2 * This file is part of the Chelsio T4 Ethernet driver for Linux.
4 * Copyright (c) 2003-2016 Chelsio Communications, Inc. All rights reserved.
6 * This software is available to you under a choice of one of two
7 * licenses. You may choose to be licensed under the terms of the GNU
8 * General Public License (GPL) Version 2, available from the file
9 * COPYING in the main directory of this source tree, or the
10 * OpenIB.org BSD license below:
12 * Redistribution and use in source and binary forms, with or
13 * without modification, are permitted provided that the following
16 * - Redistributions of source code must retain the above
17 * copyright notice, this list of conditions and the following
20 * - Redistributions in binary form must reproduce the above
21 * copyright notice, this list of conditions and the following
22 * disclaimer in the documentation and/or other materials
23 * provided with the distribution.
25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
35 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
37 #include <linux/bitmap.h>
38 #include <linux/crc32.h>
39 #include <linux/ctype.h>
40 #include <linux/debugfs.h>
41 #include <linux/err.h>
42 #include <linux/etherdevice.h>
43 #include <linux/firmware.h>
45 #include <linux/if_vlan.h>
46 #include <linux/init.h>
47 #include <linux/log2.h>
48 #include <linux/mdio.h>
49 #include <linux/module.h>
50 #include <linux/moduleparam.h>
51 #include <linux/mutex.h>
52 #include <linux/netdevice.h>
53 #include <linux/pci.h>
54 #include <linux/aer.h>
55 #include <linux/rtnetlink.h>
56 #include <linux/sched.h>
57 #include <linux/seq_file.h>
58 #include <linux/sockios.h>
59 #include <linux/vmalloc.h>
60 #include <linux/workqueue.h>
61 #include <net/neighbour.h>
62 #include <net/netevent.h>
63 #include <net/addrconf.h>
64 #include <net/bonding.h>
65 #include <net/addrconf.h>
66 #include <linux/uaccess.h>
67 #include <linux/crash_dump.h>
70 #include "cxgb4_filter.h"
72 #include "t4_values.h"
75 #include "t4fw_version.h"
76 #include "cxgb4_dcb.h"
77 #include "cxgb4_debugfs.h"
81 #include "cxgb4_tc_u32.h"
83 char cxgb4_driver_name[] = KBUILD_MODNAME;
88 #define DRV_VERSION "2.0.0-ko"
89 const char cxgb4_driver_version[] = DRV_VERSION;
90 #define DRV_DESC "Chelsio T4/T5/T6 Network Driver"
92 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
93 NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
94 NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
96 /* Macros needed to support the PCI Device ID Table ...
98 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_BEGIN \
99 static const struct pci_device_id cxgb4_pci_tbl[] = {
100 #define CH_PCI_DEVICE_ID_FUNCTION 0x4
102 /* Include PCI Device IDs for both PF4 and PF0-3 so our PCI probe() routine is
105 #define CH_PCI_DEVICE_ID_FUNCTION2 0x0
107 #define CH_PCI_ID_TABLE_ENTRY(devid) \
108 {PCI_VDEVICE(CHELSIO, (devid)), 4}
110 #define CH_PCI_DEVICE_ID_TABLE_DEFINE_END \
114 #include "t4_pci_id_tbl.h"
116 #define FW4_FNAME "cxgb4/t4fw.bin"
117 #define FW5_FNAME "cxgb4/t5fw.bin"
118 #define FW6_FNAME "cxgb4/t6fw.bin"
119 #define FW4_CFNAME "cxgb4/t4-config.txt"
120 #define FW5_CFNAME "cxgb4/t5-config.txt"
121 #define FW6_CFNAME "cxgb4/t6-config.txt"
122 #define PHY_AQ1202_FIRMWARE "cxgb4/aq1202_fw.cld"
123 #define PHY_BCM84834_FIRMWARE "cxgb4/bcm8483.bin"
124 #define PHY_AQ1202_DEVICEID 0x4409
125 #define PHY_BCM84834_DEVICEID 0x4486
127 MODULE_DESCRIPTION(DRV_DESC);
128 MODULE_AUTHOR("Chelsio Communications");
129 MODULE_LICENSE("Dual BSD/GPL");
130 MODULE_VERSION(DRV_VERSION);
131 MODULE_DEVICE_TABLE(pci, cxgb4_pci_tbl);
132 MODULE_FIRMWARE(FW4_FNAME);
133 MODULE_FIRMWARE(FW5_FNAME);
134 MODULE_FIRMWARE(FW6_FNAME);
137 * The driver uses the best interrupt scheme available on a platform in the
138 * order MSI-X, MSI, legacy INTx interrupts. This parameter determines which
139 * of these schemes the driver may consider as follows:
141 * msi = 2: choose from among all three options
142 * msi = 1: only consider MSI and INTx interrupts
143 * msi = 0: force INTx interrupts
147 module_param(msi, int, 0644);
148 MODULE_PARM_DESC(msi, "whether to use INTx (0), MSI (1) or MSI-X (2)");
151 * Normally we tell the chip to deliver Ingress Packets into our DMA buffers
152 * offset by 2 bytes in order to have the IP headers line up on 4-byte
153 * boundaries. This is a requirement for many architectures which will throw
154 * a machine check fault if an attempt is made to access one of the 4-byte IP
155 * header fields on a non-4-byte boundary. And it's a major performance issue
156 * even on some architectures which allow it like some implementations of the
157 * x86 ISA. However, some architectures don't mind this and for some very
158 * edge-case performance sensitive applications (like forwarding large volumes
159 * of small packets), setting this DMA offset to 0 will decrease the number of
160 * PCI-E Bus transfers enough to measurably affect performance.
162 static int rx_dma_offset = 2;
164 /* TX Queue select used to determine what algorithm to use for selecting TX
165 * queue. Select between the kernel provided function (select_queue=0) or user
166 * cxgb_select_queue function (select_queue=1)
168 * Default: select_queue=0
170 static int select_queue;
171 module_param(select_queue, int, 0644);
172 MODULE_PARM_DESC(select_queue,
173 "Select between kernel provided method of selecting or driver method of selecting TX queue. Default is kernel method.");
175 static struct dentry *cxgb4_debugfs_root;
177 LIST_HEAD(adapter_list);
178 DEFINE_MUTEX(uld_mutex);
180 static void link_report(struct net_device *dev)
182 if (!netif_carrier_ok(dev))
183 netdev_info(dev, "link down\n");
185 static const char *fc[] = { "no", "Rx", "Tx", "Tx/Rx" };
188 const struct port_info *p = netdev_priv(dev);
190 switch (p->link_cfg.speed) {
210 pr_info("%s: unsupported speed: %d\n",
211 dev->name, p->link_cfg.speed);
215 netdev_info(dev, "link up, %s, full-duplex, %s PAUSE\n", s,
220 #ifdef CONFIG_CHELSIO_T4_DCB
221 /* Set up/tear down Data Center Bridging Priority mapping for a net device. */
222 static void dcb_tx_queue_prio_enable(struct net_device *dev, int enable)
224 struct port_info *pi = netdev_priv(dev);
225 struct adapter *adap = pi->adapter;
226 struct sge_eth_txq *txq = &adap->sge.ethtxq[pi->first_qset];
229 /* We use a simple mapping of Port TX Queue Index to DCB
230 * Priority when we're enabling DCB.
232 for (i = 0; i < pi->nqsets; i++, txq++) {
236 name = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DMAQ) |
238 FW_PARAMS_PARAM_DMAQ_EQ_DCBPRIO_ETH) |
239 FW_PARAMS_PARAM_YZ_V(txq->q.cntxt_id));
240 value = enable ? i : 0xffffffff;
242 /* Since we can be called while atomic (from "interrupt
243 * level") we need to issue the Set Parameters Commannd
244 * without sleeping (timeout < 0).
