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 <linux/uaccess.h>
66 #include <linux/crash_dump.h>
67 #include <net/udp_tunnel.h>
71 #include "cxgb4_filter.h"
73 #include "t4_values.h"
76 #include "t4fw_version.h"
77 #include "cxgb4_dcb.h"
79 #include "cxgb4_debugfs.h"
84 #include "cxgb4_tc_u32.h"
85 #include "cxgb4_tc_flower.h"
86 #include "cxgb4_tc_mqprio.h"
87 #include "cxgb4_tc_matchall.h"
88 #include "cxgb4_ptp.h"
89 #include "cxgb4_cudbg.h"
91 char cxgb4_driver_name[] = KBUILD_MODNAME;
93 #define DRV_DESC "Chelsio T4/T5/T6 Network Driver"
95 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
96 NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
97 NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
99 /* Macros needed to support the PCI Device ID Table ...
101 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_BEGIN \
102 static const struct pci_device_id cxgb4_pci_tbl[] = {
103 #define CXGB4_UNIFIED_PF 0x4
105 #define CH_PCI_DEVICE_ID_FUNCTION CXGB4_UNIFIED_PF
107 /* Include PCI Device IDs for both PF4 and PF0-3 so our PCI probe() routine is
110 #define CH_PCI_DEVICE_ID_FUNCTION2 0x0
112 #define CH_PCI_ID_TABLE_ENTRY(devid) \
113 {PCI_VDEVICE(CHELSIO, (devid)), CXGB4_UNIFIED_PF}
115 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_END \
119 #include "t4_pci_id_tbl.h"
121 #define FW4_FNAME "cxgb4/t4fw.bin"
122 #define FW5_FNAME "cxgb4/t5fw.bin"
123 #define FW6_FNAME "cxgb4/t6fw.bin"
124 #define FW4_CFNAME "cxgb4/t4-config.txt"
125 #define FW5_CFNAME "cxgb4/t5-config.txt"
126 #define FW6_CFNAME "cxgb4/t6-config.txt"
127 #define PHY_AQ1202_FIRMWARE "cxgb4/aq1202_fw.cld"
128 #define PHY_BCM84834_FIRMWARE "cxgb4/bcm8483.bin"
129 #define PHY_AQ1202_DEVICEID 0x4409
130 #define PHY_BCM84834_DEVICEID 0x4486
132 MODULE_DESCRIPTION(DRV_DESC);
133 MODULE_AUTHOR("Chelsio Communications");
134 MODULE_LICENSE("Dual BSD/GPL");
135 MODULE_DEVICE_TABLE(pci, cxgb4_pci_tbl);
136 MODULE_FIRMWARE(FW4_FNAME);
137 MODULE_FIRMWARE(FW5_FNAME);
138 MODULE_FIRMWARE(FW6_FNAME);
141 * The driver uses the best interrupt scheme available on a platform in the
142 * order MSI-X, MSI, legacy INTx interrupts. This parameter determines which
143 * of these schemes the driver may consider as follows:
145 * msi = 2: choose from among all three options
146 * msi = 1: only consider MSI and INTx interrupts
147 * msi = 0: force INTx interrupts
151 module_param(msi, int, 0644);
152 MODULE_PARM_DESC(msi, "whether to use INTx (0), MSI (1) or MSI-X (2)");
155 * Normally we tell the chip to deliver Ingress Packets into our DMA buffers
156 * offset by 2 bytes in order to have the IP headers line up on 4-byte
157 * boundaries. This is a requirement for many architectures which will throw
158 * a machine check fault if an attempt is made to access one of the 4-byte IP
159 * header fields on a non-4-byte boundary. And it's a major performance issue
160 * even on some architectures which allow it like some implementations of the
161 * x86 ISA. However, some architectures don't mind this and for some very
162 * edge-case performance sensitive applications (like forwarding large volumes
163 * of small packets), setting this DMA offset to 0 will decrease the number of
164 * PCI-E Bus transfers enough to measurably affect performance.
166 static int rx_dma_offset = 2;
168 /* TX Queue select used to determine what algorithm to use for selecting TX
169 * queue. Select between the kernel provided function (select_queue=0) or user
170 * cxgb_select_queue function (select_queue=1)
172 * Default: select_queue=0
174 static int select_queue;
175 module_param(select_queue, int, 0644);
176 MODULE_PARM_DESC(select_queue,
177 "Select between kernel provided method of selecting or driver method of selecting TX queue. Default is kernel method.");
179 static struct dentry *cxgb4_debugfs_root;
181 LIST_HEAD(adapter_list);
182 DEFINE_MUTEX(uld_mutex);
184 static int cfg_queues(struct adapter *adap);
186 static void link_report(struct net_device *dev)
188 if (!netif_carrier_ok(dev))
189 netdev_info(dev, "link down\n");
191 static const char *fc[] = { "no", "Rx", "Tx", "Tx/Rx" };
194 const struct port_info *p = netdev_priv(dev);
196 switch (p->link_cfg.speed) {
219 pr_info("%s: unsupported speed: %d\n",
220 dev->name, p->link_cfg.speed);
224 netdev_info(dev, "link up, %s, full-duplex, %s PAUSE\n", s,
229 #ifdef CONFIG_CHELSIO_T4_DCB
230 /* Set up/tear down Data Center Bridging Priority mapping for a net device. */
231 static void dcb_tx_queue_prio_enable(struct net_device *dev, int enable)
233 struct port_info *pi = netdev_priv(dev);
234 struct adapter *adap = pi->adapter;
235 struct sge_eth_txq *txq = &adap->sge.ethtxq[pi->first_qset];
238 /* We use a simple mapping of Port TX Queue Index to DCB
239 * Priority when we're enabling DCB.
241 for (i = 0; i < pi->nqsets; i++, txq++) {
245 name = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DMAQ) |
247 FW_PARAMS_PARAM_DMAQ_EQ_DCBPRIO_ETH) |
248 FW_PARAMS_PARAM_YZ_V(txq->q.cntxt_id));
249 value = enable ? i : 0xffffffff;
251 /* Since we can be called while atomic (from "interrupt
252 * level") we need to issue the Set Parameters Commannd
253 * without sleeping (timeout < 0).
255 err = t4_set_params_timeout(adap, adap->mbox, adap->pf, 0, 1,
257 -FW_CMD_MAX_TIMEOUT);
260 dev_err(adap->pdev_dev,
261 "Can't %s DCB Priority on port %d, TX Queue %d: err=%d\n",
262 enable ? "set" : "unset", pi->port_id, i, -err);
264 txq->dcb_prio = enable ? value : 0;
268 int cxgb4_dcb_enabled(const struct net_device *dev)
270 struct port_info *pi = netdev_priv(dev);
272 if (!pi->dcb.enabled)
275 return ((pi->dcb.state == CXGB4_DCB_STATE_FW_ALLSYNCED) ||
276 (pi->dcb.state == CXGB4_DCB_STATE_HOST));
278 #endif /* CONFIG_CHELSIO_T4_DCB */
280 void t4_os_link_changed(struct adapter *adapter, int port_id, int link_stat)
282 struct net_device *dev = adapter->port[port_id];
284 /* Skip changes from disabled ports. */
285 if (netif_running(dev) && link_stat != netif_carrier_ok(dev)) {
287 netif_carrier_on(dev);
289 #ifdef CONFIG_CHELSIO_T4_DCB
290 if (cxgb4_dcb_enabled(dev)) {
291 cxgb4_dcb_reset(dev);
292 dcb_tx_queue_prio_enable(dev, false);
294 #endif /* CONFIG_CHELSIO_T4_DCB */
295 netif_carrier_off(dev);
302 void t4_os_portmod_changed(struct adapter *adap, int port_id)
304 static const char *mod_str[] = {
305 NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
308 struct net_device *dev = adap->port[port_id];
309 struct port_info *pi = netdev_priv(dev);
311 if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
312 netdev_info(dev, "port module unplugged\n");
313 else if (pi->mod_type < ARRAY_SIZE(mod_str))
314 netdev_info(dev, "%s module inserted\n", mod_str[pi->mod_type]);
315 else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
316 netdev_info(dev, "%s: unsupported port module inserted\n",
318 else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
319 netdev_info(dev, "%s: unknown port module inserted\n",
321 else if (pi->mod_type == FW_PORT_MOD_TYPE_ERROR)
322 netdev_info(dev, "%s: transceiver module error\n", dev->name);
324 netdev_info(dev, "%s: unknown module type %d inserted\n",
325 dev->name, pi->mod_type);
327 /* If the interface is running, then we'll need any "sticky" Link
328 * Parameters redone with a new Transceiver Module.
330 pi->link_cfg.redo_l1cfg = netif_running(dev);
333 int dbfifo_int_thresh = 10; /* 10 == 640 entry threshold */
334 module_param(dbfifo_int_thresh, int, 0644);
335 MODULE_PARM_DESC(dbfifo_int_thresh, "doorbell fifo interrupt threshold");
338 * usecs to sleep while draining the dbfifo
340 static int dbfifo_drain_delay = 1000;
341 module_param(dbfifo_drain_delay, int, 0644);
342 MODULE_PARM_DESC(dbfifo_drain_delay,
343 "usecs to sleep while draining the dbfifo");
345 static inline int cxgb4_set_addr_hash(struct port_info *pi)
347 struct adapter *adap = pi->adapter;
350 struct hash_mac_addr *entry;
352 /* Calculate the hash vector for the updated list and program it */
353 list_for_each_entry(entry, &adap->mac_hlist, list) {
354 ucast |= is_unicast_ether_addr(entry->addr);
355 vec |= (1ULL << hash_mac_addr(entry->addr));
357 return t4_set_addr_hash(adap, adap->mbox, pi->viid, ucast,
361 static int cxgb4_mac_sync(struct net_device *netdev, const u8 *mac_addr)
363 struct port_info *pi = netdev_priv(netdev);
364 struct adapter *adap = pi->adapter;
368 /* idx stores the index of allocated filters,
369 * its size should be modified based on the number of
370 * MAC addresses that we allocate filters for
375 bool ucast = is_unicast_ether_addr(mac_addr);
376 const u8 *maclist[1] = {mac_addr};
377 struct hash_mac_addr *new_entry;
379 ret = cxgb4_alloc_mac_filt(adap, pi->viid, free, 1, maclist,
380 idx, ucast ? &uhash : &mhash, false);
383 /* if hash != 0, then add the addr to hash addr list
384 * so on the end we will calculate the hash for the
385 * list and program it
387 if (uhash || mhash) {
388 new_entry = kzalloc(sizeof(*new_entry), GFP_ATOMIC);
391 ether_addr_copy(new_entry->addr, mac_addr);
392 list_add_tail(&new_entry->list, &adap->mac_hlist);
393 ret = cxgb4_set_addr_hash(pi);
396 return ret < 0 ? ret : 0;
399 static int cxgb4_mac_unsync(struct net_device *netdev, const u8 *mac_addr)
401 struct port_info *pi = netdev_priv(netdev);
402 struct adapter *adap = pi->adapter;
404 const u8 *maclist[1] = {mac_addr};
405 struct hash_mac_addr *entry, *tmp;
407 /* If the MAC address to be removed is in the hash addr
408 * list, delete it from the list and update hash vector
410 list_for_each_entry_safe(entry, tmp, &adap->mac_hlist, list) {
411 if (ether_addr_equal(entry->addr, mac_addr)) {
412 list_del(&entry->list);
414 return cxgb4_set_addr_hash(pi);
418 ret = cxgb4_free_mac_filt(adap, pi->viid, 1, maclist, false);
419 return ret < 0 ? -EINVAL : 0;
423 * Set Rx properties of a port, such as promiscruity, address filters, and MTU.
424 * If @mtu is -1 it is left unchanged.
426 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
428 struct port_info *pi = netdev_priv(dev);
429 struct adapter *adapter = pi->adapter;
431 __dev_uc_sync(dev, cxgb4_mac_sync, cxgb4_mac_unsync);
432 __dev_mc_sync(dev, cxgb4_mac_sync, cxgb4_mac_unsync);
434 return t4_set_rxmode(adapter, adapter->mbox, pi->viid, mtu,
435 (dev->flags & IFF_PROMISC) ? 1 : 0,
436 (dev->flags & IFF_ALLMULTI) ? 1 : 0, 1, -1,
441 * cxgb4_change_mac - Update match filter for a MAC address.
444 * @tcam_idx: TCAM index of existing filter for old value of MAC address,
446 * @addr: the new MAC address value
447 * @persist: whether a new MAC allocation should be persistent
448 * @add_smt: if true also add the address to the HW SMT
450 * Modifies an MPS filter and sets it to the new MAC address if
451 * @tcam_idx >= 0, or adds the MAC address to a new filter if
452 * @tcam_idx < 0. In the latter case the address is added persistently
453 * if @persist is %true.
454 * Addresses are programmed to hash region, if tcam runs out of entries.
457 int cxgb4_change_mac(struct port_info *pi, unsigned int viid,
458 int *tcam_idx, const u8 *addr, bool persist,
461 struct adapter *adapter = pi->adapter;
462 struct hash_mac_addr *entry, *new_entry;
465 ret = t4_change_mac(adapter, adapter->mbox, viid,
466 *tcam_idx, addr, persist, smt_idx);
467 /* We ran out of TCAM entries. try programming hash region. */
468 if (ret == -ENOMEM) {
469 /* If the MAC address to be updated is in the hash addr
470 * list, update it from the list
472 list_for_each_entry(entry, &adapter->mac_hlist, list) {
473 if (entry->iface_mac) {
474 ether_addr_copy(entry->addr, addr);
478 new_entry = kzalloc(sizeof(*new_entry), GFP_KERNEL);
481 ether_addr_copy(new_entry->addr, addr);
482 new_entry->iface_mac = true;
483 list_add_tail(&new_entry->list, &adapter->mac_hlist);
485 ret = cxgb4_set_addr_hash(pi);
486 } else if (ret >= 0) {
495 * link_start - enable a port
496 * @dev: the port to enable
498 * Performs the MAC and PHY actions needed to enable a port.
500 static int link_start(struct net_device *dev)
503 struct port_info *pi = netdev_priv(dev);
504 unsigned int mb = pi->adapter->pf;
507 * We do not set address filters and promiscuity here, the stack does
508 * that step explicitly.
510 ret = t4_set_rxmode(pi->adapter, mb, pi->viid, dev->mtu, -1, -1, -1,
511 !!(dev->features & NETIF_F_HW_VLAN_CTAG_RX), true);
513 ret = cxgb4_update_mac_filt(pi, pi->viid, &pi->xact_addr_filt,
514 dev->dev_addr, true, &pi->smt_idx);
516 ret = t4_link_l1cfg(pi->adapter, mb, pi->tx_chan,
520 ret = t4_enable_pi_params(pi->adapter, mb, pi, true,
521 true, CXGB4_DCB_ENABLED);
528 #ifdef CONFIG_CHELSIO_T4_DCB
529 /* Handle a Data Center Bridging update message from the firmware. */
530 static void dcb_rpl(struct adapter *adap, const struct fw_port_cmd *pcmd)
532 int port = FW_PORT_CMD_PORTID_G(ntohl(pcmd->op_to_portid));
533 struct net_device *dev = adap->port[adap->chan_map[port]];
534 int old_dcb_enabled = cxgb4_dcb_enabled(dev);
537 cxgb4_dcb_handle_fw_update(adap, pcmd);
538 new_dcb_enabled = cxgb4_dcb_enabled(dev);
540 /* If the DCB has become enabled or disabled on the port then we're
541 * going to need to set up/tear down DCB Priority parameters for the
542 * TX Queues associated with the port.
544 if (new_dcb_enabled != old_dcb_enabled)
545 dcb_tx_queue_prio_enable(dev, new_dcb_enabled);
547 #endif /* CONFIG_CHELSIO_T4_DCB */
549 /* Response queue handler for the FW event queue.
551 static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
552 const struct pkt_gl *gl)
554 u8 opcode = ((const struct rss_header *)rsp)->opcode;
556 rsp++; /* skip RSS header */
558 /* FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG.
560 if (unlikely(opcode == CPL_FW4_MSG &&
561 ((const struct cpl_fw4_msg *)rsp)->type == FW_TYPE_RSSCPL)) {
563 opcode = ((const struct rss_header *)rsp)->opcode;
565 if (opcode != CPL_SGE_EGR_UPDATE) {
566 dev_err(q->adap->pdev_dev, "unexpected FW4/CPL %#x on FW event queue\n"
572 if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
573 const struct cpl_sge_egr_update *p = (void *)rsp;
574 unsigned int qid = EGR_QID_G(ntohl(p->opcode_qid));
577 txq = q->adap->sge.egr_map[qid - q->adap->sge.egr_start];
579 if (txq->q_type == CXGB4_TXQ_ETH) {
580 struct sge_eth_txq *eq;
582 eq = container_of(txq, struct sge_eth_txq, q);
583 t4_sge_eth_txq_egress_update(q->adap, eq, -1);
585 struct sge_uld_txq *oq;
587 oq = container_of(txq, struct sge_uld_txq, q);
588 tasklet_schedule(&oq->qresume_tsk);
590 } else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
591 const struct cpl_fw6_msg *p = (void *)rsp;
593 #ifdef CONFIG_CHELSIO_T4_DCB
594 const struct fw_port_cmd *pcmd = (const void *)p->data;
595 unsigned int cmd = FW_CMD_OP_G(ntohl(pcmd->op_to_portid));
596 unsigned int action =
597 FW_PORT_CMD_ACTION_G(ntohl(pcmd->action_to_len16));
599 if (cmd == FW_PORT_CMD &&
600 (action == FW_PORT_ACTION_GET_PORT_INFO ||
601 action == FW_PORT_ACTION_GET_PORT_INFO32)) {
602 int port = FW_PORT_CMD_PORTID_G(
603 be32_to_cpu(pcmd->op_to_portid));
604 struct net_device *dev;
605 int dcbxdis, state_input;
607 dev = q->adap->port[q->adap->chan_map[port]];
608 dcbxdis = (action == FW_PORT_ACTION_GET_PORT_INFO
609 ? !!(pcmd->u.info.dcbxdis_pkd & FW_PORT_CMD_DCBXDIS_F)
610 : !!(be32_to_cpu(pcmd->u.info32.lstatus32_to_cbllen32)
611 & FW_PORT_CMD_DCBXDIS32_F));
612 state_input = (dcbxdis
613 ? CXGB4_DCB_INPUT_FW_DISABLED
614 : CXGB4_DCB_INPUT_FW_ENABLED);
616 cxgb4_dcb_state_fsm(dev, state_input);
619 if (cmd == FW_PORT_CMD &&
620 action == FW_PORT_ACTION_L2_DCB_CFG)
621 dcb_rpl(q->adap, pcmd);
625 t4_handle_fw_rpl(q->adap, p->data);
626 } else if (opcode == CPL_L2T_WRITE_RPL) {
627 const struct cpl_l2t_write_rpl *p = (void *)rsp;
629 do_l2t_write_rpl(q->adap, p);
630 } else if (opcode == CPL_SMT_WRITE_RPL) {
631 const struct cpl_smt_write_rpl *p = (void *)rsp;
633 do_smt_write_rpl(q->adap, p);
634 } else if (opcode == CPL_SET_TCB_RPL) {
635 const struct cpl_set_tcb_rpl *p = (void *)rsp;
637 filter_rpl(q->adap, p);
638 } else if (opcode == CPL_ACT_OPEN_RPL) {
639 const struct cpl_act_open_rpl *p = (void *)rsp;
641 hash_filter_rpl(q->adap, p);
642 } else if (opcode == CPL_ABORT_RPL_RSS) {
643 const struct cpl_abort_rpl_rss *p = (void *)rsp;
645 hash_del_filter_rpl(q->adap, p);
646 } else if (opcode == CPL_SRQ_TABLE_RPL) {
647 const struct cpl_srq_table_rpl *p = (void *)rsp;
649 do_srq_table_rpl(q->adap, p);
651 dev_err(q->adap->pdev_dev,
652 "unexpected CPL %#x on FW event queue\n", opcode);
657 static void disable_msi(struct adapter *adapter)
659 if (adapter->flags & CXGB4_USING_MSIX) {
660 pci_disable_msix(adapter->pdev);
661 adapter->flags &= ~CXGB4_USING_MSIX;
662 } else if (adapter->flags & CXGB4_USING_MSI) {
663 pci_disable_msi(adapter->pdev);
664 adapter->flags &= ~CXGB4_USING_MSI;
669 * Interrupt handler for non-data events used with MSI-X.
671 static irqreturn_t t4_nondata_intr(int irq, void *cookie)
673 struct adapter *adap = cookie;
674 u32 v = t4_read_reg(adap, MYPF_REG(PL_PF_INT_CAUSE_A));
678 t4_write_reg(adap, MYPF_REG(PL_PF_INT_CAUSE_A), v);
680 if (adap->flags & CXGB4_MASTER_PF)
681 t4_slow_intr_handler(adap);
685 int cxgb4_set_msix_aff(struct adapter *adap, unsigned short vec,
686 cpumask_var_t *aff_mask, int idx)
690 if (!zalloc_cpumask_var(aff_mask, GFP_KERNEL)) {
691 dev_err(adap->pdev_dev, "alloc_cpumask_var failed\n");
695 cpumask_set_cpu(cpumask_local_spread(idx, dev_to_node(adap->pdev_dev)),
698 rv = irq_set_affinity_hint(vec, *aff_mask);
700 dev_warn(adap->pdev_dev,
701 "irq_set_affinity_hint %u failed %d\n",
707 void cxgb4_clear_msix_aff(unsigned short vec, cpumask_var_t aff_mask)
709 irq_set_affinity_hint(vec, NULL);
710 free_cpumask_var(aff_mask);
713 static int request_msix_queue_irqs(struct adapter *adap)
715 struct sge *s = &adap->sge;
716 struct msix_info *minfo;
719 if (s->fwevtq_msix_idx < 0)
722 err = request_irq(adap->msix_info[s->fwevtq_msix_idx].vec,
724 adap->msix_info[s->fwevtq_msix_idx].desc,
729 for_each_ethrxq(s, ethqidx) {
730 minfo = s->ethrxq[ethqidx].msix;
731 err = request_irq(minfo->vec,
734 &s->ethrxq[ethqidx].rspq);
738 cxgb4_set_msix_aff(adap, minfo->vec,
739 &minfo->aff_mask, ethqidx);
744 while (--ethqidx >= 0) {
745 minfo = s->ethrxq[ethqidx].msix;
746 cxgb4_clear_msix_aff(minfo->vec, minfo->aff_mask);
747 free_irq(minfo->vec, &s->ethrxq[ethqidx].rspq);
749 free_irq(adap->msix_info[s->fwevtq_msix_idx].vec, &s->fw_evtq);
753 static void free_msix_queue_irqs(struct adapter *adap)
755 struct sge *s = &adap->sge;
756 struct msix_info *minfo;
759 free_irq(adap->msix_info[s->fwevtq_msix_idx].vec, &s->fw_evtq);
760 for_each_ethrxq(s, i) {
761 minfo = s->ethrxq[i].msix;
762 cxgb4_clear_msix_aff(minfo->vec, minfo->aff_mask);
763 free_irq(minfo->vec, &s->ethrxq[i].rspq);
767 static int setup_ppod_edram(struct adapter *adap)
769 unsigned int param, val;
772 /* Driver sends FW_PARAMS_PARAM_DEV_PPOD_EDRAM read command to check
773 * if firmware supports ppod edram feature or not. If firmware
774 * returns 1, then driver can enable this feature by sending
775 * FW_PARAMS_PARAM_DEV_PPOD_EDRAM write command with value 1 to
776 * enable ppod edram feature.
778 param = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
779 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_PPOD_EDRAM));
781 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, ¶m, &val);
783 dev_warn(adap->pdev_dev,
784 "querying PPOD_EDRAM support failed: %d\n",
792 ret = t4_set_params(adap, adap->mbox, adap->pf, 0, 1, ¶m, &val);
794 dev_err(adap->pdev_dev,
795 "setting PPOD_EDRAM failed: %d\n", ret);
801 static void adap_config_hpfilter(struct adapter *adapter)
806 /* Enable HP filter region. Older fw will fail this request and
809 param = FW_PARAM_DEV(HPFILTER_REGION_SUPPORT);
810 ret = t4_set_params(adapter, adapter->mbox, adapter->pf, 0,
813 /* An error means FW doesn't know about HP filter support,
814 * it's not a problem, don't return an error.
817 dev_err(adapter->pdev_dev,
818 "HP filter region isn't supported by FW\n");
822 * cxgb4_write_rss - write the RSS table for a given port
824 * @queues: array of queue indices for RSS
826 * Sets up the portion of the HW RSS table for the port's VI to distribute
827 * packets to the Rx queues in @queues.
828 * Should never be called before setting up sge eth rx queues
830 int cxgb4_write_rss(const struct port_info *pi, const u16 *queues)
834 struct adapter *adapter = pi->adapter;
835 const struct sge_eth_rxq *rxq;
837 rxq = &adapter->sge.ethrxq[pi->first_qset];
838 rss = kmalloc_array(pi->rss_size, sizeof(u16), GFP_KERNEL);
842 /* map the queue indices to queue ids */
843 for (i = 0; i < pi->rss_size; i++, queues++)
844 rss[i] = rxq[*queues].rspq.abs_id;
846 err = t4_config_rss_range(adapter, adapter->pf, pi->viid, 0,
847 pi->rss_size, rss, pi->rss_size);
848 /* If Tunnel All Lookup isn't specified in the global RSS
849 * Configuration, then we need to specify a default Ingress
850 * Queue for any ingress packets which aren't hashed. We'll
851 * use our first ingress queue ...
854 err = t4_config_vi_rss(adapter, adapter->mbox, pi->viid,
855 FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN_F |
856 FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN_F |
857 FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN_F |
858 FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN_F |
859 FW_RSS_VI_CONFIG_CMD_UDPEN_F,
866 * setup_rss - configure RSS
869 * Sets up RSS for each port.
871 static int setup_rss(struct adapter *adap)
875 for_each_port(adap, i) {
876 const struct port_info *pi = adap2pinfo(adap, i);
878 /* Fill default values with equal distribution */
879 for (j = 0; j < pi->rss_size; j++)
880 pi->rss[j] = j % pi->nqsets;
882 err = cxgb4_write_rss(pi, pi->rss);
890 * Return the channel of the ingress queue with the given qid.
892 static unsigned int rxq_to_chan(const struct sge *p, unsigned int qid)
894 qid -= p->ingr_start;
895 return netdev2pinfo(p->ingr_map[qid]->netdev)->tx_chan;
898 void cxgb4_quiesce_rx(struct sge_rspq *q)
901 napi_disable(&q->napi);
905 * Wait until all NAPI handlers are descheduled.
