u32 pl_rev)
4497 /* Retrieve adapter's device ID */
4498 pci_read_config_word(pdev, PCI_DEVICE_ID, &device_id);
4500 switch (device_id >> 12) {
4502 return CHELSIO_CHIP_CODE(CHELSIO_T4, pl_rev);
4504 return CHELSIO_CHIP_CODE(CHELSIO_T5, pl_rev);
4506 return CHELSIO_CHIP_CODE(CHELSIO_T6, pl_rev);
4508 dev_err(&pdev->dev, "Device %d is not supported\n",
4514 #ifdef CONFIG_PCI_IOV
4515 static void dummy_setup(struct net_device *dev)
4517 dev->type = ARPHRD_NONE;
4519 dev->hard_header_len = 0;
4521 dev->tx_queue_len = 0;
4522 dev->flags |= IFF_NOARP;
4523 dev->priv_flags |= IFF_NO_QUEUE;
4525 /* Initialize the device structure. */
4526 dev->netdev_ops = &cxgb4_mgmt_netdev_ops;
4527 dev->ethtool_ops = &cxgb4_mgmt_ethtool_ops;
4528 dev->needs_free_netdev = true;
4531 static int config_mgmt_dev(struct pci_dev *pdev)
4533 struct adapter *adap = pci_get_drvdata(pdev);
4534 struct net_device *netdev;
4535 struct port_info *pi;
4536 char name[IFNAMSIZ];
4539 snprintf(name, IFNAMSIZ, "mgmtpf%d%d", adap->adap_idx, adap->pf);
4540 netdev = alloc_netdev(sizeof(struct port_info), name, NET_NAME_UNKNOWN,
4545 pi = netdev_priv(netdev);
4547 pi->port_id = adap->pf % adap->params.nports;
4548 SET_NETDEV_DEV(netdev, &pdev->dev);
4550 adap->port[0] = netdev;
4552 err = register_netdev(adap->port[0]);
4554 pr_info("Unable to register VF mgmt netdev %s\n", name);
4555 free_netdev(adap->port[0]);
4556 adap->port[0] = NULL;
4562 static int cxgb4_iov_configure(struct pci_dev *pdev, int num_vfs)
4564 struct adapter *adap = pci_get_drvdata(pdev);
4566 int current_vfs = pci_num_vf(pdev);
4569 pcie_fw = readl(adap->regs + PCIE_FW_A);
4570 /* Check if cxgb4 is the MASTER and fw is initialized */
4571 if (!(pcie_fw & PCIE_FW_INIT_F) ||
4572 !(pcie_fw & PCIE_FW_MASTER_VLD_F) ||
4573 PCIE_FW_MASTER_G(pcie_fw) != 4) {
4574 dev_warn(&pdev->dev,
4575 "cxgb4 driver needs to be MASTER to support SRIOV\n");
4579 /* If any of the VF's is already assigned to Guest OS, then
4580 * SRIOV for the same cannot be modified
4582 if (current_vfs && pci_vfs_assigned(pdev)) {
4584 "Cannot modify SR-IOV while VFs are assigned\n");
4585 num_vfs = current_vfs;
4589 /* Disable SRIOV when zero is passed.
4590 * One needs to disable SRIOV before modifying it, else
4591 * stack throws the below warning:
4592 * " 'n' VFs already enabled. Disable before enabling 'm' VFs."