246 err = t4_set_params_timeout(adap, adap->mbox, adap->pf, 0, 1,
248 -FW_CMD_MAX_TIMEOUT);
251 dev_err(adap->pdev_dev,
252 "Can't %s DCB Priority on port %d, TX Queue %d: err=%d\n",
253 enable ? "set" : "unset", pi->port_id, i, -err);
255 txq->dcb_prio = value;
259 static int cxgb4_dcb_enabled(const struct net_device *dev)
261 struct port_info *pi = netdev_priv(dev);
263 if (!pi->dcb.enabled)
266 return ((pi->dcb.state == CXGB4_DCB_STATE_FW_ALLSYNCED) ||
267 (pi->dcb.state == CXGB4_DCB_STATE_HOST));
269 #endif /* CONFIG_CHELSIO_T4_DCB */
271 void t4_os_link_changed(struct adapter *adapter, int port_id, int link_stat)
273 struct net_device *dev = adapter->port[port_id];
275 /* Skip changes from disabled ports. */
276 if (netif_running(dev) && link_stat != netif_carrier_ok(dev)) {
278 netif_carrier_on(dev);
280 #ifdef CONFIG_CHELSIO_T4_DCB
281 if (cxgb4_dcb_enabled(dev)) {
282 cxgb4_dcb_state_init(dev);
283 dcb_tx_queue_prio_enable(dev, false);
285 #endif /* CONFIG_CHELSIO_T4_DCB */
286 netif_carrier_off(dev);
293 void t4_os_portmod_changed(const struct adapter *adap, int port_id)
295 static const char *mod_str[] = {
296 NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
299 const struct net_device *dev = adap->port[port_id];
300 const struct port_info *pi = netdev_priv(dev);
302 if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
303 netdev_info(dev, "port module unplugged\n");
304 else if (pi->mod_type < ARRAY_SIZE(mod_str))
305 netdev_info(dev, "%s module inserted\n", mod_str[pi->mod_type]);
306 else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
307 netdev_info(dev, "%s: unsupported port module inserted\n",
309 else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
310 netdev_info(dev, "%s: unknown port module inserted\n",
312 else if (pi->mod_type == FW_PORT_MOD_TYPE_ERROR)
313 netdev_info(dev, "%s: transceiver module error\n", dev->name);
315 netdev_info(dev, "%s: unknown module type %d inserted\n",
316 dev->name, pi->mod_type);
319 int dbfifo_int_thresh = 10; /* 10 == 640 entry threshold */
320 module_param(dbfifo_int_thresh, int, 0644);
321 MODULE_PARM_DESC(dbfifo_int_thresh, "doorbell fifo interrupt threshold");
324 * usecs to sleep while draining the dbfifo
326 static int dbfifo_drain_delay = 1000;
327 module_param(dbfifo_drain_delay, int, 0644);
328 MODULE_PARM_DESC(dbfifo_drain_delay,
329 "usecs to sleep while draining the dbfifo");
331 static inline int cxgb4_set_addr_hash(struct port_info *pi)
333 struct adapter *adap = pi->adapter;
336 struct hash_mac_addr *entry;
338 /* Calculate the hash vector for the updated list and program it */
339 list_for_each_entry(entry, &adap->mac_hlist, list) {
340 ucast |= is_unicast_ether_addr(entry->addr);
341 vec |= (1ULL << hash_mac_addr(entry->addr));
343 return t4_set_addr_hash(adap, adap->mbox, pi->viid, ucast,
347 static int cxgb4_mac_sync(struct net_device *netdev, const u8 *mac_addr)
349 struct port_info *pi = netdev_priv(netdev);
350 struct adapter *adap = pi->adapter;
355 bool ucast = is_unicast_ether_addr(mac_addr);
356 const u8 *maclist[1] = {mac_addr};
357 struct hash_mac_addr *new_entry;
359 ret = t4_alloc_mac_filt(adap, adap->mbox, pi->viid, free, 1, maclist,
360 NULL, ucast ? &uhash : &mhash, false);
363 /* if hash != 0, then add the addr to hash addr list
364 * so on the end we will calculate the hash for the
365 * list and program it
367 if (uhash || mhash) {
368 new_entry = kzalloc(sizeof(*new_entry), GFP_ATOMIC);
371 ether_addr_copy(new_entry->addr, mac_addr);
372 list_add_tail(&new_entry->list, &adap->mac_hlist);
373 ret = cxgb4_set_addr_hash(pi);
376 return ret < 0 ? ret : 0;
379 static int cxgb4_mac_unsync(struct net_device *netdev, const u8 *mac_addr)
381 struct port_info *pi = netdev_priv(netdev);
382 struct adapter *adap = pi->adapter;
384 const u8 *maclist[1] = {mac_addr};
385 struct hash_mac_addr *entry, *tmp;
387 /* If the MAC address to be removed is in the hash addr
388 * list, delete it from the list and update hash vector
390 list_for_each_entry_safe(entry, tmp, &adap->mac_hlist, list) {
391 if (ether_addr_equal(entry->addr, mac_addr)) {
392 list_del(&entry->list);
394 return cxgb4_set_addr_hash(pi);
398 ret = t4_free_mac_filt(adap, adap->mbox, pi->viid, 1, maclist, false);
399 return ret < 0 ? -EINVAL : 0;
403 * Set Rx properties of a port, such as promiscruity, address filters, and MTU.
404 * If @mtu is -1 it is left unchanged.
406 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
408 struct port_info *pi = netdev_priv(dev);
409 struct adapter *adapter = pi->adapter;
411 __dev_uc_sync(dev, cxgb4_mac_sync, cxgb4_mac_unsync);
412 __dev_mc_sync(dev, cxgb4_mac_sync, cxgb4_mac_unsync);
414 return t4_set_rxmode(adapter, adapter->mbox, pi->viid, mtu,
415 (dev->flags & IFF_PROMISC) ? 1 : 0,
416 (dev->flags & IFF_ALLMULTI) ? 1 : 0, 1, -1,
421 * link_start - enable a port
422 * @dev: the port to enable
424 * Performs the MAC and PHY actions needed to enable a port.
426 static int link_start(struct net_device *dev)
429 struct port_info *pi = netdev_priv(dev);
430 unsigned int mb = pi->adapter->pf;
433 * We do not set address filters and promiscuity here, the stack does
434 * that step explicitly.
436 ret = t4_set_rxmode(pi->adapter, mb, pi->viid, dev->mtu, -1, -1, -1,
437 !!(dev->features & NETIF_F_HW_VLAN_CTAG_RX), true);
439 ret = t4_change_mac(pi->adapter, mb, pi->viid,
440 pi->xact_addr_filt, dev->dev_addr, true,
443 pi->xact_addr_filt = ret;
448 ret = t4_link_l1cfg(pi->adapter, mb, pi->tx_chan,
452 ret = t4_enable_vi_params(pi->adapter, mb, pi->viid, true,
453 true, CXGB4_DCB_ENABLED);
460 #ifdef CONFIG_CHELSIO_T4_DCB
461 /* Handle a Data Center Bridging update message from the firmware. */
462 static void dcb_rpl(struct adapter *adap, const struct fw_port_cmd *pcmd)
464 int port = FW_PORT_CMD_PORTID_G(ntohl(pcmd->op_to_portid));
465 struct net_device *dev = adap->port[adap->chan_map[port]];
466 int old_dcb_enabled = cxgb4_dcb_enabled(dev);
469 cxgb4_dcb_handle_fw_update(adap, pcmd);
470 new_dcb_enabled = cxgb4_dcb_enabled(dev);
472 /* If the DCB has become enabled or disabled on the port then we're
473 * going to need to set up/tear down DCB Priority parameters for the
474 * TX Queues associated with the port.
476 if (new_dcb_enabled != old_dcb_enabled)
477 dcb_tx_queue_prio_enable(dev, new_dcb_enabled);
479 #endif /* CONFIG_CHELSIO_T4_DCB */
481 /* Response queue handler for the FW event queue.
483 static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
484 const struct pkt_gl *gl)
486 u8 opcode = ((const struct rss_header *)rsp)->opcode;
488 rsp++; /* skip RSS header */
490 /* FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG.
492 if (unlikely(opcode == CPL_FW4_MSG &&
493 ((const struct cpl_fw4_msg *)rsp)->type == FW_TYPE_RSSCPL)) {
495 opcode = ((const struct rss_header *)rsp)->opcode;
497 if (opcode != CPL_SGE_EGR_UPDATE) {
498 dev_err(q->adap->pdev_dev, "unexpected FW4/CPL %#x on FW event queue\n"
504 if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
505 const struct cpl_sge_egr_update *p = (void *)rsp;
506 unsigned int qid = EGR_QID_G(ntohl(p->opcode_qid));
509 txq = q->adap->sge.egr_map[qid - q->adap->sge.egr_start];
511 if (txq->q_type == CXGB4_TXQ_ETH) {
512 struct sge_eth_txq *eq;
514 eq = container_of(txq, struct sge_eth_txq, q);
515 netif_tx_wake_queue(eq->txq);
517 struct sge_uld_txq *oq;
519 oq = container_of(txq, struct sge_uld_txq, q);
520 tasklet_schedule(&oq->qresume_tsk);
522 } else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
523 const struct cpl_fw6_msg *p = (void *)rsp;
525 #ifdef CONFIG_CHELSIO_T4_DCB
526 const struct fw_port_cmd *pcmd = (const void *)p->data;
527 unsigned int cmd = FW_CMD_OP_G(ntohl(pcmd->op_to_portid));
528 unsigned int action =
529 FW_PORT_CMD_ACTION_G(ntohl(pcmd->action_to_len16));
531 if (cmd == FW_PORT_CMD &&
532 action == FW_PORT_ACTION_GET_PORT_INFO) {
533 int port = FW_PORT_CMD_PORTID_G(
534 be32_to_cpu(pcmd->op_to_portid));
535 struct net_device *dev =
536 q->adap->port[q->adap->chan_map[port]];
537 int state_input = ((pcmd->u.info.dcbxdis_pkd &
538 FW_PORT_CMD_DCBXDIS_F)
539 ? CXGB4_DCB_INPUT_FW_DISABLED
540 : CXGB4_DCB_INPUT_FW_ENABLED);
542 cxgb4_dcb_state_fsm(dev, state_input);
545 if (cmd == FW_PORT_CMD &&
546 action == FW_PORT_ACTION_L2_DCB_CFG)
547 dcb_rpl(q->adap, pcmd);
551 t4_handle_fw_rpl(q->adap, p->data);
552 } else if (opcode == CPL_L2T_WRITE_RPL) {
553 const struct cpl_l2t_write_rpl *p = (void *)rsp;
555 do_l2t_write_rpl(q->adap, p);
556 } else if (opcode == CPL_SET_TCB_RPL) {
557 const struct cpl_set_tcb_rpl *p = (void *)rsp;
559 filter_rpl(q->adap, p);
561 dev_err(q->adap->pdev_dev,
562 "unexpected CPL %#x on FW event queue\n", opcode);
567 static void disable_msi(struct adapter *adapter)
569 if (adapter->flags & USING_MSIX) {
570 pci_disable_msix(adapter->pdev);
571 adapter->flags &= ~USING_MSIX;
572 } else if (adapter->flags & USING_MSI) {
573 pci_disable_msi(adapter->pdev);
574 adapter->flags &= ~USING_MSI;
579 * Interrupt handler for non-data events used with MSI-X.
581 static irqreturn_t t4_nondata_intr(int irq, void *cookie)
583 struct adapter *adap = cookie;
584 u32 v = t4_read_reg(adap, MYPF_REG(PL_PF_INT_CAUSE_A));
588 t4_write_reg(adap, MYPF_REG(PL_PF_INT_CAUSE_A), v);
590 if (adap->flags & MASTER_PF)
591 t4_slow_intr_handler(adap);
596 * Name the MSI-X interrupts.