907 static void quiesce_rx(struct adapter *adap)
911 for (i = 0; i < adap->sge.ingr_sz; i++) {
912 struct sge_rspq *q = adap->sge.ingr_map[i];
921 /* Disable interrupt and napi handler */
922 static void disable_interrupts(struct adapter *adap)
924 struct sge *s = &adap->sge;
926 if (adap->flags & CXGB4_FULL_INIT_DONE) {
927 t4_intr_disable(adap);
928 if (adap->flags & CXGB4_USING_MSIX) {
929 free_msix_queue_irqs(adap);
930 free_irq(adap->msix_info[s->nd_msix_idx].vec,
933 free_irq(adap->pdev->irq, adap);
939 void cxgb4_enable_rx(struct adapter *adap, struct sge_rspq *q)
942 napi_enable(&q->napi);
944 /* 0-increment GTS to start the timer and enable interrupts */
945 t4_write_reg(adap, MYPF_REG(SGE_PF_GTS_A),
946 SEINTARM_V(q->intr_params) |
947 INGRESSQID_V(q->cntxt_id));
951 * Enable NAPI scheduling and interrupt generation for all Rx queues.
953 static void enable_rx(struct adapter *adap)
957 for (i = 0; i < adap->sge.ingr_sz; i++) {
958 struct sge_rspq *q = adap->sge.ingr_map[i];
963 cxgb4_enable_rx(adap, q);
967 static int setup_non_data_intr(struct adapter *adap)
971 adap->sge.nd_msix_idx = -1;
972 if (!(adap->flags & CXGB4_USING_MSIX))
975 /* Request MSI-X vector for non-data interrupt */
976 msix = cxgb4_get_msix_idx_from_bmap(adap);
980 snprintf(adap->msix_info[msix].desc,
981 sizeof(adap->msix_info[msix].desc),
982 "%s", adap->port[0]->name);
984 adap->sge.nd_msix_idx = msix;
988 static int setup_fw_sge_queues(struct adapter *adap)
990 struct sge *s = &adap->sge;
993 bitmap_zero(s->starving_fl, s->egr_sz);
994 bitmap_zero(s->txq_maperr, s->egr_sz);
996 if (adap->flags & CXGB4_USING_MSIX) {
997 s->fwevtq_msix_idx = -1;
998 msix = cxgb4_get_msix_idx_from_bmap(adap);
1002 snprintf(adap->msix_info[msix].desc,
1003 sizeof(adap->msix_info[msix].desc),
1004 "%s-FWeventq", adap->port[0]->name);
1006 err = t4_sge_alloc_rxq(adap, &s->intrq, false, adap->port[0], 0,
1007 NULL, NULL, NULL, -1);
1010 msix = -((int)s->intrq.abs_id + 1);
1013 err = t4_sge_alloc_rxq(adap, &s->fw_evtq, true, adap->port[0],
1014 msix, NULL, fwevtq_handler, NULL, -1);
1015 if (err && msix >= 0)
1016 cxgb4_free_msix_idx_in_bmap(adap, msix);
1018 s->fwevtq_msix_idx = msix;
1023 * setup_sge_queues - configure SGE Tx/Rx/response queues
1024 * @adap: the adapter
1026 * Determines how many sets of SGE queues to use and initializes them.
1027 * We support multiple queue sets per port if we have MSI-X, otherwise
1028 * just one queue set per port.
1030 static int setup_sge_queues(struct adapter *adap)
1032 struct sge_uld_rxq_info *rxq_info = NULL;
1033 struct sge *s = &adap->sge;
1034 unsigned int cmplqid = 0;
1035 int err, i, j, msix = 0;
1038 rxq_info = s->uld_rxq_info[CXGB4_ULD_RDMA];
1040 if (!(adap->flags & CXGB4_USING_MSIX))
1041 msix = -((int)s->intrq.abs_id + 1);
1043 for_each_port(adap, i) {
1044 struct net_device *dev = adap->port[i];
1045 struct port_info *pi = netdev_priv(dev);
1046 struct sge_eth_rxq *q = &s->ethrxq[pi->first_qset];
1047 struct sge_eth_txq *t = &s->ethtxq[pi->first_qset];
1049 for (j = 0; j < pi->nqsets; j++, q++) {
1051 msix = cxgb4_get_msix_idx_from_bmap(adap);
1057 snprintf(adap->msix_info[msix].desc,
1058 sizeof(adap->msix_info[msix].desc),
1059 "%s-Rx%d", dev->name, j);
1060 q->msix = &adap->msix_info[msix];
1063 err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev,
1067 t4_get_tp_ch_map(adap,
1072 memset(&q->stats, 0, sizeof(q->stats));
1075 q = &s->ethrxq[pi->first_qset];
1076 for (j = 0; j < pi->nqsets; j++, t++, q++) {
1077 err = t4_sge_alloc_eth_txq(adap, t, dev,
1078 netdev_get_tx_queue(dev, j),
1080 !!(adap->flags & CXGB4_SGE_DBQ_TIMER));
1086 for_each_port(adap, i) {
1087 /* Note that cmplqid below is 0 if we don't
1088 * have RDMA queues, and that's the right value.
1091 cmplqid = rxq_info->uldrxq[i].rspq.cntxt_id;
1093 err = t4_sge_alloc_ctrl_txq(adap, &s->ctrlq[i], adap->port[i],
1094 s->fw_evtq.cntxt_id, cmplqid);
1099 if (!is_t4(adap->params.chip)) {
1100 err = t4_sge_alloc_eth_txq(adap, &s->ptptxq, adap->port[0],
1101 netdev_get_tx_queue(adap->port[0], 0)
1102 , s->fw_evtq.cntxt_id, false);
1107 t4_write_reg(adap, is_t4(adap->params.chip) ?
1108 MPS_TRC_RSS_CONTROL_A :
1109 MPS_T5_TRC_RSS_CONTROL_A,
1110 RSSCONTROL_V(netdev2pinfo(adap->port[0])->tx_chan) |
1111 QUEUENUMBER_V(s->ethrxq[0].rspq.abs_id));
1114 dev_err(adap->pdev_dev, "Can't allocate queues, err=%d\n", -err);
1115 t4_free_sge_resources(adap);
1119 static u16 cxgb_select_queue(struct net_device *dev, struct sk_buff *skb,
1120 struct net_device *sb_dev)
1124 #ifdef CONFIG_CHELSIO_T4_DCB
1125 /* If a Data Center Bridging has been successfully negotiated on this
1126 * link then we'll use the skb's priority to map it to a TX Queue.
1127 * The skb's priority is determined via the VLAN Tag Priority Code
1130 if (cxgb4_dcb_enabled(dev) && !is_kdump_kernel()) {
1134 err = vlan_get_tag(skb, &vlan_tci);
1135 if (unlikely(err)) {
1136 if (net_ratelimit())
1138 "TX Packet without VLAN Tag on DCB Link\n");
1141 txq = (vlan_tci & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
1142 #ifdef CONFIG_CHELSIO_T4_FCOE
1143 if (skb->protocol == htons(ETH_P_FCOE))
1144 txq = skb->priority & 0x7;
1145 #endif /* CONFIG_CHELSIO_T4_FCOE */
1149 #endif /* CONFIG_CHELSIO_T4_DCB */
1152 struct port_info *pi = netdev2pinfo(dev);
1155 ver = ip_hdr(skb)->version;
1156 proto = (ver == 6) ? ipv6_hdr(skb)->nexthdr :
1157 ip_hdr(skb)->protocol;
1159 /* Send unsupported traffic pattern to normal NIC queues. */
1160 txq = netdev_pick_tx(dev, skb, sb_dev);
1161 if (xfrm_offload(skb) || is_ptp_enabled(skb, dev) ||
1162 skb->encapsulation ||
1163 (proto != IPPROTO_TCP && proto != IPPROTO_UDP))
1164 txq = txq % pi->nqsets;
1170 txq = (skb_rx_queue_recorded(skb)
1171 ? skb_get_rx_queue(skb)
1172 : smp_processor_id());
1174 while (unlikely(txq >= dev->real_num_tx_queues))
1175 txq -= dev->real_num_tx_queues;
1180 return netdev_pick_tx(dev, skb, NULL) % dev->real_num_tx_queues;
1183 static int closest_timer(const struct sge *s, int time)
1185 int i, delta, match = 0, min_delta = INT_MAX;
1187 for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
1188 delta = time - s->timer_val[i];
1191 if (delta < min_delta) {
1199 static int closest_thres(const struct sge *s, int thres)
1201 int i, delta, match = 0, min_delta = INT_MAX;
1203 for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
1204 delta = thres - s->counter_val[i];
1207 if (delta < min_delta) {
1216 * cxgb4_set_rspq_intr_params - set a queue's interrupt holdoff parameters
1218 * @us: the hold-off time in us, or 0 to disable timer
1219 * @cnt: the hold-off packet count, or 0 to disable counter
1221 * Sets an Rx queue's interrupt hold-off time and packet count. At least
1222 * one of the two needs to be enabled for the queue to generate interrupts.
1224 int cxgb4_set_rspq_intr_params(struct sge_rspq *q,
1225 unsigned int us, unsigned int cnt)
1227 struct adapter *adap = q->adap;
1229 if ((us | cnt) == 0)
1236 new_idx = closest_thres(&adap->sge, cnt);
1237 if (q->desc && q->pktcnt_idx != new_idx) {
1238 /* the queue has already been created, update it */
1239 v = FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DMAQ) |
1240 FW_PARAMS_PARAM_X_V(
1241 FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
1242 FW_PARAMS_PARAM_YZ_V(q->cntxt_id);
1243 err = t4_set_params(adap, adap->mbox, adap->pf, 0, 1,
1248 q->pktcnt_idx = new_idx;
1251 us = us == 0 ? 6 : closest_timer(&adap->sge, us);
1252 q->intr_params = QINTR_TIMER_IDX_V(us) | QINTR_CNT_EN_V(cnt > 0);
1256 static int cxgb_set_features(struct net_device *dev, netdev_features_t features)
1258 const struct port_info *pi = netdev_priv(dev);
1259 netdev_features_t changed = dev->features ^ features;
1262 if (!(changed & NETIF_F_HW_VLAN_CTAG_RX))
1265 err = t4_set_rxmode(pi->adapter, pi->adapter->pf, pi->viid, -1,
1267 !!(features & NETIF_F_HW_VLAN_CTAG_RX), true);
1269 dev->features = features ^ NETIF_F_HW_VLAN_CTAG_RX;
1273 static int setup_debugfs(struct adapter *adap)
1275 if (IS_ERR_OR_NULL(adap->debugfs_root))
1278 #ifdef CONFIG_DEBUG_FS
1279 t4_setup_debugfs(adap);
1285 * upper-layer driver support
1289 * Allocate an active-open TID and set it to the supplied value.
1291 int cxgb4_alloc_atid(struct tid_info *t, void *data)
1295 spin_lock_bh(&t->atid_lock);
1297 union aopen_entry *p = t->afree;
1299 atid = (p - t->atid_tab) + t->atid_base;
1304 spin_unlock_bh(&t->atid_lock);
1307 EXPORT_SYMBOL(cxgb4_alloc_atid);
1310 * Release an active-open TID.
1312 void cxgb4_free_atid(struct tid_info *t, unsigned int atid)
1314 union aopen_entry *p = &t->atid_tab[atid - t->atid_base];
1316 spin_lock_bh(&t->atid_lock);
1320 spin_unlock_bh(&t->atid_lock);
1322 EXPORT_SYMBOL(cxgb4_free_atid);
1325 * Allocate a server TID and set it to the supplied value.
1327 int cxgb4_alloc_stid(struct tid_info *t, int family, void *data)
1331 spin_lock_bh(&t->stid_lock);
1332 if (family == PF_INET) {
1333 stid = find_first_zero_bit(t->stid_bmap, t->nstids);
1334 if (stid < t->nstids)
1335 __set_bit(stid, t->stid_bmap);
1339 stid = bitmap_find_free_region(t->stid_bmap, t->nstids, 1);
1344 t->stid_tab[stid].data = data;
1345 stid += t->stid_base;
1346 /* IPv6 requires max of 520 bits or 16 cells in TCAM
1347 * This is equivalent to 4 TIDs. With CLIP enabled it
1350 if (family == PF_INET6) {
1351 t->stids_in_use += 2;
1352 t->v6_stids_in_use += 2;
1357 spin_unlock_bh(&t->stid_lock);
1360 EXPORT_SYMBOL(cxgb4_alloc_stid);
1362 /* Allocate a server filter TID and set it to the supplied value.
1364 int cxgb4_alloc_sftid(struct tid_info *t, int family, void *data)
1368 spin_lock_bh(&t->stid_lock);
1369 if (family == PF_INET) {
1370 stid = find_next_zero_bit(t->stid_bmap,
1371 t->nstids + t->nsftids, t->nstids);
1372 if (stid < (t->nstids + t->nsftids))
1373 __set_bit(stid, t->stid_bmap);
1380 t->stid_tab[stid].data = data;
1382 stid += t->sftid_base;
1385 spin_unlock_bh(&t->stid_lock);
1388 EXPORT_SYMBOL(cxgb4_alloc_sftid);
1390 /* Release a server TID.
1392 void cxgb4_free_stid(struct tid_info *t, unsigned int stid, int family)
1394 /* Is it a server filter TID? */
1395 if (t->nsftids && (stid >= t->sftid_base)) {
1396 stid -= t->sftid_base;
1399 stid -= t->stid_base;
1402 spin_lock_bh(&t->stid_lock);
1403 if (family == PF_INET)
1404 __clear_bit(stid, t->stid_bmap);
1406 bitmap_release_region(t->stid_bmap, stid, 1);
1407 t->stid_tab[stid].data = NULL;
1408 if (stid < t->nstids) {
1409 if (family == PF_INET6) {
1410 t->stids_in_use -= 2;
1411 t->v6_stids_in_use -= 2;
1419 spin_unlock_bh(&t->stid_lock);
1421 EXPORT_SYMBOL(cxgb4_free_stid);
1424 * Populate a TID_RELEASE WR. Caller must properly size the skb.
1426 static void mk_tid_release(struct sk_buff *skb, unsigned int chan,
1429 struct cpl_tid_release *req;
1431 set_wr_txq(skb, CPL_PRIORITY_SETUP, chan);
1432 req = __skb_put(skb, sizeof(*req));
1433 INIT_TP_WR(req, tid);
1434 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, tid));
1438 * Queue a TID release request and if necessary schedule a work queue to
1441 static void cxgb4_queue_tid_release(struct tid_info *t, unsigned int chan,
1444 struct adapter *adap = container_of(t, struct adapter, tids);
1445 void **p = &t->tid_tab[tid - t->tid_base];
1447 spin_lock_bh(&adap->tid_release_lock);
1448 *p = adap->tid_release_head;
1449 /* Low 2 bits encode the Tx channel number */
1450 adap->tid_release_head = (void **)((uintptr_t)p | chan);
1451 if (!adap->tid_release_task_busy) {
1452 adap->tid_release_task_busy = true;
1453 queue_work(adap->workq, &adap->tid_release_task);
1455 spin_unlock_bh(&adap->tid_release_lock);
1459 * Process the list of pending TID release requests.
1461 static void process_tid_release_list(struct work_struct *work)
1463 struct sk_buff *skb;
1464 struct adapter *adap;
1466 adap = container_of(work, struct adapter, tid_release_task);
1468 spin_lock_bh(&adap->tid_release_lock);
1469 while (adap->tid_release_head) {
1470 void **p = adap->tid_release_head;
1471 unsigned int chan = (uintptr_t)p & 3;
1472 p = (void *)p - chan;
1474 adap->tid_release_head = *p;
1476 spin_unlock_bh(&adap->tid_release_lock);
1478 while (!(skb = alloc_skb(sizeof(struct cpl_tid_release),
1480 schedule_timeout_uninterruptible(1);
1482 mk_tid_release(skb, chan, p - adap->tids.tid_tab);
1483 t4_ofld_send(adap, skb);
1484 spin_lock_bh(&adap->tid_release_lock);
1486 adap->tid_release_task_busy = false;
1487 spin_unlock_bh(&adap->tid_release_lock);
1491 * Release a TID and inform HW. If we are unable to allocate the release
1492 * message we defer to a work queue.
1494 void cxgb4_remove_tid(struct tid_info *t, unsigned int chan, unsigned int tid,
1495 unsigned short family)
1497 struct adapter *adap = container_of(t, struct adapter, tids);
1498 struct sk_buff *skb;
1500 WARN_ON(tid_out_of_range(&adap->tids, tid));
1502 if (t->tid_tab[tid - adap->tids.tid_base]) {
1503 t->tid_tab[tid - adap->tids.tid_base] = NULL;
1504 atomic_dec(&t->conns_in_use);
1505 if (t->hash_base && (tid >= t->hash_base)) {
1506 if (family == AF_INET6)
1507 atomic_sub(2, &t->hash_tids_in_use);
1509 atomic_dec(&t->hash_tids_in_use);
1511 if (family == AF_INET6)
1512 atomic_sub(2, &t->tids_in_use);
1514 atomic_dec(&t->tids_in_use);
1518 skb = alloc_skb(sizeof(struct cpl_tid_release), GFP_ATOMIC);
1520 mk_tid_release(skb, chan, tid);
1521 t4_ofld_send(adap, skb);
1523 cxgb4_queue_tid_release(t, chan, tid);
1525 EXPORT_SYMBOL(cxgb4_remove_tid);
1528 * Allocate and initialize the TID tables. Returns 0 on success.
1530 static int tid_init(struct tid_info *t)
1532 struct adapter *adap = container_of(t, struct adapter, tids);
1533 unsigned int max_ftids = t->nftids + t->nsftids;
1534 unsigned int natids = t->natids;
1535 unsigned int hpftid_bmap_size;
1536 unsigned int eotid_bmap_size;
1537 unsigned int stid_bmap_size;
1538 unsigned int ftid_bmap_size;
1541 stid_bmap_size = BITS_TO_LONGS(t->nstids + t->nsftids);
1542 ftid_bmap_size = BITS_TO_LONGS(t->nftids);
1543 hpftid_bmap_size = BITS_TO_LONGS(t->nhpftids);
1544 eotid_bmap_size = BITS_TO_LONGS(t->neotids);
1545 size = t->ntids * sizeof(*t->tid_tab) +
1546 natids * sizeof(*t->atid_tab) +
1547 t->nstids * sizeof(*t->stid_tab) +
1548 t->nsftids * sizeof(*t->stid_tab) +
1549 stid_bmap_size * sizeof(long) +
1550 t->nhpftids * sizeof(*t->hpftid_tab) +
1551 hpftid_bmap_size * sizeof(long) +
1552 max_ftids * sizeof(*t->ftid_tab) +
1553 ftid_bmap_size * sizeof(long) +
1554 t->neotids * sizeof(*t->eotid_tab) +
1555 eotid_bmap_size * sizeof(long);
1557 t->tid_tab = kvzalloc(size, GFP_KERNEL);
1561 t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
1562 t->stid_tab = (struct serv_entry *)&t->atid_tab[natids];
1563 t->stid_bmap = (unsigned long *)&t->stid_tab[t->nstids + t->nsftids];
1564 t->hpftid_tab = (struct filter_entry *)&t->stid_bmap[stid_bmap_size];
1565 t->hpftid_bmap = (unsigned long *)&t->hpftid_tab[t->nhpftids];
1566 t->ftid_tab = (struct filter_entry *)&t->hpftid_bmap[hpftid_bmap_size];
1567 t->ftid_bmap = (unsigned long *)&t->ftid_tab[max_ftids];
1568 t->eotid_tab = (struct eotid_entry *)&t->ftid_bmap[ftid_bmap_size];
1569 t->eotid_bmap = (unsigned long *)&t->eotid_tab[t->neotids];
1570 spin_lock_init(&t->stid_lock);
1571 spin_lock_init(&t->atid_lock);
1572 spin_lock_init(&t->ftid_lock);
1574 t->stids_in_use = 0;
1575 t->v6_stids_in_use = 0;
1576 t->sftids_in_use = 0;
1578 t->atids_in_use = 0;
1579 atomic_set(&t->tids_in_use, 0);
1580 atomic_set(&t->conns_in_use, 0);
1581 atomic_set(&t->hash_tids_in_use, 0);
1582 atomic_set(&t->eotids_in_use, 0);
1584 /* Setup the free list for atid_tab and clear the stid bitmap. */
1587 t->atid_tab[natids - 1].next = &t->atid_tab[natids];
1588 t->afree = t->atid_tab;
1591 if (is_offload(adap)) {
1592 bitmap_zero(t->stid_bmap, t->nstids + t->nsftids);
1593 /* Reserve stid 0 for T4/T5 adapters */
1594 if (!t->stid_base &&
1595 CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5)
1596 __set_bit(0, t->stid_bmap);
1599 bitmap_zero(t->eotid_bmap, t->neotids);
1603 bitmap_zero(t->hpftid_bmap, t->nhpftids);
1604 bitmap_zero(t->ftid_bmap, t->nftids);
1609 * cxgb4_create_server - create an IP server
1611 * @stid: the server TID
1612 * @sip: local IP address to bind server to
1613 * @sport: the server's TCP port
1614 * @queue: queue to direct messages from this server to
1616 * Create an IP server for the given port and address.
1617 * Returns <0 on error and one of the %NET_XMIT_* values on success.
1619 int cxgb4_create_server(const struct net_device *dev, unsigned int stid,
1620 __be32 sip, __be16 sport, __be16 vlan,
1624 struct sk_buff *skb;
1625 struct adapter *adap;
1626 struct cpl_pass_open_req *req;
1629 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
1633 adap = netdev2adap(dev);
1634 req = __skb_put(skb, sizeof(*req));
1636 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, stid));
1637 req->local_port = sport;
1638 req->peer_port = htons(0);
1639 req->local_ip = sip;
1640 req->peer_ip = htonl(0);
1641 chan = rxq_to_chan(&adap->sge, queue);
1642 req->opt0 = cpu_to_be64(TX_CHAN_V(chan));
1643 req->opt1 = cpu_to_be64(CONN_POLICY_V(CPL_CONN_POLICY_ASK) |
1644 SYN_RSS_ENABLE_F | SYN_RSS_QUEUE_V(queue));
1645 ret = t4_mgmt_tx(adap, skb);
1646 return net_xmit_eval(ret);
1648 EXPORT_SYMBOL(cxgb4_create_server);
1650 /* cxgb4_create_server6 - create an IPv6 server
1652 * @stid: the server TID
1653 * @sip: local IPv6 address to bind server to
1654 * @sport: the server's TCP port
1655 * @queue: queue to direct messages from this server to
1657 * Create an IPv6 server for the given port and address.
1658 * Returns <0 on error and one of the %NET_XMIT_* values on success.
1660 int cxgb4_create_server6(const struct net_device *dev, unsigned int stid,
1661 const struct in6_addr *sip, __be16 sport,
1665 struct sk_buff *skb;
1666 struct adapter *adap;
1667 struct cpl_pass_open_req6 *req;
1670 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
1674 adap = netdev2adap(dev);
1675 req = __skb_put(skb, sizeof(*req));
1677 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ6, stid));
1678 req->local_port = sport;
1679 req->peer_port = htons(0);
1680 req->local_ip_hi = *(__be64 *)(sip->s6_addr);
1681 req->local_ip_lo = *(__be64 *)(sip->s6_addr + 8);
1682 req->peer_ip_hi = cpu_to_be64(0);
1683 req->peer_ip_lo = cpu_to_be64(0);
1684 chan = rxq_to_chan(&adap->sge, queue);
1685 req->opt0 = cpu_to_be64(TX_CHAN_V(chan));
1686 req->opt1 = cpu_to_be64(CONN_POLICY_V(CPL_CONN_POLICY_ASK) |
1687 SYN_RSS_ENABLE_F | SYN_RSS_QUEUE_V(queue));
1688 ret = t4_mgmt_tx(adap, skb);
1689 return net_xmit_eval(ret);
1691 EXPORT_SYMBOL(cxgb4_create_server6);
1693 int cxgb4_remove_server(const struct net_device *dev, unsigned int stid,
1694 unsigned int queue, bool ipv6)
1696 struct sk_buff *skb;
1697 struct adapter *adap;
1698 struct cpl_close_listsvr_req *req;
1701 adap = netdev2adap(dev);
1703 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
1707 req = __skb_put(skb, sizeof(*req));
1709 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, stid));
1710 req->reply_ctrl = htons(NO_REPLY_V(0) | (ipv6 ? LISTSVR_IPV6_V(1) :
1711 LISTSVR_IPV6_V(0)) | QUEUENO_V(queue));
1712 ret = t4_mgmt_tx(adap, skb);
1713 return net_xmit_eval(ret);
1715 EXPORT_SYMBOL(cxgb4_remove_server);
1718 * cxgb4_best_mtu - find the entry in the MTU table closest to an MTU
1719 * @mtus: the HW MTU table
1720 * @mtu: the target MTU
1721 * @idx: index of selected entry in the MTU table
1723 * Returns the index and the value in the HW MTU table that is closest to
1724 * but does not exceed @mtu, unless @mtu is smaller than any value in the
1725 * table, in which case that smallest available value is selected.
1727 unsigned int cxgb4_best_mtu(const unsigned short *mtus, unsigned short mtu,
1732 while (i < NMTUS - 1 && mtus[i + 1] <= mtu)
1738 EXPORT_SYMBOL(cxgb4_best_mtu);
1741 * cxgb4_best_aligned_mtu - find best MTU, [hopefully] data size aligned
1742 * @mtus: the HW MTU table
1743 * @header_size: Header Size
1744 * @data_size_max: maximum Data Segment Size
1745 * @data_size_align: desired Data Segment Size Alignment (2^N)
1746 * @mtu_idxp: HW MTU Table Index return value pointer (possibly NULL)
1748 * Similar to cxgb4_best_mtu() but instead of searching the Hardware
1749 * MTU Table based solely on a Maximum MTU parameter, we break that
1750 * parameter up into a Header Size and Maximum Data Segment Size, and
1751 * provide a desired Data Segment Size Alignment. If we find an MTU in
1752 * the Hardware MTU Table which will result in a Data Segment Size with
1753 * the requested alignment _and_ that MTU isn't "too far" from the
1754 * closest MTU, then we'll return that rather than the closest MTU.
1756 unsigned int cxgb4_best_aligned_mtu(const unsigned short *mtus,
1757 unsigned short header_size,
1758 unsigned short data_size_max,
1759 unsigned short data_size_align,
1760 unsigned int *mtu_idxp)
1762 unsigned short max_mtu = header_size + data_size_max;
1763 unsigned short data_size_align_mask = data_size_align - 1;
1764 int mtu_idx, aligned_mtu_idx;
1766 /* Scan the MTU Table till we find an MTU which is larger than our
1767 * Maximum MTU or we reach the end of the table. Along the way,
1768 * record the last MTU found, if any, which will result in a Data
1769 * Segment Length matching the requested alignment.