4595 pci_disable_sriov(pdev);
4596 if (adap->port[0]) {
4597 unregister_netdev(adap->port[0]);
4598 adap->port[0] = NULL;
4600 /* free VF resources */
4601 kfree(adap->vfinfo);
4602 adap->vfinfo = NULL;
4607 if (num_vfs != current_vfs) {
4608 err = pci_enable_sriov(pdev, num_vfs);
4612 adap->num_vfs = num_vfs;
4613 err = config_mgmt_dev(pdev);
4618 adap->vfinfo = kcalloc(adap->num_vfs,
4619 sizeof(struct vf_info), GFP_KERNEL);
4621 fill_vf_station_mac_addr(adap);
4626 static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
4628 int func, i, err, s_qpp, qpp, num_seg;
4629 struct port_info *pi;
4630 bool highdma = false;
4631 struct adapter *adapter = NULL;
4632 struct net_device *netdev;
4635 enum chip_type chip;
4636 static int adap_idx = 1;
4637 #ifdef CONFIG_PCI_IOV
4641 printk_once(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
4643 err = pci_request_regions(pdev, KBUILD_MODNAME);
4645 /* Just info, some other driver may have claimed the device. */
4646 dev_info(&pdev->dev, "cannot obtain PCI resources\n");
4650 err = pci_enable_device(pdev);
4652 dev_err(&pdev->dev, "cannot enable PCI device\n");
4653 goto out_release_regions;
4656 regs = pci_ioremap_bar(pdev, 0);
4658 dev_err(&pdev->dev, "cannot map device registers\n");
4660 goto out_disable_device;
4663 err = t4_wait_dev_ready(regs);
4665 goto out_unmap_bar0;
4667 /* We control everything through one PF */
4668 whoami = readl(regs + PL_WHOAMI_A);
4669 pl_rev = REV_G(readl(regs + PL_REV_A));
4670 chip = get_chip_type(pdev, pl_rev);
4671 func = CHELSIO_CHIP_VERSION(chip) <= CHELSIO_T5 ?
4672 SOURCEPF_G(whoami) : T6_SOURCEPF_G(whoami);
4673 if (func != ent->driver_data) {
4674 #ifndef CONFIG_PCI_IOV
4677 pci_disable_device(pdev);
4678 pci_save_state(pdev); /* to restore SR-IOV later */
4682 if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
4684 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
4686 dev_err(&pdev->dev, "unable to obtain 64-bit DMA for "
4687 "coherent allocations\n");
4688 goto out_unmap_bar0;
4691 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
4693 dev_err(&pdev->dev, "no usable DMA configuration\n");
4694 goto out_unmap_bar0;
4698 pci_enable_pcie_error_reporting(pdev);
4699 enable_pcie_relaxed_ordering(pdev);
4700 pci_set_master(pdev);
4701 pci_save_state(pdev);
4703 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
4706 goto out_unmap_bar0;
4710 adapter->workq = create_singlethread_workqueue("cxgb4");
4711 if (!adapter->workq) {
4713 goto out_free_adapter;
4716 adapter->mbox_log = kzalloc(sizeof(*adapter->mbox_log) +
4717 (sizeof(struct mbox_cmd) *
4718 T4_OS_LOG_MBOX_CMDS),
4720 if (!adapter->mbox_log) {
4722 goto out_free_adapter;
4724 adapter->mbox_log->size = T4_OS_LOG_MBOX_CMDS;
4726 /* PCI device has been enabled */
4727 adapter->flags |= DEV_ENABLED;
4729 adapter->regs = regs;
4730 adapter->pdev = pdev;
4731 adapter->pdev_dev = &pdev->dev;
4732 adapter->name = pci_name(pdev);
4733 adapter->mbox = func;
4735 adapter->msg_enable = DFLT_MSG_ENABLE;