598 static void name_msix_vecs(struct adapter *adap)
600 int i, j, msi_idx = 2, n = sizeof(adap->msix_info[0].desc);
602 /* non-data interrupts */
603 snprintf(adap->msix_info[0].desc, n, "%s", adap->port[0]->name);
606 snprintf(adap->msix_info[1].desc, n, "%s-FWeventq",
607 adap->port[0]->name);
609 /* Ethernet queues */
610 for_each_port(adap, j) {
611 struct net_device *d = adap->port[j];
612 const struct port_info *pi = netdev_priv(d);
614 for (i = 0; i < pi->nqsets; i++, msi_idx++)
615 snprintf(adap->msix_info[msi_idx].desc, n, "%s-Rx%d",
620 static int request_msix_queue_irqs(struct adapter *adap)
622 struct sge *s = &adap->sge;
626 err = request_irq(adap->msix_info[1].vec, t4_sge_intr_msix, 0,
627 adap->msix_info[1].desc, &s->fw_evtq);
631 for_each_ethrxq(s, ethqidx) {
632 err = request_irq(adap->msix_info[msi_index].vec,
634 adap->msix_info[msi_index].desc,
635 &s->ethrxq[ethqidx].rspq);
643 while (--ethqidx >= 0)
644 free_irq(adap->msix_info[--msi_index].vec,
645 &s->ethrxq[ethqidx].rspq);
646 free_irq(adap->msix_info[1].vec, &s->fw_evtq);
650 static void free_msix_queue_irqs(struct adapter *adap)
652 int i, msi_index = 2;
653 struct sge *s = &adap->sge;
655 free_irq(adap->msix_info[1].vec, &s->fw_evtq);
656 for_each_ethrxq(s, i)
657 free_irq(adap->msix_info[msi_index++].vec, &s->ethrxq[i].rspq);
661 * cxgb4_write_rss - write the RSS table for a given port
663 * @queues: array of queue indices for RSS
665 * Sets up the portion of the HW RSS table for the port's VI to distribute
666 * packets to the Rx queues in @queues.
667 * Should never be called before setting up sge eth rx queues
669 int cxgb4_write_rss(const struct port_info *pi, const u16 *queues)
673 struct adapter *adapter = pi->adapter;
674 const struct sge_eth_rxq *rxq;
676 rxq = &adapter->sge.ethrxq[pi->first_qset];
677 rss = kmalloc(pi->rss_size * sizeof(u16), GFP_KERNEL);
681 /* map the queue indices to queue ids */
682 for (i = 0; i < pi->rss_size; i++, queues++)
683 rss[i] = rxq[*queues].rspq.abs_id;
685 err = t4_config_rss_range(adapter, adapter->pf, pi->viid, 0,
686 pi->rss_size, rss, pi->rss_size);
687 /* If Tunnel All Lookup isn't specified in the global RSS
688 * Configuration, then we need to specify a default Ingress
689 * Queue for any ingress packets which aren't hashed. We'll
690 * use our first ingress queue ...
693 err = t4_config_vi_rss(adapter, adapter->mbox, pi->viid,
694 FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN_F |
695 FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN_F |
696 FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN_F |
697 FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN_F |
698 FW_RSS_VI_CONFIG_CMD_UDPEN_F,
705 * setup_rss - configure RSS
708 * Sets up RSS for each port.
710 static int setup_rss(struct adapter *adap)
714 for_each_port(adap, i) {
715 const struct port_info *pi = adap2pinfo(adap, i);
717 /* Fill default values with equal distribution */
718 for (j = 0; j < pi->rss_size; j++)
719 pi->rss[j] = j % pi->nqsets;
721 err = cxgb4_write_rss(pi, pi->rss);
729 * Return the channel of the ingress queue with the given qid.
731 static unsigned int rxq_to_chan(const struct sge *p, unsigned int qid)
733 qid -= p->ingr_start;
734 return netdev2pinfo(p->ingr_map[qid]->netdev)->tx_chan;
738 * Wait until all NAPI handlers are descheduled.
740 static void quiesce_rx(struct adapter *adap)
744 for (i = 0; i < adap->sge.ingr_sz; i++) {
745 struct sge_rspq *q = adap->sge.ingr_map[i];
748 napi_disable(&q->napi);
752 /* Disable interrupt and napi handler */
753 static void disable_interrupts(struct adapter *adap)
755 if (adap->flags & FULL_INIT_DONE) {
756 t4_intr_disable(adap);
757 if (adap->flags & USING_MSIX) {
758 free_msix_queue_irqs(adap);
759 free_irq(adap->msix_info[0].vec, adap);
761 free_irq(adap->pdev->irq, adap);
768 * Enable NAPI scheduling and interrupt generation for all Rx queues.
770 static void enable_rx(struct adapter *adap)
774 for (i = 0; i < adap->sge.ingr_sz; i++) {
775 struct sge_rspq *q = adap->sge.ingr_map[i];
780 napi_enable(&q->napi);
782 /* 0-increment GTS to start the timer and enable interrupts */
783 t4_write_reg(adap, MYPF_REG(SGE_PF_GTS_A),
784 SEINTARM_V(q->intr_params) |
785 INGRESSQID_V(q->cntxt_id));
790 static int setup_fw_sge_queues(struct adapter *adap)
792 struct sge *s = &adap->sge;
795 bitmap_zero(s->starving_fl, s->egr_sz);
796 bitmap_zero(s->txq_maperr, s->egr_sz);
798 if (adap->flags & USING_MSIX)
799 adap->msi_idx = 1; /* vector 0 is for non-queue interrupts */
801 err = t4_sge_alloc_rxq(adap, &s->intrq, false, adap->port[0], 0,
802 NULL, NULL, NULL, -1);
805 adap->msi_idx = -((int)s->intrq.abs_id + 1);
808 err = t4_sge_alloc_rxq(adap, &s->fw_evtq, true, adap->port[0],
809 adap->msi_idx, NULL, fwevtq_handler, NULL, -1);
811 t4_free_sge_resources(adap);
816 * setup_sge_queues - configure SGE Tx/Rx/response queues
819 * Determines how many sets of SGE queues to use and initializes them.
820 * We support multiple queue sets per port if we have MSI-X, otherwise
821 * just one queue set per port.
823 static int setup_sge_queues(struct adapter *adap)
826 struct sge *s = &adap->sge;
827 struct sge_uld_rxq_info *rxq_info = s->uld_rxq_info[CXGB4_ULD_RDMA];
828 unsigned int cmplqid = 0;
830 for_each_port(adap, i) {
831 struct net_device *dev = adap->port[i];
832 struct port_info *pi = netdev_priv(dev);
833 struct sge_eth_rxq *q = &s->ethrxq[pi->first_qset];
834 struct sge_eth_txq *t = &s->ethtxq[pi->first_qset];
836 for (j = 0; j < pi->nqsets; j++, q++) {
837 if (adap->msi_idx > 0)
839 err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev,
840 adap->msi_idx, &q->fl,
843 t4_get_mps_bg_map(adap,
848 memset(&q->stats, 0, sizeof(q->stats));
850 for (j = 0; j < pi->nqsets; j++, t++) {
851 err = t4_sge_alloc_eth_txq(adap, t, dev,
852 netdev_get_tx_queue(dev, j),
853 s->fw_evtq.cntxt_id);
859 for_each_port(adap, i) {
860 /* Note that cmplqid below is 0 if we don't
861 * have RDMA queues, and that's the right value.
864 cmplqid = rxq_info->uldrxq[i].rspq.cntxt_id;
866 err = t4_sge_alloc_ctrl_txq(adap, &s->ctrlq[i], adap->port[i],
867 s->fw_evtq.cntxt_id, cmplqid);
872 t4_write_reg(adap, is_t4(adap->params.chip) ?
873 MPS_TRC_RSS_CONTROL_A :
874 MPS_T5_TRC_RSS_CONTROL_A,
875 RSSCONTROL_V(netdev2pinfo(adap->port[0])->tx_chan) |
876 QUEUENUMBER_V(s->ethrxq[0].rspq.abs_id));
879 t4_free_sge_resources(adap);
883 static u16 cxgb_select_queue(struct net_device *dev, struct sk_buff *skb,
884 void *accel_priv, select_queue_fallback_t fallback)
888 #ifdef CONFIG_CHELSIO_T4_DCB
889 /* If a Data Center Bridging has been successfully negotiated on this
890 * link then we'll use the skb's priority to map it to a TX Queue.
891 * The skb's priority is determined via the VLAN Tag Priority Code
894 if (cxgb4_dcb_enabled(dev) && !is_kdump_kernel()) {
898 err = vlan_get_tag(skb, &vlan_tci);
902 "TX Packet without VLAN Tag on DCB Link\n");
905 txq = (vlan_tci & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
906 #ifdef CONFIG_CHELSIO_T4_FCOE
907 if (skb->protocol == htons(ETH_P_FCOE))
908 txq = skb->priority & 0x7;
909 #endif /* CONFIG_CHELSIO_T4_FCOE */
913 #endif /* CONFIG_CHELSIO_T4_DCB */
916 txq = (skb_rx_queue_recorded(skb)
917 ? skb_get_rx_queue(skb)
918 : smp_processor_id());
920 while (unlikely(txq >= dev->real_num_tx_queues))
921 txq -= dev->real_num_tx_queues;
926 return fallback(dev, skb) % dev->real_num_tx_queues;
929 static int closest_timer(const struct sge *s, int time)
931 int i, delta, match = 0, min_delta = INT_MAX;
933 for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
934 delta = time - s->timer_val[i];
937 if (delta < min_delta) {
945 static int closest_thres(const struct sge *s, int thres)
947 int i, delta, match = 0, min_delta = INT_MAX;
949 for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
950 delta = thres - s->counter_val[i];
953 if (delta < min_delta) {
962 * cxgb4_set_rspq_intr_params - set a queue's interrupt holdoff parameters
964 * @us: the hold-off time in us, or 0 to disable timer
965 * @cnt: the hold-off packet count, or 0 to disable counter
967 * Sets an Rx queue's interrupt hold-off time and packet count. At least
968 * one of the two needs to be enabled for the queue to generate interrupts.