1771 for (mtu_idx = 0, aligned_mtu_idx = -1; mtu_idx < NMTUS; mtu_idx++) {
1772 unsigned short data_size = mtus[mtu_idx] - header_size;
1774 /* If this MTU minus the Header Size would result in a
1775 * Data Segment Size of the desired alignment, remember it.
1777 if ((data_size & data_size_align_mask) == 0)
1778 aligned_mtu_idx = mtu_idx;
1780 /* If we're not at the end of the Hardware MTU Table and the
1781 * next element is larger than our Maximum MTU, drop out of
1784 if (mtu_idx+1 < NMTUS && mtus[mtu_idx+1] > max_mtu)
1788 /* If we fell out of the loop because we ran to the end of the table,
1789 * then we just have to use the last [largest] entry.
1791 if (mtu_idx == NMTUS)
1794 /* If we found an MTU which resulted in the requested Data Segment
1795 * Length alignment and that's "not far" from the largest MTU which is
1796 * less than or equal to the maximum MTU, then use that.
1798 if (aligned_mtu_idx >= 0 &&
1799 mtu_idx - aligned_mtu_idx <= 1)
1800 mtu_idx = aligned_mtu_idx;
1802 /* If the caller has passed in an MTU Index pointer, pass the
1803 * MTU Index back. Return the MTU value.
1806 *mtu_idxp = mtu_idx;
1807 return mtus[mtu_idx];
1809 EXPORT_SYMBOL(cxgb4_best_aligned_mtu);
1812 * cxgb4_port_chan - get the HW channel of a port
1813 * @dev: the net device for the port
1815 * Return the HW Tx channel of the given port.
1817 unsigned int cxgb4_port_chan(const struct net_device *dev)
1819 return netdev2pinfo(dev)->tx_chan;
1821 EXPORT_SYMBOL(cxgb4_port_chan);
1824 * cxgb4_port_e2cchan - get the HW c-channel of a port
1825 * @dev: the net device for the port
1827 * Return the HW RX c-channel of the given port.
1829 unsigned int cxgb4_port_e2cchan(const struct net_device *dev)
1831 return netdev2pinfo(dev)->rx_cchan;
1833 EXPORT_SYMBOL(cxgb4_port_e2cchan);
1835 unsigned int cxgb4_dbfifo_count(const struct net_device *dev, int lpfifo)
1837 struct adapter *adap = netdev2adap(dev);
1838 u32 v1, v2, lp_count, hp_count;
1840 v1 = t4_read_reg(adap, SGE_DBFIFO_STATUS_A);
1841 v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2_A);
1842 if (is_t4(adap->params.chip)) {
1843 lp_count = LP_COUNT_G(v1);
1844 hp_count = HP_COUNT_G(v1);
1846 lp_count = LP_COUNT_T5_G(v1);
1847 hp_count = HP_COUNT_T5_G(v2);
1849 return lpfifo ? lp_count : hp_count;
1851 EXPORT_SYMBOL(cxgb4_dbfifo_count);
1854 * cxgb4_port_viid - get the VI id of a port
1855 * @dev: the net device for the port
1857 * Return the VI id of the given port.
1859 unsigned int cxgb4_port_viid(const struct net_device *dev)
1861 return netdev2pinfo(dev)->viid;
1863 EXPORT_SYMBOL(cxgb4_port_viid);
1866 * cxgb4_port_idx - get the index of a port
1867 * @dev: the net device for the port
1869 * Return the index of the given port.
1871 unsigned int cxgb4_port_idx(const struct net_device *dev)
1873 return netdev2pinfo(dev)->port_id;
1875 EXPORT_SYMBOL(cxgb4_port_idx);
1877 void cxgb4_get_tcp_stats(struct pci_dev *pdev, struct tp_tcp_stats *v4,
1878 struct tp_tcp_stats *v6)
1880 struct adapter *adap = pci_get_drvdata(pdev);
1882 spin_lock(&adap->stats_lock);
1883 t4_tp_get_tcp_stats(adap, v4, v6, false);
1884 spin_unlock(&adap->stats_lock);
1886 EXPORT_SYMBOL(cxgb4_get_tcp_stats);
1888 void cxgb4_iscsi_init(struct net_device *dev, unsigned int tag_mask,
1889 const unsigned int *pgsz_order)
1891 struct adapter *adap = netdev2adap(dev);
1893 t4_write_reg(adap, ULP_RX_ISCSI_TAGMASK_A, tag_mask);
1894 t4_write_reg(adap, ULP_RX_ISCSI_PSZ_A, HPZ0_V(pgsz_order[0]) |
1895 HPZ1_V(pgsz_order[1]) | HPZ2_V(pgsz_order[2]) |
1896 HPZ3_V(pgsz_order[3]));
1898 EXPORT_SYMBOL(cxgb4_iscsi_init);
1900 int cxgb4_flush_eq_cache(struct net_device *dev)
1902 struct adapter *adap = netdev2adap(dev);
1904 return t4_sge_ctxt_flush(adap, adap->mbox, CTXT_EGRESS);
1906 EXPORT_SYMBOL(cxgb4_flush_eq_cache);
1908 static int read_eq_indices(struct adapter *adap, u16 qid, u16 *pidx, u16 *cidx)
1910 u32 addr = t4_read_reg(adap, SGE_DBQ_CTXT_BADDR_A) + 24 * qid + 8;
1914 spin_lock(&adap->win0_lock);
1915 ret = t4_memory_rw(adap, 0, MEM_EDC0, addr,
1916 sizeof(indices), (__be32 *)&indices,
1918 spin_unlock(&adap->win0_lock);
1920 *cidx = (be64_to_cpu(indices) >> 25) & 0xffff;
1921 *pidx = (be64_to_cpu(indices) >> 9) & 0xffff;
1926 int cxgb4_sync_txq_pidx(struct net_device *dev, u16 qid, u16 pidx,
1929 struct adapter *adap = netdev2adap(dev);
1930 u16 hw_pidx, hw_cidx;
1933 ret = read_eq_indices(adap, qid, &hw_pidx, &hw_cidx);
1937 if (pidx != hw_pidx) {
1941 if (pidx >= hw_pidx)
1942 delta = pidx - hw_pidx;
1944 delta = size - hw_pidx + pidx;
1946 if (is_t4(adap->params.chip))
1947 val = PIDX_V(delta);
1949 val = PIDX_T5_V(delta);
1951 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
1957 EXPORT_SYMBOL(cxgb4_sync_txq_pidx);
1959 int cxgb4_read_tpte(struct net_device *dev, u32 stag, __be32 *tpte)
1961 u32 edc0_size, edc1_size, mc0_size, mc1_size, size;
1962 u32 edc0_end, edc1_end, mc0_end, mc1_end;
1963 u32 offset, memtype, memaddr;
1964 struct adapter *adap;
1968 adap = netdev2adap(dev);
1970 offset = ((stag >> 8) * 32) + adap->vres.stag.start;
1972 /* Figure out where the offset lands in the Memory Type/Address scheme.
1973 * This code assumes that the memory is laid out starting at offset 0
1974 * with no breaks as: EDC0, EDC1, MC0, MC1. All cards have both EDC0
1975 * and EDC1. Some cards will have neither MC0 nor MC1, most cards have
1976 * MC0, and some have both MC0 and MC1.
1978 size = t4_read_reg(adap, MA_EDRAM0_BAR_A);
1979 edc0_size = EDRAM0_SIZE_G(size) << 20;
1980 size = t4_read_reg(adap, MA_EDRAM1_BAR_A);
1981 edc1_size = EDRAM1_SIZE_G(size) << 20;
1982 size = t4_read_reg(adap, MA_EXT_MEMORY0_BAR_A);
1983 mc0_size = EXT_MEM0_SIZE_G(size) << 20;
1985 if (t4_read_reg(adap, MA_TARGET_MEM_ENABLE_A) & HMA_MUX_F) {
1986 size = t4_read_reg(adap, MA_EXT_MEMORY1_BAR_A);
1987 hma_size = EXT_MEM1_SIZE_G(size) << 20;
1989 edc0_end = edc0_size;
1990 edc1_end = edc0_end + edc1_size;
1991 mc0_end = edc1_end + mc0_size;
1993 if (offset < edc0_end) {
1996 } else if (offset < edc1_end) {
1998 memaddr = offset - edc0_end;
2000 if (hma_size && (offset < (edc1_end + hma_size))) {
2002 memaddr = offset - edc1_end;
2003 } else if (offset < mc0_end) {
2005 memaddr = offset - edc1_end;
2006 } else if (is_t5(adap->params.chip)) {
2007 size = t4_read_reg(adap, MA_EXT_MEMORY1_BAR_A);
2008 mc1_size = EXT_MEM1_SIZE_G(size) << 20;
2009 mc1_end = mc0_end + mc1_size;
2010 if (offset < mc1_end) {
2012 memaddr = offset - mc0_end;
2014 /* offset beyond the end of any memory */
2018 /* T4/T6 only has a single memory channel */
2023 spin_lock(&adap->win0_lock);
2024 ret = t4_memory_rw(adap, 0, memtype, memaddr, 32, tpte, T4_MEMORY_READ);
2025 spin_unlock(&adap->win0_lock);
2029 dev_err(adap->pdev_dev, "stag %#x, offset %#x out of range\n",
2033 EXPORT_SYMBOL(cxgb4_read_tpte);
2035 u64 cxgb4_read_sge_timestamp(struct net_device *dev)
2038 struct adapter *adap;
2040 adap = netdev2adap(dev);
2041 lo = t4_read_reg(adap, SGE_TIMESTAMP_LO_A);
2042 hi = TSVAL_G(t4_read_reg(adap, SGE_TIMESTAMP_HI_A));
2044 return ((u64)hi << 32) | (u64)lo;
2046 EXPORT_SYMBOL(cxgb4_read_sge_timestamp);
2048 int cxgb4_bar2_sge_qregs(struct net_device *dev,
2050 enum cxgb4_bar2_qtype qtype,
2053 unsigned int *pbar2_qid)
2055 return t4_bar2_sge_qregs(netdev2adap(dev),
2057 (qtype == CXGB4_BAR2_QTYPE_EGRESS
2058 ? T4_BAR2_QTYPE_EGRESS
2059 : T4_BAR2_QTYPE_INGRESS),
2064 EXPORT_SYMBOL(cxgb4_bar2_sge_qregs);
2066 static struct pci_driver cxgb4_driver;
2068 static void check_neigh_update(struct neighbour *neigh)
2070 const struct device *parent;
2071 const struct net_device *netdev = neigh->dev;
2073 if (is_vlan_dev(netdev))
2074 netdev = vlan_dev_real_dev(netdev);
2075 parent = netdev->dev.parent;
2076 if (parent && parent->driver == &cxgb4_driver.driver)
2077 t4_l2t_update(dev_get_drvdata(parent), neigh);
2080 static int netevent_cb(struct notifier_block *nb, unsigned long event,
2084 case NETEVENT_NEIGH_UPDATE:
2085 check_neigh_update(data);
2087 case NETEVENT_REDIRECT:
2094 static bool netevent_registered;
2095 static struct notifier_block cxgb4_netevent_nb = {
2096 .notifier_call = netevent_cb
2099 static void drain_db_fifo(struct adapter *adap, int usecs)
2101 u32 v1, v2, lp_count, hp_count;
2104 v1 = t4_read_reg(adap, SGE_DBFIFO_STATUS_A);
2105 v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2_A);
2106 if (is_t4(adap->params.chip)) {
2107 lp_count = LP_COUNT_G(v1);
2108 hp_count = HP_COUNT_G(v1);
2110 lp_count = LP_COUNT_T5_G(v1);
2111 hp_count = HP_COUNT_T5_G(v2);
2114 if (lp_count == 0 && hp_count == 0)
2116 set_current_state(TASK_UNINTERRUPTIBLE);
2117 schedule_timeout(usecs_to_jiffies(usecs));
2121 static void disable_txq_db(struct sge_txq *q)
2123 unsigned long flags;
2125 spin_lock_irqsave(&q->db_lock, flags);
2127 spin_unlock_irqrestore(&q->db_lock, flags);
2130 static void enable_txq_db(struct adapter *adap, struct sge_txq *q)
2132 spin_lock_irq(&q->db_lock);
2133 if (q->db_pidx_inc) {
2134 /* Make sure that all writes to the TX descriptors
2135 * are committed before we tell HW about them.
2138 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
2139 QID_V(q->cntxt_id) | PIDX_V(q->db_pidx_inc));
2143 spin_unlock_irq(&q->db_lock);
2146 static void disable_dbs(struct adapter *adap)
2150 for_each_ethrxq(&adap->sge, i)
2151 disable_txq_db(&adap->sge.ethtxq[i].q);
2152 if (is_offload(adap)) {
2153 struct sge_uld_txq_info *txq_info =
2154 adap->sge.uld_txq_info[CXGB4_TX_OFLD];
2157 for_each_ofldtxq(&adap->sge, i) {
2158 struct sge_uld_txq *txq = &txq_info->uldtxq[i];
2160 disable_txq_db(&txq->q);
2164 for_each_port(adap, i)
2165 disable_txq_db(&adap->sge.ctrlq[i].q);
2168 static void enable_dbs(struct adapter *adap)
2172 for_each_ethrxq(&adap->sge, i)
2173 enable_txq_db(adap, &adap->sge.ethtxq[i].q);
2174 if (is_offload(adap)) {
2175 struct sge_uld_txq_info *txq_info =
2176 adap->sge.uld_txq_info[CXGB4_TX_OFLD];
2179 for_each_ofldtxq(&adap->sge, i) {
2180 struct sge_uld_txq *txq = &txq_info->uldtxq[i];
2182 enable_txq_db(adap, &txq->q);
2186 for_each_port(adap, i)
2187 enable_txq_db(adap, &adap->sge.ctrlq[i].q);
2190 static void notify_rdma_uld(struct adapter *adap, enum cxgb4_control cmd)
2192 enum cxgb4_uld type = CXGB4_ULD_RDMA;
2194 if (adap->uld && adap->uld[type].handle)
2195 adap->uld[type].control(adap->uld[type].handle, cmd);
2198 static void process_db_full(struct work_struct *work)
2200 struct adapter *adap;
2202 adap = container_of(work, struct adapter, db_full_task);
2204 drain_db_fifo(adap, dbfifo_drain_delay);
2206 notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
2207 if (CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5)
2208 t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
2209 DBFIFO_HP_INT_F | DBFIFO_LP_INT_F,
2210 DBFIFO_HP_INT_F | DBFIFO_LP_INT_F);
2212 t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
2213 DBFIFO_LP_INT_F, DBFIFO_LP_INT_F);
2216 static void sync_txq_pidx(struct adapter *adap, struct sge_txq *q)
2218 u16 hw_pidx, hw_cidx;
2221 spin_lock_irq(&q->db_lock);
2222 ret = read_eq_indices(adap, (u16)q->cntxt_id, &hw_pidx, &hw_cidx);
2225 if (q->db_pidx != hw_pidx) {
2229 if (q->db_pidx >= hw_pidx)
2230 delta = q->db_pidx - hw_pidx;
2232 delta = q->size - hw_pidx + q->db_pidx;
2234 if (is_t4(adap->params.chip))
2235 val = PIDX_V(delta);
2237 val = PIDX_T5_V(delta);
2239 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
2240 QID_V(q->cntxt_id) | val);
2245 spin_unlock_irq(&q->db_lock);
2247 CH_WARN(adap, "DB drop recovery failed.\n");
2250 static void recover_all_queues(struct adapter *adap)
2254 for_each_ethrxq(&adap->sge, i)
2255 sync_txq_pidx(adap, &adap->sge.ethtxq[i].q);
2256 if (is_offload(adap)) {
2257 struct sge_uld_txq_info *txq_info =
2258 adap->sge.uld_txq_info[CXGB4_TX_OFLD];
2260 for_each_ofldtxq(&adap->sge, i) {
2261 struct sge_uld_txq *txq = &txq_info->uldtxq[i];
2263 sync_txq_pidx(adap, &txq->q);
2267 for_each_port(adap, i)
2268 sync_txq_pidx(adap, &adap->sge.ctrlq[i].q);
2271 static void process_db_drop(struct work_struct *work)
2273 struct adapter *adap;
2275 adap = container_of(work, struct adapter, db_drop_task);
2277 if (is_t4(adap->params.chip)) {
2278 drain_db_fifo(adap, dbfifo_drain_delay);
2279 notify_rdma_uld(adap, CXGB4_CONTROL_DB_DROP);
2280 drain_db_fifo(adap, dbfifo_drain_delay);
2281 recover_all_queues(adap);
2282 drain_db_fifo(adap, dbfifo_drain_delay);
2284 notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
2285 } else if (is_t5(adap->params.chip)) {
2286 u32 dropped_db = t4_read_reg(adap, 0x010ac);
2287 u16 qid = (dropped_db >> 15) & 0x1ffff;
2288 u16 pidx_inc = dropped_db & 0x1fff;
2290 unsigned int bar2_qid;
2293 ret = t4_bar2_sge_qregs(adap, qid, T4_BAR2_QTYPE_EGRESS,
2294 0, &bar2_qoffset, &bar2_qid);
2296 dev_err(adap->pdev_dev, "doorbell drop recovery: "
2297 "qid=%d, pidx_inc=%d\n", qid, pidx_inc);
2299 writel(PIDX_T5_V(pidx_inc) | QID_V(bar2_qid),
2300 adap->bar2 + bar2_qoffset + SGE_UDB_KDOORBELL);
2302 /* Re-enable BAR2 WC */
2303 t4_set_reg_field(adap, 0x10b0, 1<<15, 1<<15);
2306 if (CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5)
2307 t4_set_reg_field(adap, SGE_DOORBELL_CONTROL_A, DROPPED_DB_F, 0);
2310 void t4_db_full(struct adapter *adap)
2312 if (is_t4(adap->params.chip)) {
2314 notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
2315 t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
2316 DBFIFO_HP_INT_F | DBFIFO_LP_INT_F, 0);
2317 queue_work(adap->workq, &adap->db_full_task);
2321 void t4_db_dropped(struct adapter *adap)
2323 if (is_t4(adap->params.chip)) {
2325 notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
2327 queue_work(adap->workq, &adap->db_drop_task);
2330 void t4_register_netevent_notifier(void)
2332 if (!netevent_registered) {
2333 register_netevent_notifier(&cxgb4_netevent_nb);
2334 netevent_registered = true;
2338 static void detach_ulds(struct adapter *adap)
2342 mutex_lock(&uld_mutex);
2343 list_del(&adap->list_node);
2345 for (i = 0; i < CXGB4_ULD_MAX; i++)
2346 if (adap->uld && adap->uld[i].handle)
2347 adap->uld[i].state_change(adap->uld[i].handle,
2348 CXGB4_STATE_DETACH);
2350 if (netevent_registered && list_empty(&adapter_list)) {
2351 unregister_netevent_notifier(&cxgb4_netevent_nb);
2352 netevent_registered = false;
2354 mutex_unlock(&uld_mutex);
2357 static void notify_ulds(struct adapter *adap, enum cxgb4_state new_state)
2361 mutex_lock(&uld_mutex);
2362 for (i = 0; i < CXGB4_ULD_MAX; i++)
2363 if (adap->uld && adap->uld[i].handle)
2364 adap->uld[i].state_change(adap->uld[i].handle,
2366 mutex_unlock(&uld_mutex);
2369 #if IS_ENABLED(CONFIG_IPV6)
2370 static int cxgb4_inet6addr_handler(struct notifier_block *this,
2371 unsigned long event, void *data)
2373 struct inet6_ifaddr *ifa = data;
2374 struct net_device *event_dev = ifa->idev->dev;
2375 const struct device *parent = NULL;
2376 #if IS_ENABLED(CONFIG_BONDING)
2377 struct adapter *adap;
2379 if (is_vlan_dev(event_dev))
2380 event_dev = vlan_dev_real_dev(event_dev);
2381 #if IS_ENABLED(CONFIG_BONDING)
2382 if (event_dev->flags & IFF_MASTER) {
2383 list_for_each_entry(adap, &adapter_list, list_node) {
2386 cxgb4_clip_get(adap->port[0],
2387 (const u32 *)ifa, 1);
2390 cxgb4_clip_release(adap->port[0],
2391 (const u32 *)ifa, 1);
2402 parent = event_dev->dev.parent;
2404 if (parent && parent->driver == &cxgb4_driver.driver) {
2407 cxgb4_clip_get(event_dev, (const u32 *)ifa, 1);
2410 cxgb4_clip_release(event_dev, (const u32 *)ifa, 1);
2419 static bool inet6addr_registered;
2420 static struct notifier_block cxgb4_inet6addr_notifier = {
2421 .notifier_call = cxgb4_inet6addr_handler
2424 static void update_clip(const struct adapter *adap)
2427 struct net_device *dev;
2432 for (i = 0; i < MAX_NPORTS; i++) {
2433 dev = adap->port[i];
2437 ret = cxgb4_update_root_dev_clip(dev);
2444 #endif /* IS_ENABLED(CONFIG_IPV6) */
2447 * cxgb_up - enable the adapter
2448 * @adap: adapter being enabled
2450 * Called when the first port is enabled, this function performs the
2451 * actions necessary to make an adapter operational, such as completing
2452 * the initialization of HW modules, and enabling interrupts.
2454 * Must be called with the rtnl lock held.
2456 static int cxgb_up(struct adapter *adap)
2458 struct sge *s = &adap->sge;
2461 mutex_lock(&uld_mutex);
2462 err = setup_sge_queues(adap);
2465 err = setup_rss(adap);
2469 if (adap->flags & CXGB4_USING_MSIX) {
2470 if (s->nd_msix_idx < 0) {
2475 err = request_irq(adap->msix_info[s->nd_msix_idx].vec,
2477 adap->msix_info[s->nd_msix_idx].desc, adap);
2481 err = request_msix_queue_irqs(adap);
2483 goto irq_err_free_nd_msix;
2485 err = request_irq(adap->pdev->irq, t4_intr_handler(adap),
2486 (adap->flags & CXGB4_USING_MSI) ? 0
2488 adap->port[0]->name, adap);
2495 t4_intr_enable(adap);
2496 adap->flags |= CXGB4_FULL_INIT_DONE;
2497 mutex_unlock(&uld_mutex);
2499 notify_ulds(adap, CXGB4_STATE_UP);
2500 #if IS_ENABLED(CONFIG_IPV6)
2505 irq_err_free_nd_msix:
2506 free_irq(adap->msix_info[s->nd_msix_idx].vec, adap);
2508 dev_err(adap->pdev_dev, "request_irq failed, err %d\n", err);
2510 t4_free_sge_resources(adap);
2512 mutex_unlock(&uld_mutex);
2516 static void cxgb_down(struct adapter *adapter)
2518 cancel_work_sync(&adapter->tid_release_task);
2519 cancel_work_sync(&adapter->db_full_task);
2520 cancel_work_sync(&adapter->db_drop_task);
2521 adapter->tid_release_task_busy = false;
2522 adapter->tid_release_head = NULL;
2524 t4_sge_stop(adapter);
2525 t4_free_sge_resources(adapter);
2527 adapter->flags &= ~CXGB4_FULL_INIT_DONE;
2531 * net_device operations
2533 int cxgb_open(struct net_device *dev)
2535 struct port_info *pi = netdev_priv(dev);
2536 struct adapter *adapter = pi->adapter;
2539 netif_carrier_off(dev);
2541 if (!(adapter->flags & CXGB4_FULL_INIT_DONE)) {
2542 err = cxgb_up(adapter);
2547 /* It's possible that the basic port information could have
2548 * changed since we first read it.
2550 err = t4_update_port_info(pi);
2554 err = link_start(dev);
2556 netif_tx_start_all_queues(dev);
2560 int cxgb_close(struct net_device *dev)
2562 struct port_info *pi = netdev_priv(dev);
2563 struct adapter *adapter = pi->adapter;
2566 netif_tx_stop_all_queues(dev);
2567 netif_carrier_off(dev);
2568 ret = t4_enable_pi_params(adapter, adapter->pf, pi,
2569 false, false, false);
2570 #ifdef CONFIG_CHELSIO_T4_DCB
2571 cxgb4_dcb_reset(dev);
2572 dcb_tx_queue_prio_enable(dev, false);
2577 int cxgb4_create_server_filter(const struct net_device *dev, unsigned int stid,
2578 __be32 sip, __be16 sport, __be16 vlan,
2579 unsigned int queue, unsigned char port, unsigned char mask)
2582 struct filter_entry *f;
2583 struct adapter *adap;
2587 adap = netdev2adap(dev);
2589 /* Adjust stid to correct filter index */
2590 stid -= adap->tids.sftid_base;
2591 stid += adap->tids.nftids;
2593 /* Check to make sure the filter requested is writable ...
2595 f = &adap->tids.ftid_tab[stid];
2596 ret = writable_filter(f);
2600 /* Clear out any old resources being used by the filter before
2601 * we start constructing the new filter.
2604 clear_filter(adap, f);
2606 /* Clear out filter specifications */
2607 memset(&f->fs, 0, sizeof(struct ch_filter_specification));
2608 f->fs.val.lport = cpu_to_be16(sport);
2609 f->fs.mask.lport = ~0;
2611 if ((val[0] | val[1] | val[2] | val[3]) != 0) {
2612 for (i = 0; i < 4; i++) {
2613 f->fs.val.lip[i] = val[i];
2614 f->fs.mask.lip[i] = ~0;
2616 if (adap->params.tp.vlan_pri_map & PORT_F) {
2617 f->fs.val.iport = port;
2618 f->fs.mask.iport = mask;
2622 if (adap->params.tp.vlan_pri_map & PROTOCOL_F) {
2623 f->fs.val.proto = IPPROTO_TCP;
2624 f->fs.mask.proto = ~0;
2629 /* Mark filter as locked */
2633 /* Save the actual tid. We need this to get the corresponding
2634 * filter entry structure in filter_rpl.