4736 memset(adapter->chan_map, 0xff, sizeof(adapter->chan_map));
4738 spin_lock_init(&adapter->stats_lock);
4739 spin_lock_init(&adapter->tid_release_lock);
4740 spin_lock_init(&adapter->win0_lock);
4741 spin_lock_init(&adapter->mbox_lock);
4743 INIT_LIST_HEAD(&adapter->mlist.list);
4745 INIT_WORK(&adapter->tid_release_task, process_tid_release_list);
4746 INIT_WORK(&adapter->db_full_task, process_db_full);
4747 INIT_WORK(&adapter->db_drop_task, process_db_drop);
4749 err = t4_prep_adapter(adapter);
4751 goto out_free_adapter;
4754 if (!is_t4(adapter->params.chip)) {
4755 s_qpp = (QUEUESPERPAGEPF0_S +
4756 (QUEUESPERPAGEPF1_S - QUEUESPERPAGEPF0_S) *
4758 qpp = 1 << QUEUESPERPAGEPF0_G(t4_read_reg(adapter,
4759 SGE_EGRESS_QUEUES_PER_PAGE_PF_A) >> s_qpp);
4760 num_seg = PAGE_SIZE / SEGMENT_SIZE;
4762 /* Each segment size is 128B. Write coalescing is enabled only
4763 * when SGE_EGRESS_QUEUES_PER_PAGE_PF reg value for the
4764 * queue is less no of segments that can be accommodated in
4767 if (qpp > num_seg) {
4769 "Incorrect number of egress queues per page\n");
4771 goto out_free_adapter;
4773 adapter->bar2 = ioremap_wc(pci_resource_start(pdev, 2),
4774 pci_resource_len(pdev, 2));
4775 if (!adapter->bar2) {
4776 dev_err(&pdev->dev, "cannot map device bar2 region\n");
4778 goto out_free_adapter;
4782 setup_memwin(adapter);
4783 err = adap_init0(adapter);
4784 #ifdef CONFIG_DEBUG_FS
4785 bitmap_zero(adapter->sge.blocked_fl, adapter->sge.egr_sz);
4787 setup_memwin_rdma(adapter);
4791 /* configure SGE_STAT_CFG_A to read WC stats */
4792 if (!is_t4(adapter->params.chip))
4793 t4_write_reg(adapter, SGE_STAT_CFG_A, STATSOURCE_T5_V(7) |
4794 (is_t5(adapter->params.chip) ? STATMODE_V(0) :
4797 for_each_port(adapter, i) {
4798 netdev = alloc_etherdev_mq(sizeof(struct port_info),
4805 SET_NETDEV_DEV(netdev, &pdev->dev);
4807 adapter->port[i] = netdev;
4808 pi = netdev_priv(netdev);
4809 pi->adapter = adapter;
4810 pi->xact_addr_filt = -1;
4812 netdev->irq = pdev->irq;
4814 netdev->hw_features = NETIF_F_SG | TSO_FLAGS |
4815 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
4816 NETIF_F_RXCSUM | NETIF_F_RXHASH |
4817 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
4820 netdev->hw_features |= NETIF_F_HIGHDMA;
4821 netdev->features |= netdev->hw_features;
4822 netdev->vlan_features = netdev->features & VLAN_FEAT;
4824 netdev->priv_flags |= IFF_UNICAST_FLT;
4826 /* MTU range: 81 - 9600 */
4827 netdev->min_mtu = 81;
4828 netdev->max_mtu = MAX_MTU;
4830 netdev->netdev_ops = &cxgb4_netdev_ops;
4831 #ifdef CONFIG_CHELSIO_T4_DCB
4832 netdev->dcbnl_ops = &cxgb4_dcb_ops;
4833 cxgb4_dcb_state_init(netdev);
4835 cxgb4_set_ethtool_ops(netdev);
4838 pci_set_drvdata(pdev, adapter);
4840 if (adapter->flags & FW_OK) {
4841 err = t4_port_init(adapter, func, func, 0);
4844 } else if (adapter->params.nports == 1) {
4845 /* If we don't have a connection to the firmware -- possibly
4846 * because of an error -- grab the raw VPD parameters so we
4847 * can set the proper MAC Address on the debug network
4848 * interface that we've created.