970 int cxgb4_set_rspq_intr_params(struct sge_rspq *q,
971 unsigned int us, unsigned int cnt)
973 struct adapter *adap = q->adap;
982 new_idx = closest_thres(&adap->sge, cnt);
983 if (q->desc && q->pktcnt_idx != new_idx) {
984 /* the queue has already been created, update it */
985 v = FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DMAQ) |
987 FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
988 FW_PARAMS_PARAM_YZ_V(q->cntxt_id);
989 err = t4_set_params(adap, adap->mbox, adap->pf, 0, 1,
994 q->pktcnt_idx = new_idx;
997 us = us == 0 ? 6 : closest_timer(&adap->sge, us);
998 q->intr_params = QINTR_TIMER_IDX_V(us) | QINTR_CNT_EN_V(cnt > 0);
1002 static int cxgb_set_features(struct net_device *dev, netdev_features_t features)
1004 const struct port_info *pi = netdev_priv(dev);
1005 netdev_features_t changed = dev->features ^ features;
1008 if (!(changed & NETIF_F_HW_VLAN_CTAG_RX))
1011 err = t4_set_rxmode(pi->adapter, pi->adapter->pf, pi->viid, -1,
1013 !!(features & NETIF_F_HW_VLAN_CTAG_RX), true);
1015 dev->features = features ^ NETIF_F_HW_VLAN_CTAG_RX;
1019 static int setup_debugfs(struct adapter *adap)
1021 if (IS_ERR_OR_NULL(adap->debugfs_root))
1024 #ifdef CONFIG_DEBUG_FS
1025 t4_setup_debugfs(adap);
1031 * upper-layer driver support
1035 * Allocate an active-open TID and set it to the supplied value.
1037 int cxgb4_alloc_atid(struct tid_info *t, void *data)
1041 spin_lock_bh(&t->atid_lock);
1043 union aopen_entry *p = t->afree;
1045 atid = (p - t->atid_tab) + t->atid_base;
1050 spin_unlock_bh(&t->atid_lock);
1053 EXPORT_SYMBOL(cxgb4_alloc_atid);
1056 * Release an active-open TID.
1058 void cxgb4_free_atid(struct tid_info *t, unsigned int atid)
1060 union aopen_entry *p = &t->atid_tab[atid - t->atid_base];
1062 spin_lock_bh(&t->atid_lock);
1066 spin_unlock_bh(&t->atid_lock);
1068 EXPORT_SYMBOL(cxgb4_free_atid);
1071 * Allocate a server TID and set it to the supplied value.
1073 int cxgb4_alloc_stid(struct tid_info *t, int family, void *data)
1077 spin_lock_bh(&t->stid_lock);
1078 if (family == PF_INET) {
1079 stid = find_first_zero_bit(t->stid_bmap, t->nstids);
1080 if (stid < t->nstids)
1081 __set_bit(stid, t->stid_bmap);
1085 stid = bitmap_find_free_region(t->stid_bmap, t->nstids, 1);
1090 t->stid_tab[stid].data = data;
1091 stid += t->stid_base;
1092 /* IPv6 requires max of 520 bits or 16 cells in TCAM
1093 * This is equivalent to 4 TIDs. With CLIP enabled it
1096 if (family == PF_INET6) {
1097 t->stids_in_use += 2;
1098 t->v6_stids_in_use += 2;
1103 spin_unlock_bh(&t->stid_lock);
1106 EXPORT_SYMBOL(cxgb4_alloc_stid);
1108 /* Allocate a server filter TID and set it to the supplied value.
1110 int cxgb4_alloc_sftid(struct tid_info *t, int family, void *data)
1114 spin_lock_bh(&t->stid_lock);
1115 if (family == PF_INET) {
1116 stid = find_next_zero_bit(t->stid_bmap,
1117 t->nstids + t->nsftids, t->nstids);
1118 if (stid < (t->nstids + t->nsftids))
1119 __set_bit(stid, t->stid_bmap);
1126 t->stid_tab[stid].data = data;
1128 stid += t->sftid_base;
1131 spin_unlock_bh(&t->stid_lock);
1134 EXPORT_SYMBOL(cxgb4_alloc_sftid);
1136 /* Release a server TID.
1138 void cxgb4_free_stid(struct tid_info *t, unsigned int stid, int family)
1140 /* Is it a server filter TID? */
1141 if (t->nsftids && (stid >= t->sftid_base)) {
1142 stid -= t->sftid_base;
1145 stid -= t->stid_base;
1148 spin_lock_bh(&t->stid_lock);
1149 if (family == PF_INET)
1150 __clear_bit(stid, t->stid_bmap);
1152 bitmap_release_region(t->stid_bmap, stid, 1);
1153 t->stid_tab[stid].data = NULL;
1154 if (stid < t->nstids) {
1155 if (family == PF_INET6) {
1156 t->stids_in_use -= 2;
1157 t->v6_stids_in_use -= 2;
1165 spin_unlock_bh(&t->stid_lock);
1167 EXPORT_SYMBOL(cxgb4_free_stid);
1170 * Populate a TID_RELEASE WR. Caller must properly size the skb.
1172 static void mk_tid_release(struct sk_buff *skb, unsigned int chan,
1175 struct cpl_tid_release *req;
1177 set_wr_txq(skb, CPL_PRIORITY_SETUP, chan);
1178 req = __skb_put(skb, sizeof(*req));
1179 INIT_TP_WR(req, tid);
1180 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, tid));
1184 * Queue a TID release request and if necessary schedule a work queue to
1187 static void cxgb4_queue_tid_release(struct tid_info *t, unsigned int chan,
1190 void **p = &t->tid_tab[tid];
1191 struct adapter *adap = container_of(t, struct adapter, tids);
1193 spin_lock_bh(&adap->tid_release_lock);
1194 *p = adap->tid_release_head;
1195 /* Low 2 bits encode the Tx channel number */
1196 adap->tid_release_head = (void **)((uintptr_t)p | chan);
1197 if (!adap->tid_release_task_busy) {
1198 adap->tid_release_task_busy = true;
1199 queue_work(adap->workq, &adap->tid_release_task);
1201 spin_unlock_bh(&adap->tid_release_lock);
1205 * Process the list of pending TID release requests.
1207 static void process_tid_release_list(struct work_struct *work)
1209 struct sk_buff *skb;
1210 struct adapter *adap;
1212 adap = container_of(work, struct adapter, tid_release_task);
1214 spin_lock_bh(&adap->tid_release_lock);
1215 while (adap->tid_release_head) {
1216 void **p = adap->tid_release_head;
1217 unsigned int chan = (uintptr_t)p & 3;
1218 p = (void *)p - chan;
1220 adap->tid_release_head = *p;
1222 spin_unlock_bh(&adap->tid_release_lock);
1224 while (!(skb = alloc_skb(sizeof(struct cpl_tid_release),
1226 schedule_timeout_uninterruptible(1);
1228 mk_tid_release(skb, chan, p - adap->tids.tid_tab);
1229 t4_ofld_send(adap, skb);
1230 spin_lock_bh(&adap->tid_release_lock);
1232 adap->tid_release_task_busy = false;
1233 spin_unlock_bh(&adap->tid_release_lock);
1237 * Release a TID and inform HW. If we are unable to allocate the release
1238 * message we defer to a work queue.
1240 void cxgb4_remove_tid(struct tid_info *t, unsigned int chan, unsigned int tid,
1241 unsigned short family)
1243 struct sk_buff *skb;
1244 struct adapter *adap = container_of(t, struct adapter, tids);
1246 WARN_ON(tid >= t->ntids);
1248 if (t->tid_tab[tid]) {
1249 t->tid_tab[tid] = NULL;
1250 atomic_dec(&t->conns_in_use);
1251 if (t->hash_base && (tid >= t->hash_base)) {
1252 if (family == AF_INET6)
1253 atomic_sub(2, &t->hash_tids_in_use);
1255 atomic_dec(&t->hash_tids_in_use);
1257 if (family == AF_INET6)
1258 atomic_sub(2, &t->tids_in_use);
1260 atomic_dec(&t->tids_in_use);
1264 skb = alloc_skb(sizeof(struct cpl_tid_release), GFP_ATOMIC);
1266 mk_tid_release(skb, chan, tid);
1267 t4_ofld_send(adap, skb);
1269 cxgb4_queue_tid_release(t, chan, tid);
1271 EXPORT_SYMBOL(cxgb4_remove_tid);
1274 * Allocate and initialize the TID tables. Returns 0 on success.
1276 static int tid_init(struct tid_info *t)
1278 struct adapter *adap = container_of(t, struct adapter, tids);
1279 unsigned int max_ftids = t->nftids + t->nsftids;
1280 unsigned int natids = t->natids;
1281 unsigned int stid_bmap_size;
1282 unsigned int ftid_bmap_size;
1285 stid_bmap_size = BITS_TO_LONGS(t->nstids + t->nsftids);
1286 ftid_bmap_size = BITS_TO_LONGS(t->nftids);
1287 size = t->ntids * sizeof(*t->tid_tab) +
1288 natids * sizeof(*t->atid_tab) +
1289 t->nstids * sizeof(*t->stid_tab) +
1290 t->nsftids * sizeof(*t->stid_tab) +
1291 stid_bmap_size * sizeof(long) +
1292 max_ftids * sizeof(*t->ftid_tab) +
1293 ftid_bmap_size * sizeof(long);
1295 t->tid_tab = kvzalloc(size, GFP_KERNEL);
1299 t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
1300 t->stid_tab = (struct serv_entry *)&t->atid_tab[natids];
1301 t->stid_bmap = (unsigned long *)&t->stid_tab[t->nstids + t->nsftids];
1302 t->ftid_tab = (struct filter_entry *)&t->stid_bmap[stid_bmap_size];
1303 t->ftid_bmap = (unsigned long *)&t->ftid_tab[max_ftids];
1304 spin_lock_init(&t->stid_lock);
1305 spin_lock_init(&t->atid_lock);
1306 spin_lock_init(&t->ftid_lock);
1308 t->stids_in_use = 0;
1309 t->v6_stids_in_use = 0;
1310 t->sftids_in_use = 0;
1312 t->atids_in_use = 0;
1313 atomic_set(&t->tids_in_use, 0);
1314 atomic_set(&t->conns_in_use, 0);
1315 atomic_set(&t->hash_tids_in_use, 0);
1317 /* Setup the free list for atid_tab and clear the stid bitmap. */
1320 t->atid_tab[natids - 1].next = &t->atid_tab[natids];
1321 t->afree = t->atid_tab;
1324 if (is_offload(adap)) {
1325 bitmap_zero(t->stid_bmap, t->nstids + t->nsftids);
1326 /* Reserve stid 0 for T4/T5 adapters */
1327 if (!t->stid_base &&
1328 CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5)
1329 __set_bit(0, t->stid_bmap);
1332 bitmap_zero(t->ftid_bmap, t->nftids);
1337 * cxgb4_create_server - create an IP server
1339 * @stid: the server TID
1340 * @sip: local IP address to bind server to
1341 * @sport: the server's TCP port
1342 * @queue: queue to direct messages from this server to
1344 * Create an IP server for the given port and address.