2636 f->tid = stid + adap->tids.ftid_base;
2637 ret = set_filter_wr(adap, stid);
2639 clear_filter(adap, f);
2645 EXPORT_SYMBOL(cxgb4_create_server_filter);
2647 int cxgb4_remove_server_filter(const struct net_device *dev, unsigned int stid,
2648 unsigned int queue, bool ipv6)
2650 struct filter_entry *f;
2651 struct adapter *adap;
2653 adap = netdev2adap(dev);
2655 /* Adjust stid to correct filter index */
2656 stid -= adap->tids.sftid_base;
2657 stid += adap->tids.nftids;
2659 f = &adap->tids.ftid_tab[stid];
2660 /* Unlock the filter */
2663 return delete_filter(adap, stid);
2665 EXPORT_SYMBOL(cxgb4_remove_server_filter);
2667 static void cxgb_get_stats(struct net_device *dev,
2668 struct rtnl_link_stats64 *ns)
2670 struct port_stats stats;
2671 struct port_info *p = netdev_priv(dev);
2672 struct adapter *adapter = p->adapter;
2674 /* Block retrieving statistics during EEH error
2675 * recovery. Otherwise, the recovery might fail
2676 * and the PCI device will be removed permanently
2678 spin_lock(&adapter->stats_lock);
2679 if (!netif_device_present(dev)) {
2680 spin_unlock(&adapter->stats_lock);
2683 t4_get_port_stats_offset(adapter, p->tx_chan, &stats,
2685 spin_unlock(&adapter->stats_lock);
2687 ns->tx_bytes = stats.tx_octets;
2688 ns->tx_packets = stats.tx_frames;
2689 ns->rx_bytes = stats.rx_octets;
2690 ns->rx_packets = stats.rx_frames;
2691 ns->multicast = stats.rx_mcast_frames;
2693 /* detailed rx_errors */
2694 ns->rx_length_errors = stats.rx_jabber + stats.rx_too_long +
2696 ns->rx_over_errors = 0;
2697 ns->rx_crc_errors = stats.rx_fcs_err;
2698 ns->rx_frame_errors = stats.rx_symbol_err;
2699 ns->rx_dropped = stats.rx_ovflow0 + stats.rx_ovflow1 +
2700 stats.rx_ovflow2 + stats.rx_ovflow3 +
2701 stats.rx_trunc0 + stats.rx_trunc1 +
2702 stats.rx_trunc2 + stats.rx_trunc3;
2703 ns->rx_missed_errors = 0;
2705 /* detailed tx_errors */
2706 ns->tx_aborted_errors = 0;
2707 ns->tx_carrier_errors = 0;
2708 ns->tx_fifo_errors = 0;
2709 ns->tx_heartbeat_errors = 0;
2710 ns->tx_window_errors = 0;
2712 ns->tx_errors = stats.tx_error_frames;
2713 ns->rx_errors = stats.rx_symbol_err + stats.rx_fcs_err +
2714 ns->rx_length_errors + stats.rx_len_err + ns->rx_fifo_errors;
2717 static int cxgb_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
2720 int ret = 0, prtad, devad;
2721 struct port_info *pi = netdev_priv(dev);
2722 struct adapter *adapter = pi->adapter;
2723 struct mii_ioctl_data *data = (struct mii_ioctl_data *)&req->ifr_data;
2727 if (pi->mdio_addr < 0)
2729 data->phy_id = pi->mdio_addr;
2733 if (mdio_phy_id_is_c45(data->phy_id)) {
2734 prtad = mdio_phy_id_prtad(data->phy_id);
2735 devad = mdio_phy_id_devad(data->phy_id);
2736 } else if (data->phy_id < 32) {
2737 prtad = data->phy_id;
2739 data->reg_num &= 0x1f;
2743 mbox = pi->adapter->pf;
2744 if (cmd == SIOCGMIIREG)
2745 ret = t4_mdio_rd(pi->adapter, mbox, prtad, devad,
2746 data->reg_num, &data->val_out);
2748 ret = t4_mdio_wr(pi->adapter, mbox, prtad, devad,
2749 data->reg_num, data->val_in);
2752 return copy_to_user(req->ifr_data, &pi->tstamp_config,
2753 sizeof(pi->tstamp_config)) ?
2756 if (copy_from_user(&pi->tstamp_config, req->ifr_data,
2757 sizeof(pi->tstamp_config)))
2760 if (!is_t4(adapter->params.chip)) {
2761 switch (pi->tstamp_config.tx_type) {
2762 case HWTSTAMP_TX_OFF:
2763 case HWTSTAMP_TX_ON:
2769 switch (pi->tstamp_config.rx_filter) {
2770 case HWTSTAMP_FILTER_NONE:
2771 pi->rxtstamp = false;
2773 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
2774 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
2775 cxgb4_ptprx_timestamping(pi, pi->port_id,
2778 case HWTSTAMP_FILTER_PTP_V2_EVENT:
2779 cxgb4_ptprx_timestamping(pi, pi->port_id,
2782 case HWTSTAMP_FILTER_ALL:
2783 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
2784 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
2785 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
2786 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
2787 pi->rxtstamp = true;
2790 pi->tstamp_config.rx_filter =
2791 HWTSTAMP_FILTER_NONE;
2795 if ((pi->tstamp_config.tx_type == HWTSTAMP_TX_OFF) &&
2796 (pi->tstamp_config.rx_filter ==
2797 HWTSTAMP_FILTER_NONE)) {
2798 if (cxgb4_ptp_txtype(adapter, pi->port_id) >= 0)
2799 pi->ptp_enable = false;
2802 if (pi->tstamp_config.rx_filter !=
2803 HWTSTAMP_FILTER_NONE) {
2804 if (cxgb4_ptp_redirect_rx_packet(adapter,
2806 pi->ptp_enable = true;
2809 /* For T4 Adapters */
2810 switch (pi->tstamp_config.rx_filter) {
2811 case HWTSTAMP_FILTER_NONE:
2812 pi->rxtstamp = false;
2814 case HWTSTAMP_FILTER_ALL:
2815 pi->rxtstamp = true;
2818 pi->tstamp_config.rx_filter =
2819 HWTSTAMP_FILTER_NONE;
2823 return copy_to_user(req->ifr_data, &pi->tstamp_config,
2824 sizeof(pi->tstamp_config)) ?
2832 static void cxgb_set_rxmode(struct net_device *dev)
2834 /* unfortunately we can't return errors to the stack */
2835 set_rxmode(dev, -1, false);
2838 static int cxgb_change_mtu(struct net_device *dev, int new_mtu)
2841 struct port_info *pi = netdev_priv(dev);
2843 ret = t4_set_rxmode(pi->adapter, pi->adapter->pf, pi->viid, new_mtu, -1,
2850 #ifdef CONFIG_PCI_IOV
2851 static int cxgb4_mgmt_open(struct net_device *dev)
2853 /* Turn carrier off since we don't have to transmit anything on this
2856 netif_carrier_off(dev);
2860 /* Fill MAC address that will be assigned by the FW */
2861 static void cxgb4_mgmt_fill_vf_station_mac_addr(struct adapter *adap)
2863 u8 hw_addr[ETH_ALEN], macaddr[ETH_ALEN];
2864 unsigned int i, vf, nvfs;
2869 adap->params.pci.vpd_cap_addr = pci_find_capability(adap->pdev,
2871 err = t4_get_raw_vpd_params(adap, &adap->params.vpd);
2875 na = adap->params.vpd.na;
2876 for (i = 0; i < ETH_ALEN; i++)
2877 hw_addr[i] = (hex2val(na[2 * i + 0]) * 16 +
2878 hex2val(na[2 * i + 1]));
2880 a = (hw_addr[0] << 8) | hw_addr[1];
2881 b = (hw_addr[1] << 8) | hw_addr[2];
2883 a |= 0x0200; /* locally assigned Ethernet MAC address */
2884 a &= ~0x0100; /* not a multicast Ethernet MAC address */
2885 macaddr[0] = a >> 8;
2886 macaddr[1] = a & 0xff;
2888 for (i = 2; i < 5; i++)
2889 macaddr[i] = hw_addr[i + 1];
2891 for (vf = 0, nvfs = pci_sriov_get_totalvfs(adap->pdev);
2893 macaddr[5] = adap->pf * nvfs + vf;
2894 ether_addr_copy(adap->vfinfo[vf].vf_mac_addr, macaddr);
2898 static int cxgb4_mgmt_set_vf_mac(struct net_device *dev, int vf, u8 *mac)
2900 struct port_info *pi = netdev_priv(dev);
2901 struct adapter *adap = pi->adapter;
2904 /* verify MAC addr is valid */
2905 if (!is_valid_ether_addr(mac)) {
2906 dev_err(pi->adapter->pdev_dev,
2907 "Invalid Ethernet address %pM for VF %d\n",
2912 dev_info(pi->adapter->pdev_dev,
2913 "Setting MAC %pM on VF %d\n", mac, vf);
2914 ret = t4_set_vf_mac_acl(adap, vf + 1, 1, mac);
2916 ether_addr_copy(adap->vfinfo[vf].vf_mac_addr, mac);
2920 static int cxgb4_mgmt_get_vf_config(struct net_device *dev,
2921 int vf, struct ifla_vf_info *ivi)
2923 struct port_info *pi = netdev_priv(dev);
2924 struct adapter *adap = pi->adapter;
2925 struct vf_info *vfinfo;
2927 if (vf >= adap->num_vfs)
2929 vfinfo = &adap->vfinfo[vf];
2932 ivi->max_tx_rate = vfinfo->tx_rate;
2933 ivi->min_tx_rate = 0;
2934 ether_addr_copy(ivi->mac, vfinfo->vf_mac_addr);
2935 ivi->vlan = vfinfo->vlan;
2936 ivi->linkstate = vfinfo->link_state;
2940 static int cxgb4_mgmt_get_phys_port_id(struct net_device *dev,
2941 struct netdev_phys_item_id *ppid)
2943 struct port_info *pi = netdev_priv(dev);
2944 unsigned int phy_port_id;
2946 phy_port_id = pi->adapter->adap_idx * 10 + pi->port_id;
2947 ppid->id_len = sizeof(phy_port_id);
2948 memcpy(ppid->id, &phy_port_id, ppid->id_len);
2952 static int cxgb4_mgmt_set_vf_rate(struct net_device *dev, int vf,
2953 int min_tx_rate, int max_tx_rate)
2955 struct port_info *pi = netdev_priv(dev);
2956 struct adapter *adap = pi->adapter;
2957 unsigned int link_ok, speed, mtu;
2958 u32 fw_pfvf, fw_class;
2963 if (vf >= adap->num_vfs)
2967 dev_err(adap->pdev_dev,
2968 "Min tx rate (%d) (> 0) for VF %d is Invalid.\n",
2973 if (max_tx_rate == 0) {
2974 /* unbind VF to to any Traffic Class */
2976 (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) |
2977 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_SCHEDCLASS_ETH));
2978 fw_class = 0xffffffff;
2979 ret = t4_set_params(adap, adap->mbox, adap->pf, vf + 1, 1,
2980 &fw_pfvf, &fw_class);
2982 dev_err(adap->pdev_dev,
2983 "Err %d in unbinding PF %d VF %d from TX Rate Limiting\n",
2987 dev_info(adap->pdev_dev,
2988 "PF %d VF %d is unbound from TX Rate Limiting\n",
2990 adap->vfinfo[vf].tx_rate = 0;
2994 ret = t4_get_link_params(pi, &link_ok, &speed, &mtu);
2995 if (ret != FW_SUCCESS) {
2996 dev_err(adap->pdev_dev,
2997 "Failed to get link information for VF %d\n", vf);
3002 dev_err(adap->pdev_dev, "Link down for VF %d\n", vf);
3006 if (max_tx_rate > speed) {
3007 dev_err(adap->pdev_dev,
3008 "Max tx rate %d for VF %d can't be > link-speed %u",
3009 max_tx_rate, vf, speed);
3014 /* subtract ethhdr size and 4 bytes crc since, f/w appends it */
3015 pktsize = pktsize - sizeof(struct ethhdr) - 4;
3016 /* subtract ipv4 hdr size, tcp hdr size to get typical IPv4 MSS size */
3017 pktsize = pktsize - sizeof(struct iphdr) - sizeof(struct tcphdr);
3018 /* configure Traffic Class for rate-limiting */
3019 ret = t4_sched_params(adap, SCHED_CLASS_TYPE_PACKET,
3020 SCHED_CLASS_LEVEL_CL_RL,
3021 SCHED_CLASS_MODE_CLASS,
3022 SCHED_CLASS_RATEUNIT_BITS,
3023 SCHED_CLASS_RATEMODE_ABS,
3024 pi->tx_chan, class_id, 0,
3025 max_tx_rate * 1000, 0, pktsize, 0);
3027 dev_err(adap->pdev_dev, "Err %d for Traffic Class config\n",
3031 dev_info(adap->pdev_dev,
3032 "Class %d with MSS %u configured with rate %u\n",
3033 class_id, pktsize, max_tx_rate);
3035 /* bind VF to configured Traffic Class */
3036 fw_pfvf = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) |
3037 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_SCHEDCLASS_ETH));
3038 fw_class = class_id;
3039 ret = t4_set_params(adap, adap->mbox, adap->pf, vf + 1, 1, &fw_pfvf,
3042 dev_err(adap->pdev_dev,
3043 "Err %d in binding PF %d VF %d to Traffic Class %d\n",
3044 ret, adap->pf, vf, class_id);
3047 dev_info(adap->pdev_dev, "PF %d VF %d is bound to Class %d\n",
3048 adap->pf, vf, class_id);
3049 adap->vfinfo[vf].tx_rate = max_tx_rate;
3053 static int cxgb4_mgmt_set_vf_vlan(struct net_device *dev, int vf,
3054 u16 vlan, u8 qos, __be16 vlan_proto)
3056 struct port_info *pi = netdev_priv(dev);
3057 struct adapter *adap = pi->adapter;
3060 if (vf >= adap->num_vfs || vlan > 4095 || qos > 7)
3063 if (vlan_proto != htons(ETH_P_8021Q) || qos != 0)
3064 return -EPROTONOSUPPORT;
3066 ret = t4_set_vlan_acl(adap, adap->mbox, vf + 1, vlan);
3068 adap->vfinfo[vf].vlan = vlan;
3072 dev_err(adap->pdev_dev, "Err %d %s VLAN ACL for PF/VF %d/%d\n",
3073 ret, (vlan ? "setting" : "clearing"), adap->pf, vf);
3077 static int cxgb4_mgmt_set_vf_link_state(struct net_device *dev, int vf,
3080 struct port_info *pi = netdev_priv(dev);
3081 struct adapter *adap = pi->adapter;
3085 if (vf >= adap->num_vfs)
3089 case IFLA_VF_LINK_STATE_AUTO:
3090 val = FW_VF_LINK_STATE_AUTO;
3093 case IFLA_VF_LINK_STATE_ENABLE:
3094 val = FW_VF_LINK_STATE_ENABLE;
3097 case IFLA_VF_LINK_STATE_DISABLE:
3098 val = FW_VF_LINK_STATE_DISABLE;
3105 param = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) |
3106 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_LINK_STATE));
3107 ret = t4_set_params(adap, adap->mbox, adap->pf, vf + 1, 1,
3110 dev_err(adap->pdev_dev,
3111 "Error %d in setting PF %d VF %d link state\n",
3116 adap->vfinfo[vf].link_state = link;
3119 #endif /* CONFIG_PCI_IOV */
3121 static int cxgb_set_mac_addr(struct net_device *dev, void *p)
3124 struct sockaddr *addr = p;
3125 struct port_info *pi = netdev_priv(dev);
3127 if (!is_valid_ether_addr(addr->sa_data))
3128 return -EADDRNOTAVAIL;
3130 ret = cxgb4_update_mac_filt(pi, pi->viid, &pi->xact_addr_filt,
3131 addr->sa_data, true, &pi->smt_idx);
3135 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
3139 #ifdef CONFIG_NET_POLL_CONTROLLER
3140 static void cxgb_netpoll(struct net_device *dev)
3142 struct port_info *pi = netdev_priv(dev);
3143 struct adapter *adap = pi->adapter;
3145 if (adap->flags & CXGB4_USING_MSIX) {
3147 struct sge_eth_rxq *rx = &adap->sge.ethrxq[pi->first_qset];
3149 for (i = pi->nqsets; i; i--, rx++)
3150 t4_sge_intr_msix(0, &rx->rspq);
3152 t4_intr_handler(adap)(0, adap);
3156 static int cxgb_set_tx_maxrate(struct net_device *dev, int index, u32 rate)
3158 struct port_info *pi = netdev_priv(dev);
3159 struct adapter *adap = pi->adapter;
3160 struct ch_sched_queue qe = { 0 };
3161 struct ch_sched_params p = { 0 };
3162 struct sched_class *e;
3166 if (!can_sched(dev))
3169 if (index < 0 || index > pi->nqsets - 1)
3172 if (!(adap->flags & CXGB4_FULL_INIT_DONE)) {
3173 dev_err(adap->pdev_dev,
3174 "Failed to rate limit on queue %d. Link Down?\n",
3180 e = cxgb4_sched_queue_lookup(dev, &qe);
3181 if (e && e->info.u.params.level != SCHED_CLASS_LEVEL_CL_RL) {
3182 dev_err(adap->pdev_dev,
3183 "Queue %u already bound to class %u of type: %u\n",
3184 index, e->idx, e->info.u.params.level);
3188 /* Convert from Mbps to Kbps */
3189 req_rate = rate * 1000;
3191 /* Max rate is 100 Gbps */
3192 if (req_rate > SCHED_MAX_RATE_KBPS) {
3193 dev_err(adap->pdev_dev,
3194 "Invalid rate %u Mbps, Max rate is %u Mbps\n",
3195 rate, SCHED_MAX_RATE_KBPS / 1000);
3199 /* First unbind the queue from any existing class */
3200 memset(&qe, 0, sizeof(qe));
3202 qe.class = SCHED_CLS_NONE;
3204 err = cxgb4_sched_class_unbind(dev, (void *)(&qe), SCHED_QUEUE);
3206 dev_err(adap->pdev_dev,
3207 "Unbinding Queue %d on port %d fail. Err: %d\n",
3208 index, pi->port_id, err);
3212 /* Queue already unbound */
3216 /* Fetch any available unused or matching scheduling class */
3217 p.type = SCHED_CLASS_TYPE_PACKET;
3218 p.u.params.level = SCHED_CLASS_LEVEL_CL_RL;
3219 p.u.params.mode = SCHED_CLASS_MODE_CLASS;
3220 p.u.params.rateunit = SCHED_CLASS_RATEUNIT_BITS;
3221 p.u.params.ratemode = SCHED_CLASS_RATEMODE_ABS;
3222 p.u.params.channel = pi->tx_chan;
3223 p.u.params.class = SCHED_CLS_NONE;
3224 p.u.params.minrate = 0;
3225 p.u.params.maxrate = req_rate;
3226 p.u.params.weight = 0;
3227 p.u.params.pktsize = dev->mtu;
3229 e = cxgb4_sched_class_alloc(dev, &p);
3233 /* Bind the queue to a scheduling class */
3234 memset(&qe, 0, sizeof(qe));
3238 err = cxgb4_sched_class_bind(dev, (void *)(&qe), SCHED_QUEUE);
3240 dev_err(adap->pdev_dev,
3241 "Queue rate limiting failed. Err: %d\n", err);
3245 static int cxgb_setup_tc_flower(struct net_device *dev,
3246 struct flow_cls_offload *cls_flower)
3248 switch (cls_flower->command) {
3249 case FLOW_CLS_REPLACE:
3250 return cxgb4_tc_flower_replace(dev, cls_flower);
3251 case FLOW_CLS_DESTROY:
3252 return cxgb4_tc_flower_destroy(dev, cls_flower);
3253 case FLOW_CLS_STATS:
3254 return cxgb4_tc_flower_stats(dev, cls_flower);
3260 static int cxgb_setup_tc_cls_u32(struct net_device *dev,
3261 struct tc_cls_u32_offload *cls_u32)
3263 switch (cls_u32->command) {
3264 case TC_CLSU32_NEW_KNODE:
3265 case TC_CLSU32_REPLACE_KNODE:
3266 return cxgb4_config_knode(dev, cls_u32);
3267 case TC_CLSU32_DELETE_KNODE:
3268 return cxgb4_delete_knode(dev, cls_u32);
3274 static int cxgb_setup_tc_matchall(struct net_device *dev,
3275 struct tc_cls_matchall_offload *cls_matchall,
3278 struct adapter *adap = netdev2adap(dev);
3280 if (!adap->tc_matchall)
3283 switch (cls_matchall->command) {
3284 case TC_CLSMATCHALL_REPLACE:
3285 return cxgb4_tc_matchall_replace(dev, cls_matchall, ingress);
3286 case TC_CLSMATCHALL_DESTROY:
3287 return cxgb4_tc_matchall_destroy(dev, cls_matchall, ingress);
3288 case TC_CLSMATCHALL_STATS:
3290 return cxgb4_tc_matchall_stats(dev, cls_matchall);
3299 static int cxgb_setup_tc_block_ingress_cb(enum tc_setup_type type,
3300 void *type_data, void *cb_priv)
3302 struct net_device *dev = cb_priv;
3303 struct port_info *pi = netdev2pinfo(dev);
3304 struct adapter *adap = netdev2adap(dev);
3306 if (!(adap->flags & CXGB4_FULL_INIT_DONE)) {
3307 dev_err(adap->pdev_dev,
3308 "Failed to setup tc on port %d. Link Down?\n",
3313 if (!tc_cls_can_offload_and_chain0(dev, type_data))
3317 case TC_SETUP_CLSU32:
3318 return cxgb_setup_tc_cls_u32(dev, type_data);
3319 case TC_SETUP_CLSFLOWER:
3320 return cxgb_setup_tc_flower(dev, type_data);
3321 case TC_SETUP_CLSMATCHALL:
3322 return cxgb_setup_tc_matchall(dev, type_data, true);
3328 static int cxgb_setup_tc_block_egress_cb(enum tc_setup_type type,
3329 void *type_data, void *cb_priv)
3331 struct net_device *dev = cb_priv;
3332 struct port_info *pi = netdev2pinfo(dev);
3333 struct adapter *adap = netdev2adap(dev);
3335 if (!(adap->flags & CXGB4_FULL_INIT_DONE)) {
3336 dev_err(adap->pdev_dev,
3337 "Failed to setup tc on port %d. Link Down?\n",
3342 if (!tc_cls_can_offload_and_chain0(dev, type_data))
3346 case TC_SETUP_CLSMATCHALL:
3347 return cxgb_setup_tc_matchall(dev, type_data, false);
3355 static int cxgb_setup_tc_mqprio(struct net_device *dev,
3356 struct tc_mqprio_qopt_offload *mqprio)
3358 struct adapter *adap = netdev2adap(dev);
3360 if (!is_ethofld(adap) || !adap->tc_mqprio)
3363 return cxgb4_setup_tc_mqprio(dev, mqprio);
3366 static LIST_HEAD(cxgb_block_cb_list);
3368 static int cxgb_setup_tc_block(struct net_device *dev,
3369 struct flow_block_offload *f)
3371 struct port_info *pi = netdev_priv(dev);
3372 flow_setup_cb_t *cb;
3375 pi->tc_block_shared = f->block_shared;
3376 if (f->binder_type == FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS) {
3377 cb = cxgb_setup_tc_block_egress_cb;
3378 ingress_only = false;
3380 cb = cxgb_setup_tc_block_ingress_cb;
3381 ingress_only = true;
3384 return flow_block_cb_setup_simple(f, &cxgb_block_cb_list,
3385 cb, pi, dev, ingress_only);
3388 static int cxgb_setup_tc(struct net_device *dev, enum tc_setup_type type,
3392 case TC_SETUP_QDISC_MQPRIO:
3393 return cxgb_setup_tc_mqprio(dev, type_data);
3394 case TC_SETUP_BLOCK:
3395 return cxgb_setup_tc_block(dev, type_data);
3401 static void cxgb_del_udp_tunnel(struct net_device *netdev,
3402 struct udp_tunnel_info *ti)
3404 struct port_info *pi = netdev_priv(netdev);
3405 struct adapter *adapter = pi->adapter;
3406 unsigned int chip_ver = CHELSIO_CHIP_VERSION(adapter->params.chip);
3407 u8 match_all_mac[] = { 0, 0, 0, 0, 0, 0 };
3410 if (chip_ver < CHELSIO_T6)
3414 case UDP_TUNNEL_TYPE_VXLAN:
3415 if (!adapter->vxlan_port_cnt ||
3416 adapter->vxlan_port != ti->port)
3417 return; /* Invalid VxLAN destination port */
3419 adapter->vxlan_port_cnt--;
3420 if (adapter->vxlan_port_cnt)
3423 adapter->vxlan_port = 0;
3424 t4_write_reg(adapter, MPS_RX_VXLAN_TYPE_A, 0);
3426 case UDP_TUNNEL_TYPE_GENEVE:
3427 if (!adapter->geneve_port_cnt ||
3428 adapter->geneve_port != ti->port)
3429 return; /* Invalid GENEVE destination port */
3431 adapter->geneve_port_cnt--;
3432 if (adapter->geneve_port_cnt)
3435 adapter->geneve_port = 0;
3436 t4_write_reg(adapter, MPS_RX_GENEVE_TYPE_A, 0);
3442 /* Matchall mac entries can be deleted only after all tunnel ports
3443 * are brought down or removed.
3445 if (!adapter->rawf_cnt)
3447 for_each_port(adapter, i) {
3448 pi = adap2pinfo(adapter, i);
3449 ret = t4_free_raw_mac_filt(adapter, pi->viid,
3450 match_all_mac, match_all_mac,
3451 adapter->rawf_start +
3453 1, pi->port_id, false);
3455 netdev_info(netdev, "Failed to free mac filter entry, for port %d\n",
3462 static void cxgb_add_udp_tunnel(struct net_device *netdev,
3463 struct udp_tunnel_info *ti)
3465 struct port_info *pi = netdev_priv(netdev);
3466 struct adapter *adapter = pi->adapter;
3467 unsigned int chip_ver = CHELSIO_CHIP_VERSION(adapter->params.chip);
3468 u8 match_all_mac[] = { 0, 0, 0, 0, 0, 0 };
3471 if (chip_ver < CHELSIO_T6 || !adapter->rawf_cnt)
3475 case UDP_TUNNEL_TYPE_VXLAN:
3476 /* Callback for adding vxlan port can be called with the same
3477 * port for both IPv4 and IPv6. We should not disable the
3478 * offloading when the same port for both protocols is added
3479 * and later one of them is removed.