4850 u8 hw_addr[ETH_ALEN];
4851 u8 *na = adapter->params.vpd.na;
4853 err = t4_get_raw_vpd_params(adapter, &adapter->params.vpd);
4855 for (i = 0; i < ETH_ALEN; i++)
4856 hw_addr[i] = (hex2val(na[2 * i + 0]) * 16 +
4857 hex2val(na[2 * i + 1]));
4858 t4_set_hw_addr(adapter, 0, hw_addr);
4862 /* Configure queues and allocate tables now, they can be needed as
4863 * soon as the first register_netdev completes.
4865 cfg_queues(adapter);
4867 adapter->l2t = t4_init_l2t(adapter->l2t_start, adapter->l2t_end);
4868 if (!adapter->l2t) {
4869 /* We tolerate a lack of L2T, giving up some functionality */
4870 dev_warn(&pdev->dev, "could not allocate L2T, continuing\n");
4871 adapter->params.offload = 0;
4874 #if IS_ENABLED(CONFIG_IPV6)
4875 if ((CHELSIO_CHIP_VERSION(adapter->params.chip) <= CHELSIO_T5) &&
4876 (!(t4_read_reg(adapter, LE_DB_CONFIG_A) & ASLIPCOMPEN_F))) {
4877 /* CLIP functionality is not present in hardware,
4878 * hence disable all offload features
4880 dev_warn(&pdev->dev,
4881 "CLIP not enabled in hardware, continuing\n");
4882 adapter->params.offload = 0;
4884 adapter->clipt = t4_init_clip_tbl(adapter->clipt_start,
4885 adapter->clipt_end);
4886 if (!adapter->clipt) {
4887 /* We tolerate a lack of clip_table, giving up
4888 * some functionality
4890 dev_warn(&pdev->dev,
4891 "could not allocate Clip table, continuing\n");
4892 adapter->params.offload = 0;
4897 for_each_port(adapter, i) {
4898 pi = adap2pinfo(adapter, i);
4899 pi->sched_tbl = t4_init_sched(adapter->params.nsched_cls);
4901 dev_warn(&pdev->dev,
4902 "could not activate scheduling on port %d\n",
4906 if (tid_init(&adapter->tids) < 0) {
4907 dev_warn(&pdev->dev, "could not allocate TID table, "
4909 adapter->params.offload = 0;
4911 adapter->tc_u32 = cxgb4_init_tc_u32(adapter);
4912 if (!adapter->tc_u32)
4913 dev_warn(&pdev->dev,
4914 "could not offload tc u32, continuing\n");
4917 if (is_offload(adapter)) {
4918 if (t4_read_reg(adapter, LE_DB_CONFIG_A) & HASHEN_F) {
4919 u32 hash_base, hash_reg;
4921 if (chip <= CHELSIO_T5) {
4922 hash_reg = LE_DB_TID_HASHBASE_A;
4923 hash_base = t4_read_reg(adapter, hash_reg);
4924 adapter->tids.hash_base = hash_base / 4;
4926 hash_reg = T6_LE_DB_HASH_TID_BASE_A;
4927 hash_base = t4_read_reg(adapter, hash_reg);
4928 adapter->tids.hash_base = hash_base;
4933 /* See what interrupts we'll be using */
4934 if (msi > 1 && enable_msix(adapter) == 0)
4935 adapter->flags |= USING_MSIX;
4936 else if (msi > 0 && pci_enable_msi(pdev) == 0) {
4937 adapter->flags |= USING_MSI;
4939 free_msix_info(adapter);
4942 /* check for PCI Express bandwidth capabiltites */
4943 cxgb4_check_pcie_caps(adapter);
4945 err = init_rss(adapter);
4950 * The card is now ready to go. If any errors occur during device
4951 * registration we do not fail the whole card but rather proceed only
4952 * with the ports we manage to register successfully. However we must
4953 * register at least one net device.