1345 * Returns <0 on error and one of the %NET_XMIT_* values on success.
1347 int cxgb4_create_server(const struct net_device *dev, unsigned int stid,
1348 __be32 sip, __be16 sport, __be16 vlan,
1352 struct sk_buff *skb;
1353 struct adapter *adap;
1354 struct cpl_pass_open_req *req;
1357 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
1361 adap = netdev2adap(dev);
1362 req = __skb_put(skb, sizeof(*req));
1364 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, stid));
1365 req->local_port = sport;
1366 req->peer_port = htons(0);
1367 req->local_ip = sip;
1368 req->peer_ip = htonl(0);
1369 chan = rxq_to_chan(&adap->sge, queue);
1370 req->opt0 = cpu_to_be64(TX_CHAN_V(chan));
1371 req->opt1 = cpu_to_be64(CONN_POLICY_V(CPL_CONN_POLICY_ASK) |
1372 SYN_RSS_ENABLE_F | SYN_RSS_QUEUE_V(queue));
1373 ret = t4_mgmt_tx(adap, skb);
1374 return net_xmit_eval(ret);
1376 EXPORT_SYMBOL(cxgb4_create_server);
1378 /* cxgb4_create_server6 - create an IPv6 server
1380 * @stid: the server TID
1381 * @sip: local IPv6 address to bind server to
1382 * @sport: the server's TCP port
1383 * @queue: queue to direct messages from this server to
1385 * Create an IPv6 server for the given port and address.
1386 * Returns <0 on error and one of the %NET_XMIT_* values on success.
1388 int cxgb4_create_server6(const struct net_device *dev, unsigned int stid,
1389 const struct in6_addr *sip, __be16 sport,
1393 struct sk_buff *skb;
1394 struct adapter *adap;
1395 struct cpl_pass_open_req6 *req;
1398 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
1402 adap = netdev2adap(dev);
1403 req = __skb_put(skb, sizeof(*req));
1405 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ6, stid));
1406 req->local_port = sport;
1407 req->peer_port = htons(0);
1408 req->local_ip_hi = *(__be64 *)(sip->s6_addr);
1409 req->local_ip_lo = *(__be64 *)(sip->s6_addr + 8);
1410 req->peer_ip_hi = cpu_to_be64(0);
1411 req->peer_ip_lo = cpu_to_be64(0);
1412 chan = rxq_to_chan(&adap->sge, queue);
1413 req->opt0 = cpu_to_be64(TX_CHAN_V(chan));
1414 req->opt1 = cpu_to_be64(CONN_POLICY_V(CPL_CONN_POLICY_ASK) |
1415 SYN_RSS_ENABLE_F | SYN_RSS_QUEUE_V(queue));
1416 ret = t4_mgmt_tx(adap, skb);
1417 return net_xmit_eval(ret);
1419 EXPORT_SYMBOL(cxgb4_create_server6);
1421 int cxgb4_remove_server(const struct net_device *dev, unsigned int stid,
1422 unsigned int queue, bool ipv6)
1424 struct sk_buff *skb;
1425 struct adapter *adap;
1426 struct cpl_close_listsvr_req *req;
1429 adap = netdev2adap(dev);
1431 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
1435 req = __skb_put(skb, sizeof(*req));
1437 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, stid));
1438 req->reply_ctrl = htons(NO_REPLY_V(0) | (ipv6 ? LISTSVR_IPV6_V(1) :
1439 LISTSVR_IPV6_V(0)) | QUEUENO_V(queue));
1440 ret = t4_mgmt_tx(adap, skb);
1441 return net_xmit_eval(ret);
1443 EXPORT_SYMBOL(cxgb4_remove_server);
1446 * cxgb4_best_mtu - find the entry in the MTU table closest to an MTU
1447 * @mtus: the HW MTU table
1448 * @mtu: the target MTU
1449 * @idx: index of selected entry in the MTU table
1451 * Returns the index and the value in the HW MTU table that is closest to
1452 * but does not exceed @mtu, unless @mtu is smaller than any value in the
1453 * table, in which case that smallest available value is selected.
1455 unsigned int cxgb4_best_mtu(const unsigned short *mtus, unsigned short mtu,
1460 while (i < NMTUS - 1 && mtus[i + 1] <= mtu)
1466 EXPORT_SYMBOL(cxgb4_best_mtu);
1469 * cxgb4_best_aligned_mtu - find best MTU, [hopefully] data size aligned
1470 * @mtus: the HW MTU table
1471 * @header_size: Header Size
1472 * @data_size_max: maximum Data Segment Size
1473 * @data_size_align: desired Data Segment Size Alignment (2^N)
1474 * @mtu_idxp: HW MTU Table Index return value pointer (possibly NULL)
1476 * Similar to cxgb4_best_mtu() but instead of searching the Hardware
1477 * MTU Table based solely on a Maximum MTU parameter, we break that
1478 * parameter up into a Header Size and Maximum Data Segment Size, and
1479 * provide a desired Data Segment Size Alignment. If we find an MTU in
1480 * the Hardware MTU Table which will result in a Data Segment Size with
1481 * the requested alignment _and_ that MTU isn't "too far" from the
1482 * closest MTU, then we'll return that rather than the closest MTU.
1484 unsigned int cxgb4_best_aligned_mtu(const unsigned short *mtus,
1485 unsigned short header_size,
1486 unsigned short data_size_max,
1487 unsigned short data_size_align,
1488 unsigned int *mtu_idxp)
1490 unsigned short max_mtu = header_size + data_size_max;
1491 unsigned short data_size_align_mask = data_size_align - 1;
1492 int mtu_idx, aligned_mtu_idx;
1494 /* Scan the MTU Table till we find an MTU which is larger than our
1495 * Maximum MTU or we reach the end of the table. Along the way,
1496 * record the last MTU found, if any, which will result in a Data
1497 * Segment Length matching the requested alignment.
1499 for (mtu_idx = 0, aligned_mtu_idx = -1; mtu_idx < NMTUS; mtu_idx++) {
1500 unsigned short data_size = mtus[mtu_idx] - header_size;
1502 /* If this MTU minus the Header Size would result in a
1503 * Data Segment Size of the desired alignment, remember it.
1505 if ((data_size & data_size_align_mask) == 0)
1506 aligned_mtu_idx = mtu_idx;
1508 /* If we're not at the end of the Hardware MTU Table and the
1509 * next element is larger than our Maximum MTU, drop out of
1512 if (mtu_idx+1 < NMTUS && mtus[mtu_idx+1] > max_mtu)
1516 /* If we fell out of the loop because we ran to the end of the table,
1517 * then we just have to use the last [largest] entry.
1519 if (mtu_idx == NMTUS)
1522 /* If we found an MTU which resulted in the requested Data Segment
1523 * Length alignment and that's "not far" from the largest MTU which is
1524 * less than or equal to the maximum MTU, then use that.
1526 if (aligned_mtu_idx >= 0 &&
1527 mtu_idx - aligned_mtu_idx <= 1)
1528 mtu_idx = aligned_mtu_idx;
1530 /* If the caller has passed in an MTU Index pointer, pass the
1531 * MTU Index back. Return the MTU value.
1534 *mtu_idxp = mtu_idx;
1535 return mtus[mtu_idx];
1537 EXPORT_SYMBOL(cxgb4_best_aligned_mtu);
1540 * cxgb4_tp_smt_idx - Get the Source Mac Table index for this VI
1542 * @viid: VI id of the given port
1544 * Return the SMT index for this VI.
1546 unsigned int cxgb4_tp_smt_idx(enum chip_type chip, unsigned int viid)
1548 /* In T4/T5, SMT contains 256 SMAC entries organized in
1549 * 128 rows of 2 entries each.
1550 * In T6, SMT contains 256 SMAC entries in 256 rows.
1551 * TODO: The below code needs to be updated when we add support
1554 if (CHELSIO_CHIP_VERSION(chip) <= CHELSIO_T5)
1555 return ((viid & 0x7f) << 1);
1557 return (viid & 0x7f);
1559 EXPORT_SYMBOL(cxgb4_tp_smt_idx);
1562 * cxgb4_port_chan - get the HW channel of a port
1563 * @dev: the net device for the port
1565 * Return the HW Tx channel of the given port.
1567 unsigned int cxgb4_port_chan(const struct net_device *dev)
1569 return netdev2pinfo(dev)->tx_chan;
1571 EXPORT_SYMBOL(cxgb4_port_chan);
1573 unsigned int cxgb4_dbfifo_count(const struct net_device *dev, int lpfifo)
1575 struct adapter *adap = netdev2adap(dev);
1576 u32 v1, v2, lp_count, hp_count;
1578 v1 = t4_read_reg(adap, SGE_DBFIFO_STATUS_A);
1579 v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2_A);
1580 if (is_t4(adap->params.chip)) {
1581 lp_count = LP_COUNT_G(v1);
1582 hp_count = HP_COUNT_G(v1);
1584 lp_count = LP_COUNT_T5_G(v1);
1585 hp_count = HP_COUNT_T5_G(v2);
1587 return lpfifo ? lp_count : hp_count;
1589 EXPORT_SYMBOL(cxgb4_dbfifo_count);
1592 * cxgb4_port_viid - get the VI id of a port
1593 * @dev: the net device for the port
1595 * Return the VI id of the given port.