3481 if (adapter->vxlan_port_cnt &&
3482 adapter->vxlan_port == ti->port) {
3483 adapter->vxlan_port_cnt++;
3487 /* We will support only one VxLAN port */
3488 if (adapter->vxlan_port_cnt) {
3489 netdev_info(netdev, "UDP port %d already offloaded, not adding port %d\n",
3490 be16_to_cpu(adapter->vxlan_port),
3491 be16_to_cpu(ti->port));
3495 adapter->vxlan_port = ti->port;
3496 adapter->vxlan_port_cnt = 1;
3498 t4_write_reg(adapter, MPS_RX_VXLAN_TYPE_A,
3499 VXLAN_V(be16_to_cpu(ti->port)) | VXLAN_EN_F);
3501 case UDP_TUNNEL_TYPE_GENEVE:
3502 if (adapter->geneve_port_cnt &&
3503 adapter->geneve_port == ti->port) {
3504 adapter->geneve_port_cnt++;
3508 /* We will support only one GENEVE port */
3509 if (adapter->geneve_port_cnt) {
3510 netdev_info(netdev, "UDP port %d already offloaded, not adding port %d\n",
3511 be16_to_cpu(adapter->geneve_port),
3512 be16_to_cpu(ti->port));
3516 adapter->geneve_port = ti->port;
3517 adapter->geneve_port_cnt = 1;
3519 t4_write_reg(adapter, MPS_RX_GENEVE_TYPE_A,
3520 GENEVE_V(be16_to_cpu(ti->port)) | GENEVE_EN_F);
3526 /* Create a 'match all' mac filter entry for inner mac,
3527 * if raw mac interface is supported. Once the linux kernel provides
3528 * driver entry points for adding/deleting the inner mac addresses,
3529 * we will remove this 'match all' entry and fallback to adding
3530 * exact match filters.
3532 for_each_port(adapter, i) {
3533 pi = adap2pinfo(adapter, i);
3535 ret = t4_alloc_raw_mac_filt(adapter, pi->viid,
3538 adapter->rawf_start +
3540 1, pi->port_id, false);
3542 netdev_info(netdev, "Failed to allocate a mac filter entry, not adding port %d\n",
3543 be16_to_cpu(ti->port));
3544 cxgb_del_udp_tunnel(netdev, ti);
3550 static netdev_features_t cxgb_features_check(struct sk_buff *skb,
3551 struct net_device *dev,
3552 netdev_features_t features)
3554 struct port_info *pi = netdev_priv(dev);
3555 struct adapter *adapter = pi->adapter;
3557 if (CHELSIO_CHIP_VERSION(adapter->params.chip) < CHELSIO_T6)
3560 /* Check if hw supports offload for this packet */
3561 if (!skb->encapsulation || cxgb_encap_offload_supported(skb))
3564 /* Offload is not supported for this encapsulated packet */
3565 return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3568 static netdev_features_t cxgb_fix_features(struct net_device *dev,
3569 netdev_features_t features)
3571 /* Disable GRO, if RX_CSUM is disabled */
3572 if (!(features & NETIF_F_RXCSUM))
3573 features &= ~NETIF_F_GRO;
3578 static const struct net_device_ops cxgb4_netdev_ops = {
3579 .ndo_open = cxgb_open,
3580 .ndo_stop = cxgb_close,
3581 .ndo_start_xmit = t4_start_xmit,
3582 .ndo_select_queue = cxgb_select_queue,
3583 .ndo_get_stats64 = cxgb_get_stats,
3584 .ndo_set_rx_mode = cxgb_set_rxmode,
3585 .ndo_set_mac_address = cxgb_set_mac_addr,
3586 .ndo_set_features = cxgb_set_features,
3587 .ndo_validate_addr = eth_validate_addr,
3588 .ndo_do_ioctl = cxgb_ioctl,
3589 .ndo_change_mtu = cxgb_change_mtu,
3590 #ifdef CONFIG_NET_POLL_CONTROLLER
3591 .ndo_poll_controller = cxgb_netpoll,
3593 #ifdef CONFIG_CHELSIO_T4_FCOE
3594 .ndo_fcoe_enable = cxgb_fcoe_enable,
3595 .ndo_fcoe_disable = cxgb_fcoe_disable,
3596 #endif /* CONFIG_CHELSIO_T4_FCOE */
3597 .ndo_set_tx_maxrate = cxgb_set_tx_maxrate,
3598 .ndo_setup_tc = cxgb_setup_tc,
3599 .ndo_udp_tunnel_add = cxgb_add_udp_tunnel,
3600 .ndo_udp_tunnel_del = cxgb_del_udp_tunnel,
3601 .ndo_features_check = cxgb_features_check,
3602 .ndo_fix_features = cxgb_fix_features,
3605 #ifdef CONFIG_PCI_IOV
3606 static const struct net_device_ops cxgb4_mgmt_netdev_ops = {
3607 .ndo_open = cxgb4_mgmt_open,
3608 .ndo_set_vf_mac = cxgb4_mgmt_set_vf_mac,
3609 .ndo_get_vf_config = cxgb4_mgmt_get_vf_config,
3610 .ndo_set_vf_rate = cxgb4_mgmt_set_vf_rate,
3611 .ndo_get_phys_port_id = cxgb4_mgmt_get_phys_port_id,
3612 .ndo_set_vf_vlan = cxgb4_mgmt_set_vf_vlan,
3613 .ndo_set_vf_link_state = cxgb4_mgmt_set_vf_link_state,
3617 static void cxgb4_mgmt_get_drvinfo(struct net_device *dev,
3618 struct ethtool_drvinfo *info)
3620 struct adapter *adapter = netdev2adap(dev);
3622 strlcpy(info->driver, cxgb4_driver_name, sizeof(info->driver));
3623 strlcpy(info->bus_info, pci_name(adapter->pdev),
3624 sizeof(info->bus_info));
3627 static const struct ethtool_ops cxgb4_mgmt_ethtool_ops = {
3628 .get_drvinfo = cxgb4_mgmt_get_drvinfo,
3631 static void notify_fatal_err(struct work_struct *work)
3633 struct adapter *adap;
3635 adap = container_of(work, struct adapter, fatal_err_notify_task);
3636 notify_ulds(adap, CXGB4_STATE_FATAL_ERROR);
3639 void t4_fatal_err(struct adapter *adap)
3643 if (pci_channel_offline(adap->pdev))
3646 /* Disable the SGE since ULDs are going to free resources that
3647 * could be exposed to the adapter. RDMA MWs for example...
3649 t4_shutdown_adapter(adap);
3650 for_each_port(adap, port) {
3651 struct net_device *dev = adap->port[port];
3653 /* If we get here in very early initialization the network
3654 * devices may not have been set up yet.
3659 netif_tx_stop_all_queues(dev);
3660 netif_carrier_off(dev);
3662 dev_alert(adap->pdev_dev, "encountered fatal error, adapter stopped\n");
3663 queue_work(adap->workq, &adap->fatal_err_notify_task);
3666 static void setup_memwin(struct adapter *adap)
3668 u32 nic_win_base = t4_get_util_window(adap);
3670 t4_setup_memwin(adap, nic_win_base, MEMWIN_NIC);
3673 static void setup_memwin_rdma(struct adapter *adap)
3675 if (adap->vres.ocq.size) {
3679 start = t4_read_pcie_cfg4(adap, PCI_BASE_ADDRESS_2);
3680 start &= PCI_BASE_ADDRESS_MEM_MASK;
3681 start += OCQ_WIN_OFFSET(adap->pdev, &adap->vres);
3682 sz_kb = roundup_pow_of_two(adap->vres.ocq.size) >> 10;
3684 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN_A, 3),
3685 start | BIR_V(1) | WINDOW_V(ilog2(sz_kb)));
3687 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET_A, 3),
3688 adap->vres.ocq.start);
3690 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET_A, 3));
3694 /* HMA Definitions */
3696 /* The maximum number of address that can be send in a single FW cmd */
3697 #define HMA_MAX_ADDR_IN_CMD 5
3699 #define HMA_PAGE_SIZE PAGE_SIZE
3701 #define HMA_MAX_NO_FW_ADDRESS (16 << 10) /* FW supports 16K addresses */
3703 #define HMA_PAGE_ORDER \
3704 ((HMA_PAGE_SIZE < HMA_MAX_NO_FW_ADDRESS) ? \
3705 ilog2(HMA_MAX_NO_FW_ADDRESS / HMA_PAGE_SIZE) : 0)
3707 /* The minimum and maximum possible HMA sizes that can be specified in the FW
3708 * configuration(in units of MB).
3710 #define HMA_MIN_TOTAL_SIZE 1
3711 #define HMA_MAX_TOTAL_SIZE \
3712 (((HMA_PAGE_SIZE << HMA_PAGE_ORDER) * \
3713 HMA_MAX_NO_FW_ADDRESS) >> 20)
3715 static void adap_free_hma_mem(struct adapter *adapter)
3717 struct scatterlist *iter;
3721 if (!adapter->hma.sgt)
3724 if (adapter->hma.flags & HMA_DMA_MAPPED_FLAG) {
3725 dma_unmap_sg(adapter->pdev_dev, adapter->hma.sgt->sgl,
3726 adapter->hma.sgt->nents, PCI_DMA_BIDIRECTIONAL);
3727 adapter->hma.flags &= ~HMA_DMA_MAPPED_FLAG;
3730 for_each_sg(adapter->hma.sgt->sgl, iter,
3731 adapter->hma.sgt->orig_nents, i) {
3732 page = sg_page(iter);
3734 __free_pages(page, HMA_PAGE_ORDER);
3737 kfree(adapter->hma.phy_addr);
3738 sg_free_table(adapter->hma.sgt);
3739 kfree(adapter->hma.sgt);
3740 adapter->hma.sgt = NULL;
3743 static int adap_config_hma(struct adapter *adapter)
3745 struct scatterlist *sgl, *iter;
3746 struct sg_table *sgt;
3747 struct page *newpage;
3748 unsigned int i, j, k;
3749 u32 param, hma_size;
3755 /* HMA is supported only for T6+ cards.
3756 * Avoid initializing HMA in kdump kernels.
3758 if (is_kdump_kernel() ||
3759 CHELSIO_CHIP_VERSION(adapter->params.chip) < CHELSIO_T6)
3762 /* Get the HMA region size required by fw */
3763 param = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
3764 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_HMA_SIZE));
3765 ret = t4_query_params(adapter, adapter->mbox, adapter->pf, 0,
3766 1, ¶m, &hma_size);
3767 /* An error means card has its own memory or HMA is not supported by
3768 * the firmware. Return without any errors.
3770 if (ret || !hma_size)
3773 if (hma_size < HMA_MIN_TOTAL_SIZE ||
3774 hma_size > HMA_MAX_TOTAL_SIZE) {
3775 dev_err(adapter->pdev_dev,
3776 "HMA size %uMB beyond bounds(%u-%lu)MB\n",
3777 hma_size, HMA_MIN_TOTAL_SIZE, HMA_MAX_TOTAL_SIZE);
3781 page_size = HMA_PAGE_SIZE;
3782 page_order = HMA_PAGE_ORDER;
3783 adapter->hma.sgt = kzalloc(sizeof(*adapter->hma.sgt), GFP_KERNEL);
3784 if (unlikely(!adapter->hma.sgt)) {
3785 dev_err(adapter->pdev_dev, "HMA SG table allocation failed\n");
3788 sgt = adapter->hma.sgt;
3789 /* FW returned value will be in MB's
3791 sgt->orig_nents = (hma_size << 20) / (page_size << page_order);
3792 if (sg_alloc_table(sgt, sgt->orig_nents, GFP_KERNEL)) {
3793 dev_err(adapter->pdev_dev, "HMA SGL allocation failed\n");
3794 kfree(adapter->hma.sgt);
3795 adapter->hma.sgt = NULL;
3799 sgl = adapter->hma.sgt->sgl;
3800 node = dev_to_node(adapter->pdev_dev);
3801 for_each_sg(sgl, iter, sgt->orig_nents, i) {
3802 newpage = alloc_pages_node(node, __GFP_NOWARN | GFP_KERNEL |
3803 __GFP_ZERO, page_order);
3805 dev_err(adapter->pdev_dev,
3806 "Not enough memory for HMA page allocation\n");
3810 sg_set_page(iter, newpage, page_size << page_order, 0);
3813 sgt->nents = dma_map_sg(adapter->pdev_dev, sgl, sgt->orig_nents,
3816 dev_err(adapter->pdev_dev,
3817 "Not enough memory for HMA DMA mapping");
3821 adapter->hma.flags |= HMA_DMA_MAPPED_FLAG;
3823 adapter->hma.phy_addr = kcalloc(sgt->nents, sizeof(dma_addr_t),
3825 if (unlikely(!adapter->hma.phy_addr))
3828 for_each_sg(sgl, iter, sgt->nents, i) {
3829 newpage = sg_page(iter);
3830 adapter->hma.phy_addr[i] = sg_dma_address(iter);
3833 ncmds = DIV_ROUND_UP(sgt->nents, HMA_MAX_ADDR_IN_CMD);
3834 /* Pass on the addresses to firmware */
3835 for (i = 0, k = 0; i < ncmds; i++, k += HMA_MAX_ADDR_IN_CMD) {
3836 struct fw_hma_cmd hma_cmd;
3837 u8 naddr = HMA_MAX_ADDR_IN_CMD;
3838 u8 soc = 0, eoc = 0;
3839 u8 hma_mode = 1; /* Presently we support only Page table mode */
3841 soc = (i == 0) ? 1 : 0;
3842 eoc = (i == ncmds - 1) ? 1 : 0;
3844 /* For last cmd, set naddr corresponding to remaining
3847 if (i == ncmds - 1) {
3848 naddr = sgt->nents % HMA_MAX_ADDR_IN_CMD;
3849 naddr = naddr ? naddr : HMA_MAX_ADDR_IN_CMD;
3851 memset(&hma_cmd, 0, sizeof(hma_cmd));
3852 hma_cmd.op_pkd = htonl(FW_CMD_OP_V(FW_HMA_CMD) |
3853 FW_CMD_REQUEST_F | FW_CMD_WRITE_F);
3854 hma_cmd.retval_len16 = htonl(FW_LEN16(hma_cmd));
3856 hma_cmd.mode_to_pcie_params =
3857 htonl(FW_HMA_CMD_MODE_V(hma_mode) |
3858 FW_HMA_CMD_SOC_V(soc) | FW_HMA_CMD_EOC_V(eoc));
3860 /* HMA cmd size specified in MB's */
3861 hma_cmd.naddr_size =
3862 htonl(FW_HMA_CMD_SIZE_V(hma_size) |
3863 FW_HMA_CMD_NADDR_V(naddr));
3865 /* Total Page size specified in units of 4K */
3866 hma_cmd.addr_size_pkd =
3867 htonl(FW_HMA_CMD_ADDR_SIZE_V
3868 ((page_size << page_order) >> 12));
3870 /* Fill the 5 addresses */
3871 for (j = 0; j < naddr; j++) {
3872 hma_cmd.phy_address[j] =
3873 cpu_to_be64(adapter->hma.phy_addr[j + k]);
3875 ret = t4_wr_mbox(adapter, adapter->mbox, &hma_cmd,
3876 sizeof(hma_cmd), &hma_cmd);
3878 dev_err(adapter->pdev_dev,
3879 "HMA FW command failed with err %d\n", ret);
3885 dev_info(adapter->pdev_dev,
3886 "Reserved %uMB host memory for HMA\n", hma_size);
3890 adap_free_hma_mem(adapter);
3894 static int adap_init1(struct adapter *adap, struct fw_caps_config_cmd *c)
3899 /* Now that we've successfully configured and initialized the adapter
3900 * can ask the Firmware what resources it has provisioned for us.
3902 ret = t4_get_pfres(adap);
3904 dev_err(adap->pdev_dev,
3905 "Unable to retrieve resource provisioning information\n");
3909 /* get device capabilities */
3910 memset(c, 0, sizeof(*c));
3911 c->op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3912 FW_CMD_REQUEST_F | FW_CMD_READ_F);
3913 c->cfvalid_to_len16 = htonl(FW_LEN16(*c));
3914 ret = t4_wr_mbox(adap, adap->mbox, c, sizeof(*c), c);
3918 c->op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3919 FW_CMD_REQUEST_F | FW_CMD_WRITE_F);
3920 ret = t4_wr_mbox(adap, adap->mbox, c, sizeof(*c), NULL);
3924 ret = t4_config_glbl_rss(adap, adap->pf,
3925 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL,
3926 FW_RSS_GLB_CONFIG_CMD_TNLMAPEN_F |
3927 FW_RSS_GLB_CONFIG_CMD_TNLALLLKP_F);
3931 ret = t4_cfg_pfvf(adap, adap->mbox, adap->pf, 0, adap->sge.egr_sz, 64,
3932 MAX_INGQ, 0, 0, 4, 0xf, 0xf, 16, FW_CMD_CAP_PF,
3939 /* tweak some settings */
3940 t4_write_reg(adap, TP_SHIFT_CNT_A, 0x64f8849);
3941 t4_write_reg(adap, ULP_RX_TDDP_PSZ_A, HPZ0_V(PAGE_SHIFT - 12));
3942 t4_write_reg(adap, TP_PIO_ADDR_A, TP_INGRESS_CONFIG_A);
3943 v = t4_read_reg(adap, TP_PIO_DATA_A);
3944 t4_write_reg(adap, TP_PIO_DATA_A, v & ~CSUM_HAS_PSEUDO_HDR_F);
3946 /* first 4 Tx modulation queues point to consecutive Tx channels */
3947 adap->params.tp.tx_modq_map = 0xE4;
3948 t4_write_reg(adap, TP_TX_MOD_QUEUE_REQ_MAP_A,
3949 TX_MOD_QUEUE_REQ_MAP_V(adap->params.tp.tx_modq_map));
3951 /* associate each Tx modulation queue with consecutive Tx channels */
3953 t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
3954 &v, 1, TP_TX_SCHED_HDR_A);
3955 t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
3956 &v, 1, TP_TX_SCHED_FIFO_A);
3957 t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
3958 &v, 1, TP_TX_SCHED_PCMD_A);
3960 #define T4_TX_MODQ_10G_WEIGHT_DEFAULT 16 /* in KB units */
3961 if (is_offload(adap)) {
3962 t4_write_reg(adap, TP_TX_MOD_QUEUE_WEIGHT0_A,
3963 TX_MODQ_WEIGHT0_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3964 TX_MODQ_WEIGHT1_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3965 TX_MODQ_WEIGHT2_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3966 TX_MODQ_WEIGHT3_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
3967 t4_write_reg(adap, TP_TX_MOD_CHANNEL_WEIGHT_A,
3968 TX_MODQ_WEIGHT0_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3969 TX_MODQ_WEIGHT1_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3970 TX_MODQ_WEIGHT2_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3971 TX_MODQ_WEIGHT3_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
3974 /* get basic stuff going */
3975 return t4_early_init(adap, adap->pf);
3979 * Max # of ATIDs. The absolute HW max is 16K but we keep it lower.
3981 #define MAX_ATIDS 8192U
3984 * Phase 0 of initialization: contact FW, obtain config, perform basic init.
3986 * If the firmware we're dealing with has Configuration File support, then
3987 * we use that to perform all configuration
3991 * Tweak configuration based on module parameters, etc. Most of these have
3992 * defaults assigned to them by Firmware Configuration Files (if we're using
3993 * them) but need to be explicitly set if we're using hard-coded
3994 * initialization. But even in the case of using Firmware Configuration
3995 * Files, we'd like to expose the ability to change these via module
3996 * parameters so these are essentially common tweaks/settings for
3997 * Configuration Files and hard-coded initialization ...
3999 static int adap_init0_tweaks(struct adapter *adapter)
4002 * Fix up various Host-Dependent Parameters like Page Size, Cache
4003 * Line Size, etc. The firmware default is for a 4KB Page Size and
4004 * 64B Cache Line Size ...
4006 t4_fixup_host_params(adapter, PAGE_SIZE, L1_CACHE_BYTES);
4009 * Process module parameters which affect early initialization.
4011 if (rx_dma_offset != 2 && rx_dma_offset != 0) {
4012 dev_err(&adapter->pdev->dev,
4013 "Ignoring illegal rx_dma_offset=%d, using 2\n",
4017 t4_set_reg_field(adapter, SGE_CONTROL_A,
4018 PKTSHIFT_V(PKTSHIFT_M),
4019 PKTSHIFT_V(rx_dma_offset));
4022 * Don't include the "IP Pseudo Header" in CPL_RX_PKT checksums: Linux
4023 * adds the pseudo header itself.
4025 t4_tp_wr_bits_indirect(adapter, TP_INGRESS_CONFIG_A,
4026 CSUM_HAS_PSEUDO_HDR_F, 0);
4031 /* 10Gb/s-BT PHY Support. chip-external 10Gb/s-BT PHYs are complex chips
4032 * unto themselves and they contain their own firmware to perform their
4035 static int phy_aq1202_version(const u8 *phy_fw_data,
4040 /* At offset 0x8 you're looking for the primary image's
4041 * starting offset which is 3 Bytes wide
4043 * At offset 0xa of the primary image, you look for the offset
4044 * of the DRAM segment which is 3 Bytes wide.
4046 * The FW version is at offset 0x27e of the DRAM and is 2 Bytes
4049 #define be16(__p) (((__p)[0] << 8) | (__p)[1])
4050 #define le16(__p) ((__p)[0] | ((__p)[1] << 8))
4051 #define le24(__p) (le16(__p) | ((__p)[2] << 16))
4053 offset = le24(phy_fw_data + 0x8) << 12;
4054 offset = le24(phy_fw_data + offset + 0xa);
4055 return be16(phy_fw_data + offset + 0x27e);
4062 static struct info_10gbt_phy_fw {
4063 unsigned int phy_fw_id; /* PCI Device ID */
4064 char *phy_fw_file; /* /lib/firmware/ PHY Firmware file */
4065 int (*phy_fw_version)(const u8 *phy_fw_data, size_t phy_fw_size);
4066 int phy_flash; /* Has FLASH for PHY Firmware */
4067 } phy_info_array[] = {
4069 PHY_AQ1202_DEVICEID,
4070 PHY_AQ1202_FIRMWARE,
4075 PHY_BCM84834_DEVICEID,
4076 PHY_BCM84834_FIRMWARE,
4083 static struct info_10gbt_phy_fw *find_phy_info(int devid)
4087 for (i = 0; i < ARRAY_SIZE(phy_info_array); i++) {
4088 if (phy_info_array[i].phy_fw_id == devid)
4089 return &phy_info_array[i];
4094 /* Handle updating of chip-external 10Gb/s-BT PHY firmware. This needs to
4095 * happen after the FW_RESET_CMD but before the FW_INITIALIZE_CMD. On error
4096 * we return a negative error number. If we transfer new firmware we return 1
4097 * (from t4_load_phy_fw()). If we don't do anything we return 0.
4099 static int adap_init0_phy(struct adapter *adap)
4101 const struct firmware *phyf;
4103 struct info_10gbt_phy_fw *phy_info;
4105 /* Use the device ID to determine which PHY file to flash.
4107 phy_info = find_phy_info(adap->pdev->device);
4109 dev_warn(adap->pdev_dev,
4110 "No PHY Firmware file found for this PHY\n");
4114 /* If we have a T4 PHY firmware file under /lib/firmware/cxgb4/, then
4115 * use that. The adapter firmware provides us with a memory buffer
4116 * where we can load a PHY firmware file from the host if we want to
4117 * override the PHY firmware File in flash.
4119 ret = request_firmware_direct(&phyf, phy_info->phy_fw_file,
4122 /* For adapters without FLASH attached to PHY for their
4123 * firmware, it's obviously a fatal error if we can't get the
4124 * firmware to the adapter. For adapters with PHY firmware
4125 * FLASH storage, it's worth a warning if we can't find the
4126 * PHY Firmware but we'll neuter the error ...
4128 dev_err(adap->pdev_dev, "unable to find PHY Firmware image "
4129 "/lib/firmware/%s, error %d\n",
4130 phy_info->phy_fw_file, -ret);
4131 if (phy_info->phy_flash) {
4132 int cur_phy_fw_ver = 0;
4134 t4_phy_fw_ver(adap, &cur_phy_fw_ver);
4135 dev_warn(adap->pdev_dev, "continuing with, on-adapter "
4136 "FLASH copy, version %#x\n", cur_phy_fw_ver);
4143 /* Load PHY Firmware onto adapter.
4145 ret = t4_load_phy_fw(adap, MEMWIN_NIC, &adap->win0_lock,
4146 phy_info->phy_fw_version,
4147 (u8 *)phyf->data, phyf->size);
4149 dev_err(adap->pdev_dev, "PHY Firmware transfer error %d\n",
4152 int new_phy_fw_ver = 0;
4154 if (phy_info->phy_fw_version)
4155 new_phy_fw_ver = phy_info->phy_fw_version(phyf->data,
4157 dev_info(adap->pdev_dev, "Successfully transferred PHY "
4158 "Firmware /lib/firmware/%s, version %#x\n",
4159 phy_info->phy_fw_file, new_phy_fw_ver);
4162 release_firmware(phyf);
4168 * Attempt to initialize the adapter via a Firmware Configuration File.
4170 static int adap_init0_config(struct adapter *adapter, int reset)
4172 char *fw_config_file, fw_config_file_path[256];
4173 u32 finiver, finicsum, cfcsum, param, val;
4174 struct fw_caps_config_cmd caps_cmd;
4175 unsigned long mtype = 0, maddr = 0;
4176 const struct firmware *cf;
4177 char *config_name = NULL;
4178 int config_issued = 0;
4182 * Reset device if necessary.
4185 ret = t4_fw_reset(adapter, adapter->mbox,
4186 PIORSTMODE_F | PIORST_F);
4191 /* If this is a 10Gb/s-BT adapter make sure the chip-external
4192 * 10Gb/s-BT PHYs have up-to-date firmware. Note that this step needs
4193 * to be performed after any global adapter RESET above since some
4194 * PHYs only have local RAM copies of the PHY firmware.
4196 if (is_10gbt_device(adapter->pdev->device)) {
4197 ret = adap_init0_phy(adapter);
4202 * If we have a T4 configuration file under /lib/firmware/cxgb4/,
4203 * then use that. Otherwise, use the configuration file stored
4204 * in the adapter flash ...
4206 switch (CHELSIO_CHIP_VERSION(adapter->params.chip)) {
4208 fw_config_file = FW4_CFNAME;
4211 fw_config_file = FW5_CFNAME;
4214 fw_config_file = FW6_CFNAME;
4217 dev_err(adapter->pdev_dev, "Device %d is not supported\n",
4218 adapter->pdev->device);
4223 ret = request_firmware(&cf, fw_config_file, adapter->pdev_dev);
4225 config_name = "On FLASH";
4226 mtype = FW_MEMTYPE_CF_FLASH;
4227 maddr = t4_flash_cfg_addr(adapter);
4229 u32 params[7], val[7];
4231 sprintf(fw_config_file_path,
4232 "/lib/firmware/%s", fw_config_file);
4233 config_name = fw_config_file_path;
4235 if (cf->size >= FLASH_CFG_MAX_SIZE)
4238 params[0] = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
4239 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_CF));
4240 ret = t4_query_params(adapter, adapter->mbox,
4241 adapter->pf, 0, 1, params, val);
4244 * For t4_memory_rw() below addresses and
4245 * sizes have to be in terms of multiples of 4
4246 * bytes. So, if the Configuration File isn't
4247 * a multiple of 4 bytes in length we'll have
4248 * to write that out separately since we can't
4249 * guarantee that the bytes following the
4250 * residual byte in the buffer returned by
4251 * request_firmware() are zeroed out ...