4955 for_each_port(adapter, i) {
4956 pi = adap2pinfo(adapter, i);
4957 adapter->port[i]->dev_port = pi->lport;
4958 netif_set_real_num_tx_queues(adapter->port[i], pi->nqsets);
4959 netif_set_real_num_rx_queues(adapter->port[i], pi->nqsets);
4961 err = register_netdev(adapter->port[i]);
4964 adapter->chan_map[pi->tx_chan] = i;
4965 print_port_info(adapter->port[i]);
4968 dev_err(&pdev->dev, "could not register any net devices\n");
4972 dev_warn(&pdev->dev, "only %d net devices registered\n", i);
4976 if (cxgb4_debugfs_root) {
4977 adapter->debugfs_root = debugfs_create_dir(pci_name(pdev),
4978 cxgb4_debugfs_root);
4979 setup_debugfs(adapter);
4982 /* PCIe EEH recovery on powerpc platforms needs fundamental reset */
4983 pdev->needs_freset = 1;
4985 if (is_uld(adapter)) {
4986 mutex_lock(&uld_mutex);
4987 list_add_tail(&adapter->list_node, &adapter_list);
4988 mutex_unlock(&uld_mutex);
4991 print_adapter_info(adapter);
4992 setup_fw_sge_queues(adapter);
4996 #ifdef CONFIG_PCI_IOV
4997 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
5000 goto free_pci_region;
5003 adapter->pdev = pdev;
5004 adapter->pdev_dev = &pdev->dev;
5005 adapter->name = pci_name(pdev);
5006 adapter->mbox = func;
5008 adapter->regs = regs;
5009 adapter->adap_idx = adap_idx;
5010 adapter->mbox_log = kzalloc(sizeof(*adapter->mbox_log) +
5011 (sizeof(struct mbox_cmd) *
5012 T4_OS_LOG_MBOX_CMDS),
5014 if (!adapter->mbox_log) {
5018 spin_lock_init(&adapter->mbox_lock);
5019 INIT_LIST_HEAD(&adapter->mlist.list);
5021 v = FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
5022 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_PORTVEC);
5023 err = t4_query_params(adapter, adapter->mbox, adapter->pf, 0, 1,
5026 dev_err(adapter->pdev_dev, "Could not fetch port params\n");
5030 adapter->params.nports = hweight32(port_vec);
5031 pci_set_drvdata(pdev, adapter);
5038 pci_disable_sriov(pdev);
5039 pci_release_regions(pdev);
5046 free_some_resources(adapter);
5047 if (adapter->flags & USING_MSIX)
5048 free_msix_info(adapter);
5049 if (adapter->num_uld || adapter->num_ofld_uld)
5050 t4_uld_mem_free(adapter);
5052 if (!is_t4(adapter->params.chip))
5053 iounmap(adapter->bar2);
5056 destroy_workqueue(adapter->workq);
5058 kfree(adapter->mbox_log);
5063 pci_disable_pcie_error_reporting(pdev);
5064 pci_disable_device(pdev);
5065 out_release_regions:
5066 pci_release_regions(pdev);
5070 static void remove_one(struct pci_dev *pdev)
5072 struct adapter *adapter = pci_get_drvdata(pdev);
5075 pci_release_regions(pdev);
5079 if (adapter->pf == 4) {
5082 /* Tear down per-adapter Work Queue first since it can contain
5083 * references to our adapter data structure.