1597 unsigned int cxgb4_port_viid(const struct net_device *dev)
1599 return netdev2pinfo(dev)->viid;
1601 EXPORT_SYMBOL(cxgb4_port_viid);
1604 * cxgb4_port_idx - get the index of a port
1605 * @dev: the net device for the port
1607 * Return the index of the given port.
1609 unsigned int cxgb4_port_idx(const struct net_device *dev)
1611 return netdev2pinfo(dev)->port_id;
1613 EXPORT_SYMBOL(cxgb4_port_idx);
1615 void cxgb4_get_tcp_stats(struct pci_dev *pdev, struct tp_tcp_stats *v4,
1616 struct tp_tcp_stats *v6)
1618 struct adapter *adap = pci_get_drvdata(pdev);
1620 spin_lock(&adap->stats_lock);
1621 t4_tp_get_tcp_stats(adap, v4, v6);
1622 spin_unlock(&adap->stats_lock);
1624 EXPORT_SYMBOL(cxgb4_get_tcp_stats);
1626 void cxgb4_iscsi_init(struct net_device *dev, unsigned int tag_mask,
1627 const unsigned int *pgsz_order)
1629 struct adapter *adap = netdev2adap(dev);
1631 t4_write_reg(adap, ULP_RX_ISCSI_TAGMASK_A, tag_mask);
1632 t4_write_reg(adap, ULP_RX_ISCSI_PSZ_A, HPZ0_V(pgsz_order[0]) |
1633 HPZ1_V(pgsz_order[1]) | HPZ2_V(pgsz_order[2]) |
1634 HPZ3_V(pgsz_order[3]));
1636 EXPORT_SYMBOL(cxgb4_iscsi_init);
1638 int cxgb4_flush_eq_cache(struct net_device *dev)
1640 struct adapter *adap = netdev2adap(dev);
1642 return t4_sge_ctxt_flush(adap, adap->mbox);
1644 EXPORT_SYMBOL(cxgb4_flush_eq_cache);
1646 static int read_eq_indices(struct adapter *adap, u16 qid, u16 *pidx, u16 *cidx)
1648 u32 addr = t4_read_reg(adap, SGE_DBQ_CTXT_BADDR_A) + 24 * qid + 8;
1652 spin_lock(&adap->win0_lock);
1653 ret = t4_memory_rw(adap, 0, MEM_EDC0, addr,
1654 sizeof(indices), (__be32 *)&indices,
1656 spin_unlock(&adap->win0_lock);
1658 *cidx = (be64_to_cpu(indices) >> 25) & 0xffff;
1659 *pidx = (be64_to_cpu(indices) >> 9) & 0xffff;
1664 int cxgb4_sync_txq_pidx(struct net_device *dev, u16 qid, u16 pidx,
1667 struct adapter *adap = netdev2adap(dev);
1668 u16 hw_pidx, hw_cidx;
1671 ret = read_eq_indices(adap, qid, &hw_pidx, &hw_cidx);
1675 if (pidx != hw_pidx) {
1679 if (pidx >= hw_pidx)
1680 delta = pidx - hw_pidx;
1682 delta = size - hw_pidx + pidx;
1684 if (is_t4(adap->params.chip))
1685 val = PIDX_V(delta);
1687 val = PIDX_T5_V(delta);
1689 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
1695 EXPORT_SYMBOL(cxgb4_sync_txq_pidx);
1697 int cxgb4_read_tpte(struct net_device *dev, u32 stag, __be32 *tpte)
1699 struct adapter *adap;
1700 u32 offset, memtype, memaddr;
1701 u32 edc0_size, edc1_size, mc0_size, mc1_size, size;
1702 u32 edc0_end, edc1_end, mc0_end, mc1_end;
1705 adap = netdev2adap(dev);
1707 offset = ((stag >> 8) * 32) + adap->vres.stag.start;
1709 /* Figure out where the offset lands in the Memory Type/Address scheme.
1710 * This code assumes that the memory is laid out starting at offset 0
1711 * with no breaks as: EDC0, EDC1, MC0, MC1. All cards have both EDC0
1712 * and EDC1. Some cards will have neither MC0 nor MC1, most cards have
1713 * MC0, and some have both MC0 and MC1.
1715 size = t4_read_reg(adap, MA_EDRAM0_BAR_A);
1716 edc0_size = EDRAM0_SIZE_G(size) << 20;
1717 size = t4_read_reg(adap, MA_EDRAM1_BAR_A);
1718 edc1_size = EDRAM1_SIZE_G(size) << 20;
1719 size = t4_read_reg(adap, MA_EXT_MEMORY0_BAR_A);
1720 mc0_size = EXT_MEM0_SIZE_G(size) << 20;
1722 edc0_end = edc0_size;
1723 edc1_end = edc0_end + edc1_size;
1724 mc0_end = edc1_end + mc0_size;
1726 if (offset < edc0_end) {
1729 } else if (offset < edc1_end) {
1731 memaddr = offset - edc0_end;
1733 if (offset < mc0_end) {
1735 memaddr = offset - edc1_end;
1736 } else if (is_t5(adap->params.chip)) {
1737 size = t4_read_reg(adap, MA_EXT_MEMORY1_BAR_A);
1738 mc1_size = EXT_MEM1_SIZE_G(size) << 20;
1739 mc1_end = mc0_end + mc1_size;
1740 if (offset < mc1_end) {
1742 memaddr = offset - mc0_end;
1744 /* offset beyond the end of any memory */
1748 /* T4/T6 only has a single memory channel */
1753 spin_lock(&adap->win0_lock);
1754 ret = t4_memory_rw(adap, 0, memtype, memaddr, 32, tpte, T4_MEMORY_READ);
1755 spin_unlock(&adap->win0_lock);
1759 dev_err(adap->pdev_dev, "stag %#x, offset %#x out of range\n",
1763 EXPORT_SYMBOL(cxgb4_read_tpte);
1765 u64 cxgb4_read_sge_timestamp(struct net_device *dev)
1768 struct adapter *adap;
1770 adap = netdev2adap(dev);
1771 lo = t4_read_reg(adap, SGE_TIMESTAMP_LO_A);
1772 hi = TSVAL_G(t4_read_reg(adap, SGE_TIMESTAMP_HI_A));
1774 return ((u64)hi << 32) | (u64)lo;
1776 EXPORT_SYMBOL(cxgb4_read_sge_timestamp);
1778 int cxgb4_bar2_sge_qregs(struct net_device *dev,
1780 enum cxgb4_bar2_qtype qtype,
1783 unsigned int *pbar2_qid)
1785 return t4_bar2_sge_qregs(netdev2adap(dev),
1787 (qtype == CXGB4_BAR2_QTYPE_EGRESS
1788 ? T4_BAR2_QTYPE_EGRESS
1789 : T4_BAR2_QTYPE_INGRESS),
1794 EXPORT_SYMBOL(cxgb4_bar2_sge_qregs);
1796 static struct pci_driver cxgb4_driver;
1798 static void check_neigh_update(struct neighbour *neigh)
1800 const struct device *parent;
1801 const struct net_device *netdev = neigh->dev;
1803 if (is_vlan_dev(netdev))
1804 netdev = vlan_dev_real_dev(netdev);
1805 parent = netdev->dev.parent;
1806 if (parent && parent->driver == &cxgb4_driver.driver)
1807 t4_l2t_update(dev_get_drvdata(parent), neigh);
1810 static int netevent_cb(struct notifier_block *nb, unsigned long event,
1814 case NETEVENT_NEIGH_UPDATE:
1815 check_neigh_update(data);
1817 case NETEVENT_REDIRECT:
1824 static bool netevent_registered;
1825 static struct notifier_block cxgb4_netevent_nb = {
1826 .notifier_call = netevent_cb
1829 static void drain_db_fifo(struct adapter *adap, int usecs)
1831 u32 v1, v2, lp_count, hp_count;
1834 v1 = t4_read_reg(adap, SGE_DBFIFO_STATUS_A);
1835 v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2_A);
1836 if (is_t4(adap->params.chip)) {
1837 lp_count = LP_COUNT_G(v1);
1838 hp_count = HP_COUNT_G(v1);
1840 lp_count = LP_COUNT_T5_G(v1);
1841 hp_count = HP_COUNT_T5_G(v2);
1844 if (lp_count == 0 && hp_count == 0)
1846 set_current_state(TASK_UNINTERRUPTIBLE);
1847 schedule_timeout(usecs_to_jiffies(usecs));
1851 static void disable_txq_db(struct sge_txq *q)
1853 unsigned long flags;
1855 spin_lock_irqsave(&q->db_lock, flags);
1857 spin_unlock_irqrestore(&q->db_lock, flags);
1860 static void enable_txq_db(struct adapter *adap, struct sge_txq *q)
1862 spin_lock_irq(&q->db_lock);
1863 if (q->db_pidx_inc) {
1864 /* Make sure that all writes to the TX descriptors
1865 * are committed before we tell HW about them.