4253 size_t resid = cf->size & 0x3;
4254 size_t size = cf->size & ~0x3;
4255 __be32 *data = (__be32 *)cf->data;
4257 mtype = FW_PARAMS_PARAM_Y_G(val[0]);
4258 maddr = FW_PARAMS_PARAM_Z_G(val[0]) << 16;
4260 spin_lock(&adapter->win0_lock);
4261 ret = t4_memory_rw(adapter, 0, mtype, maddr,
4262 size, data, T4_MEMORY_WRITE);
4263 if (ret == 0 && resid != 0) {
4270 last.word = data[size >> 2];
4271 for (i = resid; i < 4; i++)
4273 ret = t4_memory_rw(adapter, 0, mtype,
4278 spin_unlock(&adapter->win0_lock);
4282 release_firmware(cf);
4289 /* Ofld + Hash filter is supported. Older fw will fail this request and
4292 param = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
4293 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_HASHFILTER_WITH_OFLD));
4294 ret = t4_set_params(adapter, adapter->mbox, adapter->pf, 0,
4297 /* FW doesn't know about Hash filter + ofld support,
4298 * it's not a problem, don't return an error.
4301 dev_warn(adapter->pdev_dev,
4302 "Hash filter with ofld is not supported by FW\n");
4306 * Issue a Capability Configuration command to the firmware to get it
4307 * to parse the Configuration File. We don't use t4_fw_config_file()
4308 * because we want the ability to modify various features after we've
4309 * processed the configuration file ...
4311 memset(&caps_cmd, 0, sizeof(caps_cmd));
4312 caps_cmd.op_to_write =
4313 htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
4316 caps_cmd.cfvalid_to_len16 =
4317 htonl(FW_CAPS_CONFIG_CMD_CFVALID_F |
4318 FW_CAPS_CONFIG_CMD_MEMTYPE_CF_V(mtype) |
4319 FW_CAPS_CONFIG_CMD_MEMADDR64K_CF_V(maddr >> 16) |
4320 FW_LEN16(caps_cmd));
4321 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
4324 /* If the CAPS_CONFIG failed with an ENOENT (for a Firmware
4325 * Configuration File in FLASH), our last gasp effort is to use the
4326 * Firmware Configuration File which is embedded in the firmware. A
4327 * very few early versions of the firmware didn't have one embedded
4328 * but we can ignore those.
4330 if (ret == -ENOENT) {
4331 memset(&caps_cmd, 0, sizeof(caps_cmd));
4332 caps_cmd.op_to_write =
4333 htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
4336 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
4337 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd,
4338 sizeof(caps_cmd), &caps_cmd);
4339 config_name = "Firmware Default";
4346 finiver = ntohl(caps_cmd.finiver);
4347 finicsum = ntohl(caps_cmd.finicsum);
4348 cfcsum = ntohl(caps_cmd.cfcsum);
4349 if (finicsum != cfcsum)
4350 dev_warn(adapter->pdev_dev, "Configuration File checksum "\
4351 "mismatch: [fini] csum=%#x, computed csum=%#x\n",
4355 * And now tell the firmware to use the configuration we just loaded.
4357 caps_cmd.op_to_write =
4358 htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
4361 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
4362 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
4368 * Tweak configuration based on system architecture, module
4371 ret = adap_init0_tweaks(adapter);
4375 /* We will proceed even if HMA init fails. */
4376 ret = adap_config_hma(adapter);
4378 dev_err(adapter->pdev_dev,
4379 "HMA configuration failed with error %d\n", ret);
4381 if (is_t6(adapter->params.chip)) {
4382 adap_config_hpfilter(adapter);
4383 ret = setup_ppod_edram(adapter);
4385 dev_info(adapter->pdev_dev, "Successfully enabled "
4386 "ppod edram feature\n");
4390 * And finally tell the firmware to initialize itself using the
4391 * parameters from the Configuration File.
4393 ret = t4_fw_initialize(adapter, adapter->mbox);
4397 /* Emit Firmware Configuration File information and return
4400 dev_info(adapter->pdev_dev, "Successfully configured using Firmware "\
4401 "Configuration File \"%s\", version %#x, computed checksum %#x\n",
4402 config_name, finiver, cfcsum);
4406 * Something bad happened. Return the error ... (If the "error"
4407 * is that there's no Configuration File on the adapter we don't
4408 * want to issue a warning since this is fairly common.)
4411 if (config_issued && ret != -ENOENT)
4412 dev_warn(adapter->pdev_dev, "\"%s\" configuration file error %d\n",
4417 static struct fw_info fw_info_array[] = {
4420 .fs_name = FW4_CFNAME,
4421 .fw_mod_name = FW4_FNAME,
4423 .chip = FW_HDR_CHIP_T4,
4424 .fw_ver = __cpu_to_be32(FW_VERSION(T4)),
4425 .intfver_nic = FW_INTFVER(T4, NIC),
4426 .intfver_vnic = FW_INTFVER(T4, VNIC),
4427 .intfver_ri = FW_INTFVER(T4, RI),
4428 .intfver_iscsi = FW_INTFVER(T4, ISCSI),
4429 .intfver_fcoe = FW_INTFVER(T4, FCOE),
4433 .fs_name = FW5_CFNAME,
4434 .fw_mod_name = FW5_FNAME,
4436 .chip = FW_HDR_CHIP_T5,
4437 .fw_ver = __cpu_to_be32(FW_VERSION(T5)),
4438 .intfver_nic = FW_INTFVER(T5, NIC),
4439 .intfver_vnic = FW_INTFVER(T5, VNIC),
4440 .intfver_ri = FW_INTFVER(T5, RI),
4441 .intfver_iscsi = FW_INTFVER(T5, ISCSI),
4442 .intfver_fcoe = FW_INTFVER(T5, FCOE),
4446 .fs_name = FW6_CFNAME,
4447 .fw_mod_name = FW6_FNAME,
4449 .chip = FW_HDR_CHIP_T6,
4450 .fw_ver = __cpu_to_be32(FW_VERSION(T6)),
4451 .intfver_nic = FW_INTFVER(T6, NIC),
4452 .intfver_vnic = FW_INTFVER(T6, VNIC),
4453 .intfver_ofld = FW_INTFVER(T6, OFLD),
4454 .intfver_ri = FW_INTFVER(T6, RI),
4455 .intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
4456 .intfver_iscsi = FW_INTFVER(T6, ISCSI),
4457 .intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
4458 .intfver_fcoe = FW_INTFVER(T6, FCOE),
4464 static struct fw_info *find_fw_info(int chip)
4468 for (i = 0; i < ARRAY_SIZE(fw_info_array); i++) {
4469 if (fw_info_array[i].chip == chip)
4470 return &fw_info_array[i];
4476 * Phase 0 of initialization: contact FW, obtain config, perform basic init.
4478 static int adap_init0(struct adapter *adap, int vpd_skip)
4480 struct fw_caps_config_cmd caps_cmd;
4481 u32 params[7], val[7];
4482 enum dev_state state;
4487 /* Grab Firmware Device Log parameters as early as possible so we have
4488 * access to it for debugging, etc.
4490 ret = t4_init_devlog_params(adap);
4494 /* Contact FW, advertising Master capability */
4495 ret = t4_fw_hello(adap, adap->mbox, adap->mbox,
4496 is_kdump_kernel() ? MASTER_MUST : MASTER_MAY, &state);
4498 dev_err(adap->pdev_dev, "could not connect to FW, error %d\n",
4502 if (ret == adap->mbox)
4503 adap->flags |= CXGB4_MASTER_PF;
4506 * If we're the Master PF Driver and the device is uninitialized,
4507 * then let's consider upgrading the firmware ... (We always want
4508 * to check the firmware version number in order to A. get it for
4509 * later reporting and B. to warn if the currently loaded firmware
4510 * is excessively mismatched relative to the driver.)
4513 t4_get_version_info(adap);
4514 ret = t4_check_fw_version(adap);
4515 /* If firmware is too old (not supported by driver) force an update. */
4517 state = DEV_STATE_UNINIT;
4518 if ((adap->flags & CXGB4_MASTER_PF) && state != DEV_STATE_INIT) {
4519 struct fw_info *fw_info;
4520 struct fw_hdr *card_fw;
4521 const struct firmware *fw;
4522 const u8 *fw_data = NULL;
4523 unsigned int fw_size = 0;
4525 /* This is the firmware whose headers the driver was compiled
4528 fw_info = find_fw_info(CHELSIO_CHIP_VERSION(adap->params.chip));
4529 if (fw_info == NULL) {
4530 dev_err(adap->pdev_dev,
4531 "unable to get firmware info for chip %d.\n",
4532 CHELSIO_CHIP_VERSION(adap->params.chip));
4536 /* allocate memory to read the header of the firmware on the
4539 card_fw = kvzalloc(sizeof(*card_fw), GFP_KERNEL);
4545 /* Get FW from from /lib/firmware/ */
4546 ret = request_firmware(&fw, fw_info->fw_mod_name,
4549 dev_err(adap->pdev_dev,
4550 "unable to load firmware image %s, error %d\n",
4551 fw_info->fw_mod_name, ret);
4557 /* upgrade FW logic */
4558 ret = t4_prep_fw(adap, fw_info, fw_data, fw_size, card_fw,
4562 release_firmware(fw);
4569 /* If the firmware is initialized already, emit a simply note to that
4570 * effect. Otherwise, it's time to try initializing the adapter.
4572 if (state == DEV_STATE_INIT) {
4573 ret = adap_config_hma(adap);
4575 dev_err(adap->pdev_dev,
4576 "HMA configuration failed with error %d\n",
4578 dev_info(adap->pdev_dev, "Coming up as %s: "\
4579 "Adapter already initialized\n",
4580 adap->flags & CXGB4_MASTER_PF ? "MASTER" : "SLAVE");
4582 dev_info(adap->pdev_dev, "Coming up as MASTER: "\
4583 "Initializing adapter\n");
4585 /* Find out whether we're dealing with a version of the
4586 * firmware which has configuration file support.
4588 params[0] = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
4589 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_CF));
4590 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1,
4593 /* If the firmware doesn't support Configuration Files,
4597 dev_err(adap->pdev_dev, "firmware doesn't support "
4598 "Firmware Configuration Files\n");
4602 /* The firmware provides us with a memory buffer where we can
4603 * load a Configuration File from the host if we want to
4604 * override the Configuration File in flash.
4606 ret = adap_init0_config(adap, reset);
4607 if (ret == -ENOENT) {
4608 dev_err(adap->pdev_dev, "no Configuration File "
4609 "present on adapter.\n");
4613 dev_err(adap->pdev_dev, "could not initialize "
4614 "adapter, error %d\n", -ret);
4619 /* Now that we've successfully configured and initialized the adapter
4620 * (or found it already initialized), we can ask the Firmware what
4621 * resources it has provisioned for us.
4623 ret = t4_get_pfres(adap);
4625 dev_err(adap->pdev_dev,
4626 "Unable to retrieve resource provisioning information\n");
4630 /* Grab VPD parameters. This should be done after we establish a
4631 * connection to the firmware since some of the VPD parameters
4632 * (notably the Core Clock frequency) are retrieved via requests to
4633 * the firmware. On the other hand, we need these fairly early on
4634 * so we do this right after getting ahold of the firmware.
4636 * We need to do this after initializing the adapter because someone
4637 * could have FLASHed a new VPD which won't be read by the firmware
4638 * until we do the RESET ...
4641 ret = t4_get_vpd_params(adap, &adap->params.vpd);
4646 /* Find out what ports are available to us. Note that we need to do
4647 * this before calling adap_init0_no_config() since it needs nports
4651 FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
4652 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_PORTVEC);
4653 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, &v, &port_vec);
4657 adap->params.nports = hweight32(port_vec);
4658 adap->params.portvec = port_vec;
4660 /* Give the SGE code a chance to pull in anything that it needs ...
4661 * Note that this must be called after we retrieve our VPD parameters
4662 * in order to know how to convert core ticks to seconds, etc.
4664 ret = t4_sge_init(adap);
4668 /* Grab the SGE Doorbell Queue Timer values. If successful, that
4669 * indicates that the Firmware and Hardware support this.
4671 params[0] = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
4672 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_DBQ_TIMERTICK));
4673 ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
4677 adap->sge.dbqtimer_tick = val[0];
4678 ret = t4_read_sge_dbqtimers(adap,
4679 ARRAY_SIZE(adap->sge.dbqtimer_val),
4680 adap->sge.dbqtimer_val);
4684 adap->flags |= CXGB4_SGE_DBQ_TIMER;
4686 if (is_bypass_device(adap->pdev->device))
4687 adap->params.bypass = 1;
4690 * Grab some of our basic fundamental operating parameters.
4692 params[0] = FW_PARAM_PFVF(EQ_START);
4693 params[1] = FW_PARAM_PFVF(L2T_START);
4694 params[2] = FW_PARAM_PFVF(L2T_END);
4695 params[3] = FW_PARAM_PFVF(FILTER_START);
4696 params[4] = FW_PARAM_PFVF(FILTER_END);
4697 params[5] = FW_PARAM_PFVF(IQFLINT_START);
4698 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6, params, val);
4701 adap->sge.egr_start = val[0];
4702 adap->l2t_start = val[1];
4703 adap->l2t_end = val[2];
4704 adap->tids.ftid_base = val[3];
4705 adap->tids.nftids = val[4] - val[3] + 1;
4706 adap->sge.ingr_start = val[5];
4708 if (CHELSIO_CHIP_VERSION(adap->params.chip) > CHELSIO_T5) {
4709 params[0] = FW_PARAM_PFVF(HPFILTER_START);
4710 params[1] = FW_PARAM_PFVF(HPFILTER_END);
4711 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
4716 adap->tids.hpftid_base = val[0];
4717 adap->tids.nhpftids = val[1] - val[0] + 1;
4719 /* Read the raw mps entries. In T6, the last 2 tcam entries
4720 * are reserved for raw mac addresses (rawf = 2, one per port).
4722 params[0] = FW_PARAM_PFVF(RAWF_START);
4723 params[1] = FW_PARAM_PFVF(RAWF_END);
4724 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
4727 adap->rawf_start = val[0];
4728 adap->rawf_cnt = val[1] - val[0] + 1;
4731 adap->tids.tid_base =
4732 t4_read_reg(adap, LE_DB_ACTIVE_TABLE_START_INDEX_A);
4735 /* qids (ingress/egress) returned from firmware can be anywhere
4736 * in the range from EQ(IQFLINT)_START to EQ(IQFLINT)_END.
4737 * Hence driver needs to allocate memory for this range to
4738 * store the queue info. Get the highest IQFLINT/EQ index returned
4739 * in FW_EQ_*_CMD.alloc command.
4741 params[0] = FW_PARAM_PFVF(EQ_END);
4742 params[1] = FW_PARAM_PFVF(IQFLINT_END);
4743 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
4746 adap->sge.egr_sz = val[0] - adap->sge.egr_start + 1;
4747 adap->sge.ingr_sz = val[1] - adap->sge.ingr_start + 1;
4749 adap->sge.egr_map = kcalloc(adap->sge.egr_sz,
4750 sizeof(*adap->sge.egr_map), GFP_KERNEL);
4751 if (!adap->sge.egr_map) {
4756 adap->sge.ingr_map = kcalloc(adap->sge.ingr_sz,
4757 sizeof(*adap->sge.ingr_map), GFP_KERNEL);
4758 if (!adap->sge.ingr_map) {
4763 /* Allocate the memory for the vaious egress queue bitmaps
4764 * ie starving_fl, txq_maperr and blocked_fl.
4766 adap->sge.starving_fl = kcalloc(BITS_TO_LONGS(adap->sge.egr_sz),
4767 sizeof(long), GFP_KERNEL);
4768 if (!adap->sge.starving_fl) {
4773 adap->sge.txq_maperr = kcalloc(BITS_TO_LONGS(adap->sge.egr_sz),
4774 sizeof(long), GFP_KERNEL);
4775 if (!adap->sge.txq_maperr) {
4780 #ifdef CONFIG_DEBUG_FS
4781 adap->sge.blocked_fl = kcalloc(BITS_TO_LONGS(adap->sge.egr_sz),
4782 sizeof(long), GFP_KERNEL);
4783 if (!adap->sge.blocked_fl) {
4789 params[0] = FW_PARAM_PFVF(CLIP_START);
4790 params[1] = FW_PARAM_PFVF(CLIP_END);
4791 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
4794 adap->clipt_start = val[0];
4795 adap->clipt_end = val[1];
4797 /* Get the supported number of traffic classes */
4798 params[0] = FW_PARAM_DEV(NUM_TM_CLASS);
4799 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, params, val);
4801 /* We couldn't retrieve the number of Traffic Classes
4802 * supported by the hardware/firmware. So we hard
4805 adap->params.nsched_cls = is_t4(adap->params.chip) ? 15 : 16;
4807 adap->params.nsched_cls = val[0];
4810 /* query params related to active filter region */
4811 params[0] = FW_PARAM_PFVF(ACTIVE_FILTER_START);
4812 params[1] = FW_PARAM_PFVF(ACTIVE_FILTER_END);
4813 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
4814 /* If Active filter size is set we enable establishing
4815 * offload connection through firmware work request
4817 if ((val[0] != val[1]) && (ret >= 0)) {
4818 adap->flags |= CXGB4_FW_OFLD_CONN;
4819 adap->tids.aftid_base = val[0];
4820 adap->tids.aftid_end = val[1];
4823 /* If we're running on newer firmware, let it know that we're
4824 * prepared to deal with encapsulated CPL messages. Older
4825 * firmware won't understand this and we'll just get
4826 * unencapsulated messages ...
4828 params[0] = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
4830 (void)t4_set_params(adap, adap->mbox, adap->pf, 0, 1, params, val);
4833 * Find out whether we're allowed to use the T5+ ULPTX MEMWRITE DSGL
4834 * capability. Earlier versions of the firmware didn't have the
4835 * ULPTX_MEMWRITE_DSGL so we'll interpret a query failure as no
4836 * permission to use ULPTX MEMWRITE DSGL.
4838 if (is_t4(adap->params.chip)) {
4839 adap->params.ulptx_memwrite_dsgl = false;
4841 params[0] = FW_PARAM_DEV(ULPTX_MEMWRITE_DSGL);
4842 ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
4844 adap->params.ulptx_memwrite_dsgl = (ret == 0 && val[0] != 0);
4847 /* See if FW supports FW_RI_FR_NSMR_TPTE_WR work request */
4848 params[0] = FW_PARAM_DEV(RI_FR_NSMR_TPTE_WR);
4849 ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
4851 adap->params.fr_nsmr_tpte_wr_support = (ret == 0 && val[0] != 0);
4853 /* See if FW supports FW_FILTER2 work request */
4854 if (is_t4(adap->params.chip)) {
4855 adap->params.filter2_wr_support = 0;
4857 params[0] = FW_PARAM_DEV(FILTER2_WR);
4858 ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
4860 adap->params.filter2_wr_support = (ret == 0 && val[0] != 0);
4863 /* Check if FW supports returning vin and smt index.
4864 * If this is not supported, driver will interpret
4865 * these values from viid.
4867 params[0] = FW_PARAM_DEV(OPAQUE_VIID_SMT_EXTN);
4868 ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
4870 adap->params.viid_smt_extn_support = (ret == 0 && val[0] != 0);
4873 * Get device capabilities so we can determine what resources we need
4876 memset(&caps_cmd, 0, sizeof(caps_cmd));
4877 caps_cmd.op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
4878 FW_CMD_REQUEST_F | FW_CMD_READ_F);
4879 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
4880 ret = t4_wr_mbox(adap, adap->mbox, &caps_cmd, sizeof(caps_cmd),
4885 /* hash filter has some mandatory register settings to be tested and for
4886 * that it needs to test whether offload is enabled or not, hence
4887 * checking and setting it here.
4889 if (caps_cmd.ofldcaps)
4890 adap->params.offload = 1;
4892 if (caps_cmd.ofldcaps ||
4893 (caps_cmd.niccaps & htons(FW_CAPS_CONFIG_NIC_HASHFILTER)) ||
4894 (caps_cmd.niccaps & htons(FW_CAPS_CONFIG_NIC_ETHOFLD))) {
4895 /* query offload-related parameters */
4896 params[0] = FW_PARAM_DEV(NTID);
4897 params[1] = FW_PARAM_PFVF(SERVER_START);
4898 params[2] = FW_PARAM_PFVF(SERVER_END);
4899 params[3] = FW_PARAM_PFVF(TDDP_START);
4900 params[4] = FW_PARAM_PFVF(TDDP_END);
4901 params[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
4902 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6,
4906 adap->tids.ntids = val[0];
4907 adap->tids.natids = min(adap->tids.ntids / 2, MAX_ATIDS);
4908 adap->tids.stid_base = val[1];
4909 adap->tids.nstids = val[2] - val[1] + 1;
4911 * Setup server filter region. Divide the available filter
4912 * region into two parts. Regular filters get 1/3rd and server
4913 * filters get 2/3rd part. This is only enabled if workarond
4915 * 1. For regular filters.
4916 * 2. Server filter: This are special filters which are used
4917 * to redirect SYN packets to offload queue.
4919 if (adap->flags & CXGB4_FW_OFLD_CONN && !is_bypass(adap)) {
4920 adap->tids.sftid_base = adap->tids.ftid_base +
4921 DIV_ROUND_UP(adap->tids.nftids, 3);
4922 adap->tids.nsftids = adap->tids.nftids -
4923 DIV_ROUND_UP(adap->tids.nftids, 3);
4924 adap->tids.nftids = adap->tids.sftid_base -
4925 adap->tids.ftid_base;
4927 adap->vres.ddp.start = val[3];
4928 adap->vres.ddp.size = val[4] - val[3] + 1;
4929 adap->params.ofldq_wr_cred = val[5];
4931 if (caps_cmd.niccaps & htons(FW_CAPS_CONFIG_NIC_HASHFILTER)) {
4932 init_hash_filter(adap);
4934 adap->num_ofld_uld += 1;
4937 if (caps_cmd.niccaps & htons(FW_CAPS_CONFIG_NIC_ETHOFLD)) {
4938 params[0] = FW_PARAM_PFVF(ETHOFLD_START);
4939 params[1] = FW_PARAM_PFVF(ETHOFLD_END);
4940 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
4943 adap->tids.eotid_base = val[0];
4944 adap->tids.neotids = min_t(u32, MAX_ATIDS,
4945 val[1] - val[0] + 1);
4946 adap->params.ethofld = 1;
4950 if (caps_cmd.rdmacaps) {
4951 params[0] = FW_PARAM_PFVF(STAG_START);
4952 params[1] = FW_PARAM_PFVF(STAG_END);
4953 params[2] = FW_PARAM_PFVF(RQ_START);
4954 params[3] = FW_PARAM_PFVF(RQ_END);
4955 params[4] = FW_PARAM_PFVF(PBL_START);
4956 params[5] = FW_PARAM_PFVF(PBL_END);
4957 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6,
4961 adap->vres.stag.start = val[0];
4962 adap->vres.stag.size = val[1] - val[0] + 1;
4963 adap->vres.rq.start = val[2];
4964 adap->vres.rq.size = val[3] - val[2] + 1;
4965 adap->vres.pbl.start = val[4];
4966 adap->vres.pbl.size = val[5] - val[4] + 1;
4968 params[0] = FW_PARAM_PFVF(SRQ_START);
4969 params[1] = FW_PARAM_PFVF(SRQ_END);
4970 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
4973 adap->vres.srq.start = val[0];
4974 adap->vres.srq.size = val[1] - val[0] + 1;
4976 if (adap->vres.srq.size) {
4977 adap->srq = t4_init_srq(adap->vres.srq.size);
4979 dev_warn(&adap->pdev->dev, "could not allocate SRQ, continuing\n");
4982 params[0] = FW_PARAM_PFVF(SQRQ_START);
4983 params[1] = FW_PARAM_PFVF(SQRQ_END);
4984 params[2] = FW_PARAM_PFVF(CQ_START);
4985 params[3] = FW_PARAM_PFVF(CQ_END);
4986 params[4] = FW_PARAM_PFVF(OCQ_START);
4987 params[5] = FW_PARAM_PFVF(OCQ_END);
4988 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6, params,
4992 adap->vres.qp.start = val[0];
4993 adap->vres.qp.size = val[1] - val[0] + 1;
4994 adap->vres.cq.start = val[2];
4995 adap->vres.cq.size = val[3] - val[2] + 1;
4996 adap->vres.ocq.start = val[4];
4997 adap->vres.ocq.size = val[5] - val[4] + 1;
4999 params[0] = FW_PARAM_DEV(MAXORDIRD_QP);
5000 params[1] = FW_PARAM_DEV(MAXIRD_ADAPTER);
5001 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params,
5004 adap->params.max_ordird_qp = 8;
5005 adap->params.max_ird_adapter = 32 * adap->tids.ntids;
5008 adap->params.max_ordird_qp = val[0];
5009 adap->params.max_ird_adapter = val[1];
5011 dev_info(adap->pdev_dev,
5012 "max_ordird_qp %d max_ird_adapter %d\n",
5013 adap->params.max_ordird_qp,
5014 adap->params.max_ird_adapter);
5016 /* Enable write_with_immediate if FW supports it */
5017 params[0] = FW_PARAM_DEV(RDMA_WRITE_WITH_IMM);
5018 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, params,
5020 adap->params.write_w_imm_support = (ret == 0 && val[0] != 0);
5022 /* Enable write_cmpl if FW supports it */
5023 params[0] = FW_PARAM_DEV(RI_WRITE_CMPL_WR);
5024 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, params,
5026 adap->params.write_cmpl_support = (ret == 0 && val[0] != 0);
5027 adap->num_ofld_uld += 2;
5029 if (caps_cmd.iscsicaps) {
5030 params[0] = FW_PARAM_PFVF(ISCSI_START);
5031 params[1] = FW_PARAM_PFVF(ISCSI_END);
5032 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
5036 adap->vres.iscsi.start = val[0];
5037 adap->vres.iscsi.size = val[1] - val[0] + 1;
5038 if (is_t6(adap->params.chip)) {
5039 params[0] = FW_PARAM_PFVF(PPOD_EDRAM_START);
5040 params[1] = FW_PARAM_PFVF(PPOD_EDRAM_END);
5041 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
5044 adap->vres.ppod_edram.start = val[0];
5045 adap->vres.ppod_edram.size =
5046 val[1] - val[0] + 1;
5048 dev_info(adap->pdev_dev,
5049 "ppod edram start 0x%x end 0x%x size 0x%x\n",
5051 adap->vres.ppod_edram.size);
5054 /* LIO target and cxgb4i initiaitor */
5055 adap->num_ofld_uld += 2;
5057 if (caps_cmd.cryptocaps) {
5058 if (ntohs(caps_cmd.cryptocaps) &
5059 FW_CAPS_CONFIG_CRYPTO_LOOKASIDE) {
5060 params[0] = FW_PARAM_PFVF(NCRYPTO_LOOKASIDE);
5061 ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
5067 adap->vres.ncrypto_fc = val[0];
5069 adap->num_ofld_uld += 1;
5071 if (ntohs(caps_cmd.cryptocaps) &
5072 FW_CAPS_CONFIG_TLS_INLINE) {
5073 params[0] = FW_PARAM_PFVF(TLS_START);
5074 params[1] = FW_PARAM_PFVF(TLS_END);
5075 ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
5079 adap->vres.key.start = val[0];
5080 adap->vres.key.size = val[1] - val[0] + 1;
5083 adap->params.crypto = ntohs(caps_cmd.cryptocaps);
5086 /* The MTU/MSS Table is initialized by now, so load their values. If
5087 * we're initializing the adapter, then we'll make any modifications
5088 * we want to the MTU/MSS Table and also initialize the congestion
5091 t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
5092 if (state != DEV_STATE_INIT) {
5095 /* The default MTU Table contains values 1492 and 1500.