5085 destroy_workqueue(adapter->workq);
5087 if (is_uld(adapter))
5088 detach_ulds(adapter);
5090 disable_interrupts(adapter);
5092 for_each_port(adapter, i)
5093 if (adapter->port[i]->reg_state == NETREG_REGISTERED)
5094 unregister_netdev(adapter->port[i]);
5096 debugfs_remove_recursive(adapter->debugfs_root);
5098 /* If we allocated filters, free up state associated with any
5101 clear_all_filters(adapter);
5103 if (adapter->flags & FULL_INIT_DONE)
5106 if (adapter->flags & USING_MSIX)
5107 free_msix_info(adapter);
5108 if (adapter->num_uld || adapter->num_ofld_uld)
5109 t4_uld_mem_free(adapter);
5110 free_some_resources(adapter);
5111 #if IS_ENABLED(CONFIG_IPV6)
5112 t4_cleanup_clip_tbl(adapter);
5114 iounmap(adapter->regs);
5115 if (!is_t4(adapter->params.chip))
5116 iounmap(adapter->bar2);
5117 pci_disable_pcie_error_reporting(pdev);
5118 if ((adapter->flags & DEV_ENABLED)) {
5119 pci_disable_device(pdev);
5120 adapter->flags &= ~DEV_ENABLED;
5122 pci_release_regions(pdev);
5123 kfree(adapter->mbox_log);
5127 #ifdef CONFIG_PCI_IOV
5129 if (adapter->port[0])
5130 unregister_netdev(adapter->port[0]);
5131 iounmap(adapter->regs);
5132 kfree(adapter->vfinfo);
5134 pci_disable_sriov(pdev);
5135 pci_release_regions(pdev);
5140 /* "Shutdown" quiesces the device, stopping Ingress Packet and Interrupt
5141 * delivery. This is essentially a stripped down version of the PCI remove()
5142 * function where we do the minimal amount of work necessary to shutdown any
5145 static void shutdown_one(struct pci_dev *pdev)
5147 struct adapter *adapter = pci_get_drvdata(pdev);
5149 /* As with remove_one() above (see extended comment), we only want do
5150 * do cleanup on PCI Devices which went all the way through init_one()
5154 pci_release_regions(pdev);
5158 if (adapter->pf == 4) {
5161 for_each_port(adapter, i)
5162 if (adapter->port[i]->reg_state == NETREG_REGISTERED)
5163 cxgb_close(adapter->port[i]);
5165 t4_uld_clean_up(adapter);
5166 disable_interrupts(adapter);
5167 disable_msi(adapter);
5169 t4_sge_stop(adapter);
5170 if (adapter->flags & FW_OK)
5171 t4_fw_bye(adapter, adapter->mbox);
5173 #ifdef CONFIG_PCI_IOV
5175 if (adapter->port[0])
5176 unregister_netdev(adapter->port[0]);
5177 iounmap(adapter->regs);
5178 kfree(adapter->vfinfo);
5180 pci_disable_sriov(pdev);
5181 pci_release_regions(pdev);
5186 static struct pci_driver cxgb4_driver = {
5187 .name = KBUILD_MODNAME,
5188 .id_table = cxgb4_pci_tbl,
5190 .remove = remove_one,
5191 .shutdown = shutdown_one,
5192 #ifdef CONFIG_PCI_IOV
5193 .sriov_configure = cxgb4_iov_configure,
5195 .err_handler = &cxgb4_eeh,
5198 static int __init cxgb4_init_module(void)
5202 /* Debugfs support is optional, just warn if this fails */
5203 cxgb4_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
5204 if (!cxgb4_debugfs_root)
5205 pr_warn("could not create debugfs entry, continuing\n");
5207 ret = pci_register_driver(&cxgb4_driver);
5209 debugfs_remove(cxgb4_debugfs_root);
5211 #if IS_ENABLED(CONFIG_IPV6)
5212 if (!inet6addr_registered) {
5213 register_inet6addr_notifier(&cxgb4_inet6addr_notifier);
5214 inet6addr_registered = true;
5221 static void __exit cxgb4_cleanup_module(void)
5223 #if IS_ENABLED(CONFIG_IPV6)
5224 if (inet6addr_registered) {
5225 unregister_inet6addr_notifier(&cxgb4_inet6addr_notifier);
5226 inet6addr_registered = false;
5229 pci_unregister_driver(&cxgb4_driver);
5230 debugfs_remove(cxgb4_debugfs_root); /* NULL ok */
5233 module_init(cxgb4_init_module);
5234 module_exit(cxgb4_cleanup_module);