1868 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
1869 QID_V(q->cntxt_id) | PIDX_V(q->db_pidx_inc));
1873 spin_unlock_irq(&q->db_lock);
1876 static void disable_dbs(struct adapter *adap)
1880 for_each_ethrxq(&adap->sge, i)
1881 disable_txq_db(&adap->sge.ethtxq[i].q);
1882 if (is_offload(adap)) {
1883 struct sge_uld_txq_info *txq_info =
1884 adap->sge.uld_txq_info[CXGB4_TX_OFLD];
1887 for_each_ofldtxq(&adap->sge, i) {
1888 struct sge_uld_txq *txq = &txq_info->uldtxq[i];
1890 disable_txq_db(&txq->q);
1894 for_each_port(adap, i)
1895 disable_txq_db(&adap->sge.ctrlq[i].q);
1898 static void enable_dbs(struct adapter *adap)
1902 for_each_ethrxq(&adap->sge, i)
1903 enable_txq_db(adap, &adap->sge.ethtxq[i].q);
1904 if (is_offload(adap)) {
1905 struct sge_uld_txq_info *txq_info =
1906 adap->sge.uld_txq_info[CXGB4_TX_OFLD];
1909 for_each_ofldtxq(&adap->sge, i) {
1910 struct sge_uld_txq *txq = &txq_info->uldtxq[i];
1912 enable_txq_db(adap, &txq->q);
1916 for_each_port(adap, i)
1917 enable_txq_db(adap, &adap->sge.ctrlq[i].q);
1920 static void notify_rdma_uld(struct adapter *adap, enum cxgb4_control cmd)
1922 enum cxgb4_uld type = CXGB4_ULD_RDMA;
1924 if (adap->uld && adap->uld[type].handle)
1925 adap->uld[type].control(adap->uld[type].handle, cmd);
1928 static void process_db_full(struct work_struct *work)
1930 struct adapter *adap;
1932 adap = container_of(work, struct adapter, db_full_task);
1934 drain_db_fifo(adap, dbfifo_drain_delay);
1936 notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
1937 if (CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5)
1938 t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
1939 DBFIFO_HP_INT_F | DBFIFO_LP_INT_F,
1940 DBFIFO_HP_INT_F | DBFIFO_LP_INT_F);
1942 t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
1943 DBFIFO_LP_INT_F, DBFIFO_LP_INT_F);
1946 static void sync_txq_pidx(struct adapter *adap, struct sge_txq *q)
1948 u16 hw_pidx, hw_cidx;
1951 spin_lock_irq(&q->db_lock);
1952 ret = read_eq_indices(adap, (u16)q->cntxt_id, &hw_pidx, &hw_cidx);
1955 if (q->db_pidx != hw_pidx) {
1959 if (q->db_pidx >= hw_pidx)
1960 delta = q->db_pidx - hw_pidx;
1962 delta = q->size - hw_pidx + q->db_pidx;
1964 if (is_t4(adap->params.chip))
1965 val = PIDX_V(delta);
1967 val = PIDX_T5_V(delta);
1969 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
1970 QID_V(q->cntxt_id) | val);
1975 spin_unlock_irq(&q->db_lock);
1977 CH_WARN(adap, "DB drop recovery failed.\n");
1980 static void recover_all_queues(struct adapter *adap)
1984 for_each_ethrxq(&adap->sge, i)
1985 sync_txq_pidx(adap, &adap->sge.ethtxq[i].q);
1986 if (is_offload(adap)) {
1987 struct sge_uld_txq_info *txq_info =
1988 adap->sge.uld_txq_info[CXGB4_TX_OFLD];
1990 for_each_ofldtxq(&adap->sge, i) {
1991 struct sge_uld_txq *txq = &txq_info->uldtxq[i];
1993 sync_txq_pidx(adap, &txq->q);
1997 for_each_port(adap, i)
1998 sync_txq_pidx(adap, &adap->sge.ctrlq[i].q);
2001 static void process_db_drop(struct work_struct *work)
2003 struct adapter *adap;
2005 adap = container_of(work, struct adapter, db_drop_task);
2007 if (is_t4(adap->params.chip)) {
2008 drain_db_fifo(adap, dbfifo_drain_delay);
2009 notify_rdma_uld(adap, CXGB4_CONTROL_DB_DROP);
2010 drain_db_fifo(adap, dbfifo_drain_delay);
2011 recover_all_queues(adap);
2012 drain_db_fifo(adap, dbfifo_drain_delay);
2014 notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
2015 } else if (is_t5(adap->params.chip)) {
2016 u32 dropped_db = t4_read_reg(adap, 0x010ac);
2017 u16 qid = (dropped_db >> 15) & 0x1ffff;
2018 u16 pidx_inc = dropped_db & 0x1fff;
2020 unsigned int bar2_qid;
2023 ret = t4_bar2_sge_qregs(adap, qid, T4_BAR2_QTYPE_EGRESS,
2024 0, &bar2_qoffset, &bar2_qid);
2026 dev_err(adap->pdev_dev, "doorbell drop recovery: "
2027 "qid=%d, pidx_inc=%d\n", qid, pidx_inc);
2029 writel(PIDX_T5_V(pidx_inc) | QID_V(bar2_qid),
2030 adap->bar2 + bar2_qoffset + SGE_UDB_KDOORBELL);
2032 /* Re-enable BAR2 WC */
2033 t4_set_reg_field(adap, 0x10b0, 1<<15, 1<<15);
2036 if (CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5)
2037 t4_set_reg_field(adap, SGE_DOORBELL_CONTROL_A, DROPPED_DB_F, 0);
2040 void t4_db_full(struct adapter *adap)
2042 if (is_t4(adap->params.chip)) {
2044 notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
2045 t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
2046 DBFIFO_HP_INT_F | DBFIFO_LP_INT_F, 0);
2047 queue_work(adap->workq, &adap->db_full_task);
2051 void t4_db_dropped(struct adapter *adap)
2053 if (is_t4(adap->params.chip)) {
2055 notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
2057 queue_work(adap->workq, &adap->db_drop_task);
2060 void t4_register_netevent_notifier(void)
2062 if (!netevent_registered) {
2063 register_netevent_notifier(&cxgb4_netevent_nb);
2064 netevent_registered = true;
2068 static void detach_ulds(struct adapter *adap)
2072 mutex_lock(&uld_mutex);
2073 list_del(&adap->list_node);
2074 for (i = 0; i < CXGB4_ULD_MAX; i++)
2075 if (adap->uld && adap->uld[i].handle) {
2076 adap->uld[i].state_change(adap->uld[i].handle,
2077 CXGB4_STATE_DETACH);
2078 adap->uld[i].handle = NULL;
2080 if (netevent_registered && list_empty(&adapter_list)) {
2081 unregister_netevent_notifier(&cxgb4_netevent_nb);
2082 netevent_registered = false;
2084 mutex_unlock(&uld_mutex);
2087 static void notify_ulds(struct adapter *adap, enum cxgb4_state new_state)
2091 mutex_lock(&uld_mutex);
2092 for (i = 0; i < CXGB4_ULD_MAX; i++)
2093 if (adap->uld && adap->uld[i].handle)
2094 adap->uld[i].state_change(adap->uld[i].handle,
2096 mutex_unlock(&uld_mutex);
2099 #if IS_ENABLED(CONFIG_IPV6)
2100 static int cxgb4_inet6addr_handler(struct notifier_block *this,
2101 unsigned long event, void *data)
2103 struct inet6_ifaddr *ifa = data;
2104 struct net_device *event_dev = ifa->idev->dev;
2105 const struct device *parent = NULL;
2106 #if IS_ENABLED(CONFIG_BONDING)
2107 struct adapter *adap;
2109 if (is_vlan_dev(event_dev))
2110 event_dev = vlan_dev_real_dev(event_dev);
2111 #if IS_ENABLED(CONFIG_BONDING)
2112 if (event_dev->flags & IFF_MASTER) {
2113 list_for_each_entry(adap, &adapter_list, list_node) {
2116 cxgb4_clip_get(adap->port[0],
2117 (const u32 *)ifa, 1);
2120 cxgb4_clip_release(adap->port[0],
2121 (const u32 *)ifa, 1);
2132 parent = event_dev->dev.parent;
2134 if (parent && parent->driver == &cxgb4_driver.driver) {
2137 cxgb4_clip_get(event_dev, (const u32 *)ifa, 1);
2140 cxgb4_clip_release(event_dev, (const u32 *)ifa, 1);
2149 static bool inet6addr_registered;
2150 static struct notifier_block cxgb4_inet6addr_notifier = {
2151 .notifier_call = cxgb4_inet6addr_handler
2154 static void update_clip(const struct adapter *adap)
2157 struct net_device *dev;
2162 for (i = 0; i < MAX_NPORTS; i++) {
2163 dev = adap->port[i];
2167 ret = cxgb4_update_root_dev_clip(dev);
2174 #endif /* IS_ENABLED(CONFIG_IPV6) */
2177 * cxgb_up - enable the adapter
2178 * @adap: adapter being enabled
2180 * Called when the first port is enabled, this function performs the
2181 * actions necessary to make an adapter operational, such as completing
2182 * the initialization of HW modules, and enabling interrupts.
2184 * Must be called with the rtnl lock held.
2186 static int cxgb_up(struct adapter *adap)
2190 err = setup_sge_queues(adap);
2193 err = setup_rss(adap);
2197 if (adap->flags & USING_MSIX) {
2198 name_msix_vecs(adap);
2199 err = request_irq(adap->msix_info[0].vec, t4_nondata_intr, 0,
2200 adap->msix_info[0].desc, adap);
2203 err = request_msix_queue_irqs(adap);
2205 free_irq(adap->msix_info[0].vec, adap);
2209 err = request_irq(adap->pdev->irq, t4_intr_handler(adap),
2210 (adap->flags & USING_MSI) ? 0 : IRQF_SHARED,
2211 adap->port[0]->name, adap);
2216 mutex_lock(&uld_mutex);
2219 t4_intr_enable(adap);
2220 adap->flags |= FULL_INIT_DONE;
2221 mutex_unlock(&uld_mutex);
2223 notify_ulds(adap, CXGB4_STATE_UP);
2224 #if IS_ENABLED(CONFIG_IPV6)
2227 /* Initialize hash mac addr list*/
2228 INIT_LIST_HEAD(&adap->mac_hlist);
2232 dev_err(adap->pdev_dev, "request_irq failed, err %d\n", err);
2234 t4_free_sge_resources(adap);
2238 static void cxgb_down(struct adapter *adapter)
2240 cancel_work_sync(&adapter->tid_release_task);
2241 cancel_work_sync(&adapter->db_full_task);
2242 cancel_work_sync(&adapter->db_drop_task);
2243 adapter->tid_release_task_busy = false;
2244 adapter->tid_release_head = NULL;
2246 t4_sge_stop(adapter);
2247 t4_free_sge_resources(adapter);
2248 adapter->flags &= ~FULL_INIT_DONE;
2252 * net_device operations
2254 static int cxgb_open(struct net_device *dev)
2257 struct port_info *pi = netdev_priv(dev);
2258 struct adapter *adapter = pi->adapter;
2260 netif_carrier_off(dev);
2262 if (!(adapter->flags & FULL_INIT_DONE)) {
2263 err = cxgb_up(adapter);
2268 /* It's possible that the basic port information could have
2269 * changed since we first read it.