5096 * However, for TCP, it's better to have two values which are
5097 * a multiple of 8 +/- 4 bytes apart near this popular MTU.
5098 * This allows us to have a TCP Data Payload which is a
5099 * multiple of 8 regardless of what combination of TCP Options
5100 * are in use (always a multiple of 4 bytes) which is
5101 * important for performance reasons. For instance, if no
5102 * options are in use, then we have a 20-byte IP header and a
5103 * 20-byte TCP header. In this case, a 1500-byte MSS would
5104 * result in a TCP Data Payload of 1500 - 40 == 1460 bytes
5105 * which is not a multiple of 8. So using an MSS of 1488 in
5106 * this case results in a TCP Data Payload of 1448 bytes which
5107 * is a multiple of 8. On the other hand, if 12-byte TCP Time
5108 * Stamps have been negotiated, then an MTU of 1500 bytes
5109 * results in a TCP Data Payload of 1448 bytes which, as
5110 * above, is a multiple of 8 bytes ...
5112 for (i = 0; i < NMTUS; i++)
5113 if (adap->params.mtus[i] == 1492) {
5114 adap->params.mtus[i] = 1488;
5118 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
5119 adap->params.b_wnd);
5121 t4_init_sge_params(adap);
5122 adap->flags |= CXGB4_FW_OK;
5123 t4_init_tp_params(adap, true);
5127 * Something bad happened. If a command timed out or failed with EIO
5128 * FW does not operate within its spec or something catastrophic
5129 * happened to HW/FW, stop issuing commands.
5132 adap_free_hma_mem(adap);
5133 kfree(adap->sge.egr_map);
5134 kfree(adap->sge.ingr_map);
5135 kfree(adap->sge.starving_fl);
5136 kfree(adap->sge.txq_maperr);
5137 #ifdef CONFIG_DEBUG_FS
5138 kfree(adap->sge.blocked_fl);
5140 if (ret != -ETIMEDOUT && ret != -EIO)
5141 t4_fw_bye(adap, adap->mbox);
5147 static pci_ers_result_t eeh_err_detected(struct pci_dev *pdev,
5148 pci_channel_state_t state)
5151 struct adapter *adap = pci_get_drvdata(pdev);
5157 adap->flags &= ~CXGB4_FW_OK;
5158 notify_ulds(adap, CXGB4_STATE_START_RECOVERY);
5159 spin_lock(&adap->stats_lock);
5160 for_each_port(adap, i) {
5161 struct net_device *dev = adap->port[i];
5163 netif_device_detach(dev);
5164 netif_carrier_off(dev);
5167 spin_unlock(&adap->stats_lock);
5168 disable_interrupts(adap);
5169 if (adap->flags & CXGB4_FULL_INIT_DONE)
5172 if ((adap->flags & CXGB4_DEV_ENABLED)) {
5173 pci_disable_device(pdev);
5174 adap->flags &= ~CXGB4_DEV_ENABLED;
5176 out: return state == pci_channel_io_perm_failure ?
5177 PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
5180 static pci_ers_result_t eeh_slot_reset(struct pci_dev *pdev)
5183 struct fw_caps_config_cmd c;
5184 struct adapter *adap = pci_get_drvdata(pdev);
5187 pci_restore_state(pdev);
5188 pci_save_state(pdev);
5189 return PCI_ERS_RESULT_RECOVERED;
5192 if (!(adap->flags & CXGB4_DEV_ENABLED)) {
5193 if (pci_enable_device(pdev)) {
5194 dev_err(&pdev->dev, "Cannot reenable PCI "
5195 "device after reset\n");
5196 return PCI_ERS_RESULT_DISCONNECT;
5198 adap->flags |= CXGB4_DEV_ENABLED;
5201 pci_set_master(pdev);
5202 pci_restore_state(pdev);
5203 pci_save_state(pdev);
5205 if (t4_wait_dev_ready(adap->regs) < 0)
5206 return PCI_ERS_RESULT_DISCONNECT;
5207 if (t4_fw_hello(adap, adap->mbox, adap->pf, MASTER_MUST, NULL) < 0)
5208 return PCI_ERS_RESULT_DISCONNECT;
5209 adap->flags |= CXGB4_FW_OK;
5210 if (adap_init1(adap, &c))
5211 return PCI_ERS_RESULT_DISCONNECT;
5213 for_each_port(adap, i) {
5214 struct port_info *pi = adap2pinfo(adap, i);
5215 u8 vivld = 0, vin = 0;
5217 ret = t4_alloc_vi(adap, adap->mbox, pi->tx_chan, adap->pf, 0, 1,
5218 NULL, NULL, &vivld, &vin);
5220 return PCI_ERS_RESULT_DISCONNECT;
5222 pi->xact_addr_filt = -1;
5223 /* If fw supports returning the VIN as part of FW_VI_CMD,
5224 * save the returned values.
5226 if (adap->params.viid_smt_extn_support) {
5230 /* Retrieve the values from VIID */
5231 pi->vivld = FW_VIID_VIVLD_G(pi->viid);
5232 pi->vin = FW_VIID_VIN_G(pi->viid);
5236 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
5237 adap->params.b_wnd);
5240 return PCI_ERS_RESULT_DISCONNECT;
5241 return PCI_ERS_RESULT_RECOVERED;
5244 static void eeh_resume(struct pci_dev *pdev)
5247 struct adapter *adap = pci_get_drvdata(pdev);
5253 for_each_port(adap, i) {
5254 struct net_device *dev = adap->port[i];
5256 if (netif_running(dev)) {
5258 cxgb_set_rxmode(dev);
5260 netif_device_attach(dev);
5266 static void eeh_reset_prepare(struct pci_dev *pdev)
5268 struct adapter *adapter = pci_get_drvdata(pdev);
5271 if (adapter->pf != 4)
5274 adapter->flags &= ~CXGB4_FW_OK;
5276 notify_ulds(adapter, CXGB4_STATE_DOWN);
5278 for_each_port(adapter, i)
5279 if (adapter->port[i]->reg_state == NETREG_REGISTERED)
5280 cxgb_close(adapter->port[i]);
5282 disable_interrupts(adapter);
5283 cxgb4_free_mps_ref_entries(adapter);
5285 adap_free_hma_mem(adapter);
5287 if (adapter->flags & CXGB4_FULL_INIT_DONE)
5291 static void eeh_reset_done(struct pci_dev *pdev)
5293 struct adapter *adapter = pci_get_drvdata(pdev);
5296 if (adapter->pf != 4)
5299 err = t4_wait_dev_ready(adapter->regs);
5301 dev_err(adapter->pdev_dev,
5302 "Device not ready, err %d", err);
5306 setup_memwin(adapter);
5308 err = adap_init0(adapter, 1);
5310 dev_err(adapter->pdev_dev,
5311 "Adapter init failed, err %d", err);
5315 setup_memwin_rdma(adapter);
5317 if (adapter->flags & CXGB4_FW_OK) {
5318 err = t4_port_init(adapter, adapter->pf, adapter->pf, 0);
5320 dev_err(adapter->pdev_dev,
5321 "Port init failed, err %d", err);
5326 err = cfg_queues(adapter);
5328 dev_err(adapter->pdev_dev,
5329 "Config queues failed, err %d", err);
5333 cxgb4_init_mps_ref_entries(adapter);
5335 err = setup_fw_sge_queues(adapter);
5337 dev_err(adapter->pdev_dev,
5338 "FW sge queue allocation failed, err %d", err);
5342 for_each_port(adapter, i)
5343 if (adapter->port[i]->reg_state == NETREG_REGISTERED)
5344 cxgb_open(adapter->port[i]);
5347 static const struct pci_error_handlers cxgb4_eeh = {
5348 .error_detected = eeh_err_detected,
5349 .slot_reset = eeh_slot_reset,
5350 .resume = eeh_resume,
5351 .reset_prepare = eeh_reset_prepare,
5352 .reset_done = eeh_reset_done,
5355 /* Return true if the Link Configuration supports "High Speeds" (those greater
5358 static inline bool is_x_10g_port(const struct link_config *lc)
5360 unsigned int speeds, high_speeds;
5362 speeds = FW_PORT_CAP32_SPEED_V(FW_PORT_CAP32_SPEED_G(lc->pcaps));
5363 high_speeds = speeds &
5364 ~(FW_PORT_CAP32_SPEED_100M | FW_PORT_CAP32_SPEED_1G);
5366 return high_speeds != 0;
5369 /* Perform default configuration of DMA queues depending on the number and type
5370 * of ports we found and the number of available CPUs. Most settings can be
5371 * modified by the admin prior to actual use.
5373 static int cfg_queues(struct adapter *adap)
5375 u32 avail_qsets, avail_eth_qsets, avail_uld_qsets;
5376 u32 i, n10g = 0, qidx = 0, n1g = 0;
5377 u32 ncpus = num_online_cpus();
5378 u32 niqflint, neq, num_ulds;
5379 struct sge *s = &adap->sge;
5382 /* Reduce memory usage in kdump environment, disable all offload. */
5383 if (is_kdump_kernel() || (is_uld(adap) && t4_uld_mem_alloc(adap))) {
5384 adap->params.offload = 0;
5385 adap->params.crypto = 0;
5386 adap->params.ethofld = 0;
5389 /* Calculate the number of Ethernet Queue Sets available based on
5390 * resources provisioned for us. We always have an Asynchronous
5391 * Firmware Event Ingress Queue. If we're operating in MSI or Legacy
5392 * IRQ Pin Interrupt mode, then we'll also have a Forwarded Interrupt
5393 * Ingress Queue. Meanwhile, we need two Egress Queues for each
5394 * Queue Set: one for the Free List and one for the Ethernet TX Queue.
5396 * Note that we should also take into account all of the various
5397 * Offload Queues. But, in any situation where we're operating in
5398 * a Resource Constrained Provisioning environment, doing any Offload
5399 * at all is problematic ...
5401 niqflint = adap->params.pfres.niqflint - 1;
5402 if (!(adap->flags & CXGB4_USING_MSIX))
5404 neq = adap->params.pfres.neq / 2;
5405 avail_qsets = min(niqflint, neq);
5407 if (avail_qsets < adap->params.nports) {
5408 dev_err(adap->pdev_dev, "avail_eth_qsets=%d < nports=%d\n",
5409 avail_qsets, adap->params.nports);
5413 /* Count the number of 10Gb/s or better ports */
5414 for_each_port(adap, i)
5415 n10g += is_x_10g_port(&adap2pinfo(adap, i)->link_cfg);
5417 avail_eth_qsets = min_t(u32, avail_qsets, MAX_ETH_QSETS);
5419 /* We default to 1 queue per non-10G port and up to # of cores queues
5423 q10g = (avail_eth_qsets - (adap->params.nports - n10g)) / n10g;
5425 n1g = adap->params.nports - n10g;
5426 #ifdef CONFIG_CHELSIO_T4_DCB
5427 /* For Data Center Bridging support we need to be able to support up
5428 * to 8 Traffic Priorities; each of which will be assigned to its
5429 * own TX Queue in order to prevent Head-Of-Line Blocking.
5432 if (adap->params.nports * 8 > avail_eth_qsets) {
5433 dev_err(adap->pdev_dev, "DCB avail_eth_qsets=%d < %d!\n",
5434 avail_eth_qsets, adap->params.nports * 8);
5438 if (adap->params.nports * ncpus < avail_eth_qsets)
5439 q10g = max(8U, ncpus);
5441 q10g = max(8U, q10g);
5443 while ((q10g * n10g) > (avail_eth_qsets - n1g * q1g))
5446 #else /* !CONFIG_CHELSIO_T4_DCB */
5448 q10g = min(q10g, ncpus);
5449 #endif /* !CONFIG_CHELSIO_T4_DCB */
5450 if (is_kdump_kernel()) {
5455 for_each_port(adap, i) {
5456 struct port_info *pi = adap2pinfo(adap, i);
5458 pi->first_qset = qidx;
5459 pi->nqsets = is_x_10g_port(&pi->link_cfg) ? q10g : q1g;
5464 s->max_ethqsets = qidx; /* MSI-X may lower it later */
5465 avail_qsets -= qidx;
5468 /* For offload we use 1 queue/channel if all ports are up to 1G,
5469 * otherwise we divide all available queues amongst the channels
5470 * capped by the number of available cores.
5472 num_ulds = adap->num_uld + adap->num_ofld_uld;
5473 i = min_t(u32, MAX_OFLD_QSETS, ncpus);
5474 avail_uld_qsets = roundup(i, adap->params.nports);
5475 if (avail_qsets < num_ulds * adap->params.nports) {
5476 adap->params.offload = 0;
5477 adap->params.crypto = 0;
5479 } else if (avail_qsets < num_ulds * avail_uld_qsets || !n10g) {
5480 s->ofldqsets = adap->params.nports;
5482 s->ofldqsets = avail_uld_qsets;
5485 avail_qsets -= num_ulds * s->ofldqsets;
5488 /* ETHOFLD Queues used for QoS offload should follow same
5489 * allocation scheme as normal Ethernet Queues.
5491 if (is_ethofld(adap)) {
5492 if (avail_qsets < s->max_ethqsets) {
5493 adap->params.ethofld = 0;
5496 s->eoqsets = s->max_ethqsets;
5498 avail_qsets -= s->eoqsets;
5501 for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
5502 struct sge_eth_rxq *r = &s->ethrxq[i];
5504 init_rspq(adap, &r->rspq, 5, 10, 1024, 64);
5508 for (i = 0; i < ARRAY_SIZE(s->ethtxq); i++)
5509 s->ethtxq[i].q.size = 1024;
5511 for (i = 0; i < ARRAY_SIZE(s->ctrlq); i++)
5512 s->ctrlq[i].q.size = 512;
5514 if (!is_t4(adap->params.chip))
5515 s->ptptxq.q.size = 8;
5517 init_rspq(adap, &s->fw_evtq, 0, 1, 1024, 64);
5518 init_rspq(adap, &s->intrq, 0, 1, 512, 64);
5524 * Reduce the number of Ethernet queues across all ports to at most n.
5525 * n provides at least one queue per port.
5527 static void reduce_ethqs(struct adapter *adap, int n)
5530 struct port_info *pi;
5532 while (n < adap->sge.ethqsets)
5533 for_each_port(adap, i) {
5534 pi = adap2pinfo(adap, i);
5535 if (pi->nqsets > 1) {
5537 adap->sge.ethqsets--;
5538 if (adap->sge.ethqsets <= n)
5544 for_each_port(adap, i) {
5545 pi = adap2pinfo(adap, i);
5551 static int alloc_msix_info(struct adapter *adap, u32 num_vec)
5553 struct msix_info *msix_info;
5555 msix_info = kcalloc(num_vec, sizeof(*msix_info), GFP_KERNEL);
5559 adap->msix_bmap.msix_bmap = kcalloc(BITS_TO_LONGS(num_vec),
5560 sizeof(long), GFP_KERNEL);
5561 if (!adap->msix_bmap.msix_bmap) {
5566 spin_lock_init(&adap->msix_bmap.lock);
5567 adap->msix_bmap.mapsize = num_vec;
5569 adap->msix_info = msix_info;
5573 static void free_msix_info(struct adapter *adap)
5575 kfree(adap->msix_bmap.msix_bmap);
5576 kfree(adap->msix_info);
5579 int cxgb4_get_msix_idx_from_bmap(struct adapter *adap)
5581 struct msix_bmap *bmap = &adap->msix_bmap;
5582 unsigned int msix_idx;
5583 unsigned long flags;
5585 spin_lock_irqsave(&bmap->lock, flags);
5586 msix_idx = find_first_zero_bit(bmap->msix_bmap, bmap->mapsize);
5587 if (msix_idx < bmap->mapsize) {
5588 __set_bit(msix_idx, bmap->msix_bmap);
5590 spin_unlock_irqrestore(&bmap->lock, flags);
5594 spin_unlock_irqrestore(&bmap->lock, flags);
5598 void cxgb4_free_msix_idx_in_bmap(struct adapter *adap,
5599 unsigned int msix_idx)
5601 struct msix_bmap *bmap = &adap->msix_bmap;
5602 unsigned long flags;
5604 spin_lock_irqsave(&bmap->lock, flags);
5605 __clear_bit(msix_idx, bmap->msix_bmap);
5606 spin_unlock_irqrestore(&bmap->lock, flags);
5609 /* 2 MSI-X vectors needed for the FW queue and non-data interrupts */
5610 #define EXTRA_VECS 2
5612 static int enable_msix(struct adapter *adap)
5614 u32 eth_need, uld_need = 0, ethofld_need = 0;
5615 u32 ethqsets = 0, ofldqsets = 0, eoqsets = 0;
5616 u8 num_uld = 0, nchan = adap->params.nports;
5617 u32 i, want, need, num_vec;
5618 struct sge *s = &adap->sge;
5619 struct msix_entry *entries;
5620 struct port_info *pi;
5623 want = s->max_ethqsets;
5624 #ifdef CONFIG_CHELSIO_T4_DCB
5625 /* For Data Center Bridging we need 8 Ethernet TX Priority Queues for
5634 num_uld = adap->num_ofld_uld + adap->num_uld;
5635 want += num_uld * s->ofldqsets;
5636 uld_need = num_uld * nchan;
5640 if (is_ethofld(adap)) {
5642 ethofld_need = eth_need;
5643 need += ethofld_need;
5649 entries = kmalloc_array(want, sizeof(*entries), GFP_KERNEL);
5653 for (i = 0; i < want; i++)
5654 entries[i].entry = i;
5656 allocated = pci_enable_msix_range(adap->pdev, entries, need, want);
5657 if (allocated < 0) {
5658 /* Disable offload and attempt to get vectors for NIC
5661 want = s->max_ethqsets + EXTRA_VECS;
5662 need = eth_need + EXTRA_VECS;
5663 allocated = pci_enable_msix_range(adap->pdev, entries,
5665 if (allocated < 0) {
5666 dev_info(adap->pdev_dev,
5667 "Disabling MSI-X due to insufficient MSI-X vectors\n");
5672 dev_info(adap->pdev_dev,
5673 "Disabling offload due to insufficient MSI-X vectors\n");
5674 adap->params.offload = 0;
5675 adap->params.crypto = 0;
5676 adap->params.ethofld = 0;
5683 num_vec = allocated;
5684 if (num_vec < want) {
5685 /* Distribute available vectors to the various queue groups.
5686 * Every group gets its minimum requirement and NIC gets top
5687 * priority for leftovers.
5689 ethqsets = eth_need;
5692 if (is_ethofld(adap))
5693 eoqsets = ethofld_need;
5697 if (num_vec < eth_need + ethofld_need ||
5698 ethqsets > s->max_ethqsets)
5701 for_each_port(adap, i) {
5702 pi = adap2pinfo(adap, i);
5717 if (num_vec < uld_need ||
5718 ofldqsets > s->ofldqsets)
5722 num_vec -= uld_need;
5726 ethqsets = s->max_ethqsets;
5728 ofldqsets = s->ofldqsets;
5729 if (is_ethofld(adap))
5730 eoqsets = s->eoqsets;
5733 if (ethqsets < s->max_ethqsets) {
5734 s->max_ethqsets = ethqsets;
5735 reduce_ethqs(adap, ethqsets);
5739 s->ofldqsets = ofldqsets;
5740 s->nqs_per_uld = s->ofldqsets;
5743 if (is_ethofld(adap))
5744 s->eoqsets = eoqsets;
5747 ret = alloc_msix_info(adap, allocated);
5749 goto out_disable_msix;
5751 for (i = 0; i < allocated; i++) {
5752 adap->msix_info[i].vec = entries[i].vector;
5753 adap->msix_info[i].idx = i;
5756 dev_info(adap->pdev_dev,
5757 "%d MSI-X vectors allocated, nic %d eoqsets %d per uld %d\n",
5758 allocated, s->max_ethqsets, s->eoqsets, s->nqs_per_uld);
5764 pci_disable_msix(adap->pdev);
5773 static int init_rss(struct adapter *adap)
5778 err = t4_init_rss_mode(adap, adap->mbox);
5782 for_each_port(adap, i) {
5783 struct port_info *pi = adap2pinfo(adap, i);
5785 pi->rss = kcalloc(pi->rss_size, sizeof(u16), GFP_KERNEL);
5792 /* Dump basic information about the adapter */
5793 static void print_adapter_info(struct adapter *adapter)
5795 /* Hardware/Firmware/etc. Version/Revision IDs */
5796 t4_dump_version_info(adapter);
5798 /* Software/Hardware configuration */
5799 dev_info(adapter->pdev_dev, "Configuration: %sNIC %s, %s capable\n",
5800 is_offload(adapter) ? "R" : "",
5801 ((adapter->flags & CXGB4_USING_MSIX) ? "MSI-X" :
5802 (adapter->flags & CXGB4_USING_MSI) ? "MSI" : ""),
5803 is_offload(adapter) ? "Offload" : "non-Offload");
5806 static void print_port_info(const struct net_device *dev)
5810 const struct port_info *pi = netdev_priv(dev);
5811 const struct adapter *adap = pi->adapter;
5813 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_100M)
5814 bufp += sprintf(bufp, "100M/");
5815 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_1G)
5816 bufp += sprintf(bufp, "1G/");
5817 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_10G)
5818 bufp += sprintf(bufp, "10G/");
5819 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_25G)
5820 bufp += sprintf(bufp, "25G/");
5821 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_40G)
5822 bufp += sprintf(bufp, "40G/");
5823 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_50G)
5824 bufp += sprintf(bufp, "50G/");
5825 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_100G)
5826 bufp += sprintf(bufp, "100G/");
5827 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_200G)
5828 bufp += sprintf(bufp, "200G/");
5829 if (pi->link_cfg.pcaps & FW_PORT_CAP32_SPEED_400G)
5830 bufp += sprintf(bufp, "400G/");
5833 sprintf(bufp, "BASE-%s", t4_get_port_type_description(pi->port_type));
5835 netdev_info(dev, "%s: Chelsio %s (%s) %s\n",
5836 dev->name, adap->params.vpd.id, adap->name, buf);
5840 * Free the following resources:
5841 * - memory used for tables
5844 * - resources FW is holding for us
5846 static void free_some_resources(struct adapter *adapter)
5850 kvfree(adapter->smt);
5851 kvfree(adapter->l2t);
5852 kvfree(adapter->srq);
5853 t4_cleanup_sched(adapter);
5854 kvfree(adapter->tids.tid_tab);
5855 cxgb4_cleanup_tc_matchall(adapter);
5856 cxgb4_cleanup_tc_mqprio(adapter);
5857 cxgb4_cleanup_tc_flower(adapter);
5858 cxgb4_cleanup_tc_u32(adapter);
5859 kfree(adapter->sge.egr_map);
5860 kfree(adapter->sge.ingr_map);
5861 kfree(adapter->sge.starving_fl);
5862 kfree(adapter->sge.txq_maperr);
5863 #ifdef CONFIG_DEBUG_FS
5864 kfree(adapter->sge.blocked_fl);
5866 disable_msi(adapter);
5868 for_each_port(adapter, i)
5869 if (adapter->port[i]) {
5870 struct port_info *pi = adap2pinfo(adapter, i);
5873 t4_free_vi(adapter, adapter->mbox, adapter->pf,
5875 kfree(adap2pinfo(adapter, i)->rss);
5876 free_netdev(adapter->port[i]);
5878 if (adapter->flags & CXGB4_FW_OK)
5879 t4_fw_bye(adapter, adapter->pf);
5882 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN | \
5884 #define VLAN_FEAT (NETIF_F_SG | NETIF_F_IP_CSUM | TSO_FLAGS | \
5885 NETIF_F_GRO | NETIF_F_IPV6_CSUM | NETIF_F_HIGHDMA)
5886 #define SEGMENT_SIZE 128
5888 static int t4_get_chip_type(struct adapter *adap, int ver)
5890 u32 pl_rev = REV_G(t4_read_reg(adap, PL_REV_A));
5894 return CHELSIO_CHIP_CODE(CHELSIO_T4, pl_rev);
5896 return CHELSIO_CHIP_CODE(CHELSIO_T5, pl_rev);
5898 return CHELSIO_CHIP_CODE(CHELSIO_T6, pl_rev);
5905 #ifdef CONFIG_PCI_IOV
5906 static void cxgb4_mgmt_setup(struct net_device *dev)
5908 dev->type = ARPHRD_NONE;
5910 dev->hard_header_len = 0;
5912 dev->tx_queue_len = 0;
5913 dev->flags |= IFF_NOARP;
5914 dev->priv_flags |= IFF_NO_QUEUE;
5916 /* Initialize the device structure. */
5917 dev->netdev_ops = &cxgb4_mgmt_netdev_ops;
5918 dev->ethtool_ops = &cxgb4_mgmt_ethtool_ops;
5921 static int cxgb4_iov_configure(struct pci_dev *pdev, int num_vfs)
5923 struct adapter *adap = pci_get_drvdata(pdev);
5925 int current_vfs = pci_num_vf(pdev);
5928 pcie_fw = readl(adap->regs + PCIE_FW_A);
5929 /* Check if fw is initialized */
5930 if (!(pcie_fw & PCIE_FW_INIT_F)) {
5931 dev_warn(&pdev->dev, "Device not initialized\n");
5935 /* If any of the VF's is already assigned to Guest OS, then
5936 * SRIOV for the same cannot be modified
5938 if (current_vfs && pci_vfs_assigned(pdev)) {
5940 "Cannot modify SR-IOV while VFs are assigned\n");
5943 /* Note that the upper-level code ensures that we're never called with
5944 * a non-zero "num_vfs" when we already have VFs instantiated. But
5945 * it never hurts to code defensively.
5947 if (num_vfs != 0 && current_vfs != 0)
5950 /* Nothing to do for no change. */
5951 if (num_vfs == current_vfs)
5954 /* Disable SRIOV when zero is passed. */
5956 pci_disable_sriov(pdev);
5957 /* free VF Management Interface */
5958 unregister_netdev(adap->port[0]);
5959 free_netdev(adap->port[0]);
5960 adap->port[0] = NULL;
5962 /* free VF resources */
5964 kfree(adap->vfinfo);
5965 adap->vfinfo = NULL;
5970 struct fw_pfvf_cmd port_cmd, port_rpl;
5971 struct net_device *netdev;
5972 unsigned int pmask, port;
5973 struct pci_dev *pbridge;
5974 struct port_info *pi;
5975 char name[IFNAMSIZ];
5979 /* If we want to instantiate Virtual Functions, then our
5980 * parent bridge's PCI-E needs to support Alternative Routing
5981 * ID (ARI) because our VFs will show up at function offset 8
5984 pbridge = pdev->bus->self;
5985 pcie_capability_read_word(pbridge, PCI_EXP_FLAGS, &flags);
5986 pcie_capability_read_dword(pbridge, PCI_EXP_DEVCAP2, &devcap2);
5988 if ((flags & PCI_EXP_FLAGS_VERS) < 2 ||
5989 !(devcap2 & PCI_EXP_DEVCAP2_ARI)) {
5990 /* Our parent bridge does not support ARI so issue a
5991 * warning and skip instantiating the VFs. They
5992 * won't be reachable.
5994 dev_warn(&pdev->dev, "Parent bridge %02x:%02x.%x doesn't support ARI; can't instantiate Virtual Functions\n",
5995 pbridge->bus->number, PCI_SLOT(pbridge->devfn),
5996 PCI_FUNC(pbridge->devfn));
5999 memset(&port_cmd, 0, sizeof(port_cmd));
6000 port_cmd.op_to_vfn = cpu_to_be32(FW_CMD_OP_V(FW_PFVF_CMD) |
6003 FW_PFVF_CMD_PFN_V(adap->pf) |
6004 FW_PFVF_CMD_VFN_V(0));
6005 port_cmd.retval_len16 = cpu_to_be32(FW_LEN16(port_cmd));
6006 err = t4_wr_mbox(adap, adap->mbox, &port_cmd, sizeof(port_cmd),
6010 pmask = FW_PFVF_CMD_PMASK_G(be32_to_cpu(port_rpl.type_to_neq));
6011 port = ffs(pmask) - 1;
6012 /* Allocate VF Management Interface. */
6013 snprintf(name, IFNAMSIZ, "mgmtpf%d,%d", adap->adap_idx,
6015 netdev = alloc_netdev(sizeof(struct port_info),
6016 name, NET_NAME_UNKNOWN, cxgb4_mgmt_setup);
6020 pi = netdev_priv(netdev);
6024 SET_NETDEV_DEV(netdev, &pdev->dev);
6026 adap->port[0] = netdev;
6029 err = register_netdev(adap->port[0]);
6031 pr_info("Unable to register VF mgmt netdev %s\n", name);
6032 free_netdev(adap->port[0]);
6033 adap->port[0] = NULL;
6036 /* Allocate and set up VF Information. */
6037 adap->vfinfo = kcalloc(pci_sriov_get_totalvfs(pdev),
6038 sizeof(struct vf_info), GFP_KERNEL);
6039 if (!adap->vfinfo) {
6040 unregister_netdev(adap->port[0]);
6041 free_netdev(adap->port[0]);
6042 adap->port[0] = NULL;
6045 cxgb4_mgmt_fill_vf_station_mac_addr(adap);
6047 /* Instantiate the requested number of VFs. */
6048 err = pci_enable_sriov(pdev, num_vfs);
6050 pr_info("Unable to instantiate %d VFs\n", num_vfs);
6052 unregister_netdev(adap->port[0]);
6053 free_netdev(adap->port[0]);
6054 adap->port[0] = NULL;
6055 kfree(adap->vfinfo);
6056 adap->vfinfo = NULL;
6061 adap->num_vfs = num_vfs;
6064 #endif /* CONFIG_PCI_IOV */
6066 static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
6068 struct net_device *netdev;
6069 struct adapter *adapter;
6070 static int adap_idx = 1;
6071 int s_qpp, qpp, num_seg;
6072 struct port_info *pi;
6073 bool highdma = false;
6074 enum chip_type chip;
6081 err = pci_request_regions(pdev, KBUILD_MODNAME);
6083 /* Just info, some other driver may have claimed the device. */
6084 dev_info(&pdev->dev, "cannot obtain PCI resources\n");
6088 err = pci_enable_device(pdev);
6090 dev_err(&pdev->dev, "cannot enable PCI device\n");
6091 goto out_release_regions;
6094 regs = pci_ioremap_bar(pdev, 0);
6096 dev_err(&pdev->dev, "cannot map device registers\n");
6098 goto out_disable_device;
6101 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
6104 goto out_unmap_bar0;
6107 adapter->regs = regs;
6108 err = t4_wait_dev_ready(regs);
6110 goto out_free_adapter;
6112 /* We control everything through one PF */
6113 whoami = t4_read_reg(adapter, PL_WHOAMI_A);
6114 pci_read_config_word(pdev, PCI_DEVICE_ID, &device_id);
6115 chip = t4_get_chip_type(adapter, CHELSIO_PCI_ID_VER(device_id));
6116 if ((int)chip < 0) {
6117 dev_err(&pdev->dev, "Device %d is not supported\n", device_id);
6119 goto out_free_adapter;
6121 chip_ver = CHELSIO_CHIP_VERSION(chip);
6122 func = chip_ver <= CHELSIO_T5 ?
6123 SOURCEPF_G(whoami) : T6_SOURCEPF_G(whoami);
6125 adapter->pdev = pdev;
6126 adapter->pdev_dev = &pdev->dev;
6127 adapter->name = pci_name(pdev);
6128 adapter->mbox = func;
6130 adapter->params.chip = chip;
6131 adapter->adap_idx = adap_idx;
6132 adapter->msg_enable = DFLT_MSG_ENABLE;
6133 adapter->mbox_log = kzalloc(sizeof(*adapter->mbox_log) +
6134 (sizeof(struct mbox_cmd) *
6135 T4_OS_LOG_MBOX_CMDS),
6137 if (!adapter->mbox_log) {
6139 goto out_free_adapter;
6141 spin_lock_init(&adapter->mbox_lock);
6142 INIT_LIST_HEAD(&adapter->mlist.list);
6143 adapter->mbox_log->size = T4_OS_LOG_MBOX_CMDS;
6144 pci_set_drvdata(pdev, adapter);
6146 if (func != ent->driver_data) {
6147 pci_disable_device(pdev);
6148 pci_save_state(pdev); /* to restore SR-IOV later */
6152 if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
6154 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
6156 dev_err(&pdev->dev, "unable to obtain 64-bit DMA for "
6157 "coherent allocations\n");
6158 goto out_free_adapter;
6161 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
6163 dev_err(&pdev->dev, "no usable DMA configuration\n");
6164 goto out_free_adapter;
6168 pci_enable_pcie_error_reporting(pdev);
6169 pci_set_master(pdev);
6170 pci_save_state(pdev);
6172 adapter->workq = create_singlethread_workqueue("cxgb4");
6173 if (!adapter->workq) {
6175 goto out_free_adapter;
6178 /* PCI device has been enabled */
6179 adapter->flags |= CXGB4_DEV_ENABLED;
6180 memset(adapter->chan_map, 0xff, sizeof(adapter->chan_map));
6182 /* If possible, we use PCIe Relaxed Ordering Attribute to deliver
6183 * Ingress Packet Data to Free List Buffers in order to allow for
6184 * chipset performance optimizations between the Root Complex and
6185 * Memory Controllers. (Messages to the associated Ingress Queue
6186 * notifying new Packet Placement in the Free Lists Buffers will be
6187 * send without the Relaxed Ordering Attribute thus guaranteeing that
6188 * all preceding PCIe Transaction Layer Packets will be processed
6189 * first.) But some Root Complexes have various issues with Upstream
6190 * Transaction Layer Packets with the Relaxed Ordering Attribute set.
6191 * The PCIe devices which under the Root Complexes will be cleared the
6192 * Relaxed Ordering bit in the configuration space, So we check our
6193 * PCIe configuration space to see if it's flagged with advice against
6194 * using Relaxed Ordering.
6196 if (!pcie_relaxed_ordering_enabled(pdev))
6197 adapter->flags |= CXGB4_ROOT_NO_RELAXED_ORDERING;
6199 spin_lock_init(&adapter->stats_lock);
6200 spin_lock_init(&adapter->tid_release_lock);
6201 spin_lock_init(&adapter->win0_lock);
6203 INIT_WORK(&adapter->tid_release_task, process_tid_release_list);
6204 INIT_WORK(&adapter->db_full_task, process_db_full);
6205 INIT_WORK(&adapter->db_drop_task, process_db_drop);
6206 INIT_WORK(&adapter->fatal_err_notify_task, notify_fatal_err);
6208 err = t4_prep_adapter(adapter);
6210 goto out_free_adapter;
6212 if (is_kdump_kernel()) {
6213 /* Collect hardware state and append to /proc/vmcore */
6214 err = cxgb4_cudbg_vmcore_add_dump(adapter);
6216 dev_warn(adapter->pdev_dev,
6217 "Fail collecting vmcore device dump, err: %d. Continuing\n",
6223 if (!is_t4(adapter->params.chip)) {
6224 s_qpp = (QUEUESPERPAGEPF0_S +
6225 (QUEUESPERPAGEPF1_S - QUEUESPERPAGEPF0_S) *
6227 qpp = 1 << QUEUESPERPAGEPF0_G(t4_read_reg(adapter,
6228 SGE_EGRESS_QUEUES_PER_PAGE_PF_A) >> s_qpp);
6229 num_seg = PAGE_SIZE / SEGMENT_SIZE;
6231 /* Each segment size is 128B. Write coalescing is enabled only
6232 * when SGE_EGRESS_QUEUES_PER_PAGE_PF reg value for the
6233 * queue is less no of segments that can be accommodated in
6236 if (qpp > num_seg) {
6238 "Incorrect number of egress queues per page\n");
6240 goto out_free_adapter;
6242 adapter->bar2 = ioremap_wc(pci_resource_start(pdev, 2),
6243 pci_resource_len(pdev, 2));
6244 if (!adapter->bar2) {
6245 dev_err(&pdev->dev, "cannot map device bar2 region\n");
6247 goto out_free_adapter;
6251 setup_memwin(adapter);
6252 err = adap_init0(adapter, 0);
6253 #ifdef CONFIG_DEBUG_FS
6254 bitmap_zero(adapter->sge.blocked_fl, adapter->sge.egr_sz);
6256 setup_memwin_rdma(adapter);
6260 /* configure SGE_STAT_CFG_A to read WC stats */
6261 if (!is_t4(adapter->params.chip))
6262 t4_write_reg(adapter, SGE_STAT_CFG_A, STATSOURCE_T5_V(7) |
6263 (is_t5(adapter->params.chip) ? STATMODE_V(0) :
6266 /* Initialize hash mac addr list */
6267 INIT_LIST_HEAD(&adapter->mac_hlist);
6269 for_each_port(adapter, i) {
6270 /* For supporting MQPRIO Offload, need some extra
6271 * queues for each ETHOFLD TIDs. Keep it equal to
6272 * MAX_ATIDs for now. Once we connect to firmware
6273 * later and query the EOTID params, we'll come to
6274 * know the actual # of EOTIDs supported.
6276 netdev = alloc_etherdev_mq(sizeof(struct port_info),
6277 MAX_ETH_QSETS + MAX_ATIDS);
6283 SET_NETDEV_DEV(netdev, &pdev->dev);
6285 adapter->port[i] = netdev;
6286 pi = netdev_priv(netdev);
6287 pi->adapter = adapter;
6288 pi->xact_addr_filt = -1;
6290 netdev->irq = pdev->irq;
6292 netdev->hw_features = NETIF_F_SG | TSO_FLAGS |
6293 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
6294 NETIF_F_RXCSUM | NETIF_F_RXHASH | NETIF_F_GRO |
6295 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
6298 if (chip_ver > CHELSIO_T5) {
6299 netdev->hw_enc_features |= NETIF_F_IP_CSUM |
6302 NETIF_F_GSO_UDP_TUNNEL |
6303 NETIF_F_GSO_UDP_TUNNEL_CSUM |
6304 NETIF_F_TSO | NETIF_F_TSO6;
6306 netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL |
6307 NETIF_F_GSO_UDP_TUNNEL_CSUM |
6308 NETIF_F_HW_TLS_RECORD;
6312 netdev->hw_features |= NETIF_F_HIGHDMA;
6313 netdev->features |= netdev->hw_features;
6314 netdev->vlan_features = netdev->features & VLAN_FEAT;
6316 netdev->priv_flags |= IFF_UNICAST_FLT;
6318 /* MTU range: 81 - 9600 */
6319 netdev->min_mtu = 81; /* accommodate SACK */
6320 netdev->max_mtu = MAX_MTU;
6322 netdev->netdev_ops = &cxgb4_netdev_ops;
6323 #ifdef CONFIG_CHELSIO_T4_DCB
6324 netdev->dcbnl_ops = &cxgb4_dcb_ops;
6325 cxgb4_dcb_state_init(netdev);
6326 cxgb4_dcb_version_init(netdev);
6328 cxgb4_set_ethtool_ops(netdev);
6331 cxgb4_init_ethtool_dump(adapter);
6333 pci_set_drvdata(pdev, adapter);
6335 if (adapter->flags & CXGB4_FW_OK) {
6336 err = t4_port_init(adapter, func, func, 0);
6339 } else if (adapter->params.nports == 1) {
6340 /* If we don't have a connection to the firmware -- possibly
6341 * because of an error -- grab the raw VPD parameters so we
6342 * can set the proper MAC Address on the debug network
6343 * interface that we've created.
6345 u8 hw_addr[ETH_ALEN];
6346 u8 *na = adapter->params.vpd.na;
6348 err = t4_get_raw_vpd_params(adapter, &adapter->params.vpd);
6350 for (i = 0; i < ETH_ALEN; i++)
6351 hw_addr[i] = (hex2val(na[2 * i + 0]) * 16 +
6352 hex2val(na[2 * i + 1]));
6353 t4_set_hw_addr(adapter, 0, hw_addr);
6357 if (!(adapter->flags & CXGB4_FW_OK))
6358 goto fw_attach_fail;
6360 /* Configure queues and allocate tables now, they can be needed as
6361 * soon as the first register_netdev completes.
6363 err = cfg_queues(adapter);
6367 adapter->smt = t4_init_smt();
6368 if (!adapter->smt) {
6369 /* We tolerate a lack of SMT, giving up some functionality */
6370 dev_warn(&pdev->dev, "could not allocate SMT, continuing\n");
6373 adapter->l2t = t4_init_l2t(adapter->l2t_start, adapter->l2t_end);
6374 if (!adapter->l2t) {
6375 /* We tolerate a lack of L2T, giving up some functionality */
6376 dev_warn(&pdev->dev, "could not allocate L2T, continuing\n");
6377 adapter->params.offload = 0;
6380 #if IS_ENABLED(CONFIG_IPV6)
6381 if (chip_ver <= CHELSIO_T5 &&
6382 (!(t4_read_reg(adapter, LE_DB_CONFIG_A) & ASLIPCOMPEN_F))) {
6383 /* CLIP functionality is not present in hardware,
6384 * hence disable all offload features
6386 dev_warn(&pdev->dev,
6387 "CLIP not enabled in hardware, continuing\n");
6388 adapter->params.offload = 0;
6390 adapter->clipt = t4_init_clip_tbl(adapter->clipt_start,
6391 adapter->clipt_end);
6392 if (!adapter->clipt) {
6393 /* We tolerate a lack of clip_table, giving up
6394 * some functionality
6396 dev_warn(&pdev->dev,
6397 "could not allocate Clip table, continuing\n");
6398 adapter->params.offload = 0;
6403 for_each_port(adapter, i) {
6404 pi = adap2pinfo(adapter, i);
6405 pi->sched_tbl = t4_init_sched(adapter->params.nsched_cls);
6407 dev_warn(&pdev->dev,
6408 "could not activate scheduling on port %d\n",
6412 if (tid_init(&adapter->tids) < 0) {
6413 dev_warn(&pdev->dev, "could not allocate TID table, "
6415 adapter->params.offload = 0;
6417 adapter->tc_u32 = cxgb4_init_tc_u32(adapter);
6418 if (!adapter->tc_u32)
6419 dev_warn(&pdev->dev,
6420 "could not offload tc u32, continuing\n");
6422 if (cxgb4_init_tc_flower(adapter))
6423 dev_warn(&pdev->dev,
6424 "could not offload tc flower, continuing\n");
6426 if (cxgb4_init_tc_mqprio(adapter))
6427 dev_warn(&pdev->dev,
6428 "could not offload tc mqprio, continuing\n");
6430 if (cxgb4_init_tc_matchall(adapter))
6431 dev_warn(&pdev->dev,
6432 "could not offload tc matchall, continuing\n");
6435 if (is_offload(adapter) || is_hashfilter(adapter)) {
6436 if (t4_read_reg(adapter, LE_DB_CONFIG_A) & HASHEN_F) {
6437 u32 hash_base, hash_reg;
6439 if (chip_ver <= CHELSIO_T5) {
6440 hash_reg = LE_DB_TID_HASHBASE_A;
6441 hash_base = t4_read_reg(adapter, hash_reg);
6442 adapter->tids.hash_base = hash_base / 4;
6444 hash_reg = T6_LE_DB_HASH_TID_BASE_A;
6445 hash_base = t4_read_reg(adapter, hash_reg);
6446 adapter->tids.hash_base = hash_base;
6451 /* See what interrupts we'll be using */
6452 if (msi > 1 && enable_msix(adapter) == 0)
6453 adapter->flags |= CXGB4_USING_MSIX;
6454 else if (msi > 0 && pci_enable_msi(pdev) == 0) {
6455 adapter->flags |= CXGB4_USING_MSI;
6457 free_msix_info(adapter);
6460 /* check for PCI Express bandwidth capabiltites */
6461 pcie_print_link_status(pdev);
6463 cxgb4_init_mps_ref_entries(adapter);
6465 err = init_rss(adapter);
6469 err = setup_non_data_intr(adapter);
6471 dev_err(adapter->pdev_dev,
6472 "Non Data interrupt allocation failed, err: %d\n", err);
6476 err = setup_fw_sge_queues(adapter);
6478 dev_err(adapter->pdev_dev,
6479 "FW sge queue allocation failed, err %d", err);
6485 * The card is now ready to go. If any errors occur during device
6486 * registration we do not fail the whole card but rather proceed only
6487 * with the ports we manage to register successfully. However we must
6488 * register at least one net device.
6490 for_each_port(adapter, i) {
6491 pi = adap2pinfo(adapter, i);
6492 adapter->port[i]->dev_port = pi->lport;
6493 netif_set_real_num_tx_queues(adapter->port[i], pi->nqsets);
6494 netif_set_real_num_rx_queues(adapter->port[i], pi->nqsets);
6496 netif_carrier_off(adapter->port[i]);
6498 err = register_netdev(adapter->port[i]);
6501 adapter->chan_map[pi->tx_chan] = i;
6502 print_port_info(adapter->port[i]);
6505 dev_err(&pdev->dev, "could not register any net devices\n");
6509 dev_warn(&pdev->dev, "only %d net devices registered\n", i);
6513 if (cxgb4_debugfs_root) {
6514 adapter->debugfs_root = debugfs_create_dir(pci_name(pdev),
6515 cxgb4_debugfs_root);
6516 setup_debugfs(adapter);
6519 /* PCIe EEH recovery on powerpc platforms needs fundamental reset */
6520 pdev->needs_freset = 1;
6522 if (is_uld(adapter)) {
6523 mutex_lock(&uld_mutex);
6524 list_add_tail(&adapter->list_node, &adapter_list);
6525 mutex_unlock(&uld_mutex);
6528 if (!is_t4(adapter->params.chip))
6529 cxgb4_ptp_init(adapter);
6531 if (IS_REACHABLE(CONFIG_THERMAL) &&
6532 !is_t4(adapter->params.chip) && (adapter->flags & CXGB4_FW_OK))
6533 cxgb4_thermal_init(adapter);
6535 print_adapter_info(adapter);
6539 t4_free_sge_resources(adapter);
6540 free_some_resources(adapter);
6541 if (adapter->flags & CXGB4_USING_MSIX)
6542 free_msix_info(adapter);
6543 if (adapter->num_uld || adapter->num_ofld_uld)
6544 t4_uld_mem_free(adapter);
6546 if (!is_t4(adapter->params.chip))
6547 iounmap(adapter->bar2);
6550 destroy_workqueue(adapter->workq);
6552 kfree(adapter->mbox_log);
6557 pci_disable_pcie_error_reporting(pdev);
6558 pci_disable_device(pdev);
6559 out_release_regions:
6560 pci_release_regions(pdev);
6564 static void remove_one(struct pci_dev *pdev)
6566 struct adapter *adapter = pci_get_drvdata(pdev);
6567 struct hash_mac_addr *entry, *tmp;
6570 pci_release_regions(pdev);
6574 /* If we allocated filters, free up state associated with any
6577 clear_all_filters(adapter);
6579 adapter->flags |= CXGB4_SHUTTING_DOWN;
6581 if (adapter->pf == 4) {
6584 /* Tear down per-adapter Work Queue first since it can contain
6585 * references to our adapter data structure.
6587 destroy_workqueue(adapter->workq);
6589 if (is_uld(adapter)) {
6590 detach_ulds(adapter);
6591 t4_uld_clean_up(adapter);
6594 adap_free_hma_mem(adapter);
6596 disable_interrupts(adapter);
6598 cxgb4_free_mps_ref_entries(adapter);
6600 for_each_port(adapter, i)
6601 if (adapter->port[i]->reg_state == NETREG_REGISTERED)
6602 unregister_netdev(adapter->port[i]);
6604 debugfs_remove_recursive(adapter->debugfs_root);
6606 if (!is_t4(adapter->params.chip))
6607 cxgb4_ptp_stop(adapter);
6608 if (IS_REACHABLE(CONFIG_THERMAL))
6609 cxgb4_thermal_remove(adapter);
6611 if (adapter->flags & CXGB4_FULL_INIT_DONE)
6614 if (adapter->flags & CXGB4_USING_MSIX)
6615 free_msix_info(adapter);
6616 if (adapter->num_uld || adapter->num_ofld_uld)
6617 t4_uld_mem_free(adapter);
6618 free_some_resources(adapter);
6619 list_for_each_entry_safe(entry, tmp, &adapter->mac_hlist,
6621 list_del(&entry->list);
6625 #if IS_ENABLED(CONFIG_IPV6)
6626 t4_cleanup_clip_tbl(adapter);
6628 if (!is_t4(adapter->params.chip))
6629 iounmap(adapter->bar2);
6631 #ifdef CONFIG_PCI_IOV
6633 cxgb4_iov_configure(adapter->pdev, 0);
6636 iounmap(adapter->regs);
6637 pci_disable_pcie_error_reporting(pdev);
6638 if ((adapter->flags & CXGB4_DEV_ENABLED)) {
6639 pci_disable_device(pdev);
6640 adapter->flags &= ~CXGB4_DEV_ENABLED;
6642 pci_release_regions(pdev);
6643 kfree(adapter->mbox_log);
6648 /* "Shutdown" quiesces the device, stopping Ingress Packet and Interrupt
6649 * delivery. This is essentially a stripped down version of the PCI remove()
6650 * function where we do the minimal amount of work necessary to shutdown any
6653 static void shutdown_one(struct pci_dev *pdev)
6655 struct adapter *adapter = pci_get_drvdata(pdev);
6657 /* As with remove_one() above (see extended comment), we only want do
6658 * do cleanup on PCI Devices which went all the way through init_one()
6662 pci_release_regions(pdev);
6666 adapter->flags |= CXGB4_SHUTTING_DOWN;
6668 if (adapter->pf == 4) {
6671 for_each_port(adapter, i)
6672 if (adapter->port[i]->reg_state == NETREG_REGISTERED)
6673 cxgb_close(adapter->port[i]);
6676 cxgb4_mqprio_stop_offload(adapter);
6679 if (is_uld(adapter)) {
6680 detach_ulds(adapter);
6681 t4_uld_clean_up(adapter);
6684 disable_interrupts(adapter);
6685 disable_msi(adapter);
6687 t4_sge_stop(adapter);
6688 if (adapter->flags & CXGB4_FW_OK)
6689 t4_fw_bye(adapter, adapter->mbox);
6693 static struct pci_driver cxgb4_driver = {
6694 .name = KBUILD_MODNAME,
6695 .id_table = cxgb4_pci_tbl,
6697 .remove = remove_one,
6698 .shutdown = shutdown_one,
6699 #ifdef CONFIG_PCI_IOV
6700 .sriov_configure = cxgb4_iov_configure,
6702 .err_handler = &cxgb4_eeh,
6705 static int __init cxgb4_init_module(void)
6709 cxgb4_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
6711 ret = pci_register_driver(&cxgb4_driver);
6715 #if IS_ENABLED(CONFIG_IPV6)
6716 if (!inet6addr_registered) {
6717 ret = register_inet6addr_notifier(&cxgb4_inet6addr_notifier);
6719 pci_unregister_driver(&cxgb4_driver);
6721 inet6addr_registered = true;
6729 debugfs_remove(cxgb4_debugfs_root);
6734 static void __exit cxgb4_cleanup_module(void)
6736 #if IS_ENABLED(CONFIG_IPV6)
6737 if (inet6addr_registered) {
6738 unregister_inet6addr_notifier(&cxgb4_inet6addr_notifier);
6739 inet6addr_registered = false;
6742 pci_unregister_driver(&cxgb4_driver);
6743 debugfs_remove(cxgb4_debugfs_root); /* NULL ok */
6746 module_init(cxgb4_init_module);
6747 module_exit(cxgb4_cleanup_module);