2271 err = t4_update_port_info(pi);
2275 err = link_start(dev);
2277 netif_tx_start_all_queues(dev);
2281 static int cxgb_close(struct net_device *dev)
2283 struct port_info *pi = netdev_priv(dev);
2284 struct adapter *adapter = pi->adapter;
2286 netif_tx_stop_all_queues(dev);
2287 netif_carrier_off(dev);
2288 return t4_enable_vi(adapter, adapter->pf, pi->viid, false, false);
2291 int cxgb4_create_server_filter(const struct net_device *dev, unsigned int stid,
2292 __be32 sip, __be16 sport, __be16 vlan,
2293 unsigned int queue, unsigned char port, unsigned char mask)
2296 struct filter_entry *f;
2297 struct adapter *adap;
2301 adap = netdev2adap(dev);
2303 /* Adjust stid to correct filter index */
2304 stid -= adap->tids.sftid_base;
2305 stid += adap->tids.nftids;
2307 /* Check to make sure the filter requested is writable ...
2309 f = &adap->tids.ftid_tab[stid];
2310 ret = writable_filter(f);
2314 /* Clear out any old resources being used by the filter before
2315 * we start constructing the new filter.
2318 clear_filter(adap, f);
2320 /* Clear out filter specifications */
2321 memset(&f->fs, 0, sizeof(struct ch_filter_specification));
2322 f->fs.val.lport = cpu_to_be16(sport);
2323 f->fs.mask.lport = ~0;
2325 if ((val[0] | val[1] | val[2] | val[3]) != 0) {
2326 for (i = 0; i < 4; i++) {
2327 f->fs.val.lip[i] = val[i];
2328 f->fs.mask.lip[i] = ~0;
2330 if (adap->params.tp.vlan_pri_map & PORT_F) {
2331 f->fs.val.iport = port;
2332 f->fs.mask.iport = mask;
2336 if (adap->params.tp.vlan_pri_map & PROTOCOL_F) {
2337 f->fs.val.proto = IPPROTO_TCP;
2338 f->fs.mask.proto = ~0;
2343 /* Mark filter as locked */
2347 /* Save the actual tid. We need this to get the corresponding
2348 * filter entry structure in filter_rpl.
2350 f->tid = stid + adap->tids.ftid_base;
2351 ret = set_filter_wr(adap, stid);
2353 clear_filter(adap, f);
2359 EXPORT_SYMBOL(cxgb4_create_server_filter);
2361 int cxgb4_remove_server_filter(const struct net_device *dev, unsigned int stid,
2362 unsigned int queue, bool ipv6)
2364 struct filter_entry *f;
2365 struct adapter *adap;
2367 adap = netdev2adap(dev);
2369 /* Adjust stid to correct filter index */
2370 stid -= adap->tids.sftid_base;
2371 stid += adap->tids.nftids;
2373 f = &adap->tids.ftid_tab[stid];
2374 /* Unlock the filter */
2377 return delete_filter(adap, stid);
2379 EXPORT_SYMBOL(cxgb4_remove_server_filter);
2381 static void cxgb_get_stats(struct net_device *dev,
2382 struct rtnl_link_stats64 *ns)
2384 struct port_stats stats;
2385 struct port_info *p = netdev_priv(dev);
2386 struct adapter *adapter = p->adapter;
2388 /* Block retrieving statistics during EEH error
2389 * recovery. Otherwise, the recovery might fail
2390 * and the PCI device will be removed permanently
2392 spin_lock(&adapter->stats_lock);
2393 if (!netif_device_present(dev)) {
2394 spin_unlock(&adapter->stats_lock);
2397 t4_get_port_stats_offset(adapter, p->tx_chan, &stats,
2399 spin_unlock(&adapter->stats_lock);
2401 ns->tx_bytes = stats.tx_octets;
2402 ns->tx_packets = stats.tx_frames;
2403 ns->rx_bytes = stats.rx_octets;
2404 ns->rx_packets = stats.rx_frames;
2405 ns->multicast = stats.rx_mcast_frames;
2407 /* detailed rx_errors */
2408 ns->rx_length_errors = stats.rx_jabber + stats.rx_too_long +
2410 ns->rx_over_errors = 0;
2411 ns->rx_crc_errors = stats.rx_fcs_err;
2412 ns->rx_frame_errors = stats.rx_symbol_err;
2413 ns->rx_dropped = stats.rx_ovflow0 + stats.rx_ovflow1 +
2414 stats.rx_ovflow2 + stats.rx_ovflow3 +
2415 stats.rx_trunc0 + stats.rx_trunc1 +
2416 stats.rx_trunc2 + stats.rx_trunc3;
2417 ns->rx_missed_errors = 0;
2419 /* detailed tx_errors */
2420 ns->tx_aborted_errors = 0;
2421 ns->tx_carrier_errors = 0;
2422 ns->tx_fifo_errors = 0;
2423 ns->tx_heartbeat_errors = 0;
2424 ns->tx_window_errors = 0;
2426 ns->tx_errors = stats.tx_error_frames;
2427 ns->rx_errors = stats.rx_symbol_err + stats.rx_fcs_err +
2428 ns->rx_length_errors + stats.rx_len_err + ns->rx_fifo_errors;
2431 static int cxgb_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
2434 int ret = 0, prtad, devad;
2435 struct port_info *pi = netdev_priv(dev);
2436 struct mii_ioctl_data *data = (struct mii_ioctl_data *)&req->ifr_data;
2440 if (pi->mdio_addr < 0)
2442 data->phy_id = pi->mdio_addr;
2446 if (mdio_phy_id_is_c45(data->phy_id)) {
2447 prtad = mdio_phy_id_prtad(data->phy_id);
2448 devad = mdio_phy_id_devad(data->phy_id);
2449 } else if (data->phy_id < 32) {
2450 prtad = data->phy_id;
2452 data->reg_num &= 0x1f;
2456 mbox = pi->adapter->pf;
2457 if (cmd == SIOCGMIIREG)
2458 ret = t4_mdio_rd(pi->adapter, mbox, prtad, devad,
2459 data->reg_num, &data->val_out);
2461 ret = t4_mdio_wr(pi->adapter, mbox, prtad, devad,
2462 data->reg_num, data->val_in);
2465 return copy_to_user(req->ifr_data, &pi->tstamp_config,
2466 sizeof(pi->tstamp_config)) ?
2469 if (copy_from_user(&pi->tstamp_config, req->ifr_data,
2470 sizeof(pi->tstamp_config)))
2473 switch (pi->tstamp_config.rx_filter) {
2474 case HWTSTAMP_FILTER_NONE:
2475 pi->rxtstamp = false;
2477 case HWTSTAMP_FILTER_ALL:
2478 pi->rxtstamp = true;
2481 pi->tstamp_config.rx_filter = HWTSTAMP_FILTER_NONE;
2485 return copy_to_user(req->ifr_data, &pi->tstamp_config,
2486 sizeof(pi->tstamp_config)) ?
2494 static void cxgb_set_rxmode(struct net_device *dev)
2496 /* unfortunately we can't return errors to the stack */
2497 set_rxmode(dev, -1, false);
2500 static int cxgb_change_mtu(struct net_device *dev, int new_mtu)
2503 struct port_info *pi = netdev_priv(dev);
2505 ret = t4_set_rxmode(pi->adapter, pi->adapter->pf, pi->viid, new_mtu, -1,
2512 #ifdef CONFIG_PCI_IOV
2513 static int dummy_open(struct net_device *dev)
2515 /* Turn carrier off since we don't have to transmit anything on this
2518 netif_carrier_off(dev);
2522 /* Fill MAC address that will be assigned by the FW */
2523 static void fill_vf_station_mac_addr(struct adapter *adap)
2526 u8 hw_addr[ETH_ALEN], macaddr[ETH_ALEN];
2531 err = t4_get_raw_vpd_params(adap, &adap->params.vpd);
2533 na = adap->params.vpd.na;
2534 for (i = 0; i < ETH_ALEN; i++)
2535 hw_addr[i] = (hex2val(na[2 * i + 0]) * 16 +
2536 hex2val(na[2 * i + 1]));
2537 a = (hw_addr[0] << 8) | hw_addr[1];
2538 b = (hw_addr[1] << 8) | hw_addr[2];
2540 a |= 0x0200; /* locally assigned Ethernet MAC address */
2541 a &= ~0x0100; /* not a multicast Ethernet MAC address */
2542 macaddr[0] = a >> 8;
2543 macaddr[1] = a & 0xff;
2545 for (i = 2; i < 5; i++)
2546 macaddr[i] = hw_addr[i + 1];
2548 for (i = 0; i < adap->num_vfs; i++) {
2549 macaddr[5] = adap->pf * 16 + i;
2550 ether_addr_copy(adap->vfinfo[i].vf_mac_addr, macaddr);
2555 static int cxgb_set_vf_mac(struct net_device *dev, int vf, u8 *mac)
2557 struct port_info *pi = netdev_priv(dev);
2558 struct adapter *adap = pi->adapter;
2561 /* verify MAC addr is valid */
2562 if (!is_valid_ether_addr(mac)) {
2563 dev_err(pi->adapter->pdev_dev,
2564 "Invalid Ethernet address %pM for VF %d\n",
2569 dev_info(pi->adapter->pdev_dev,
2570 "Setting MAC %pM on VF %d\n", mac, vf);
2571 ret = t4_set_vf_mac_acl(adap, vf + 1, 1, mac);
2573 ether_addr_copy(adap->vfinfo[vf].vf_mac_addr, mac);
2577 static int cxgb_get_vf_config(struct net_device *dev,
2578 int vf, struct ifla_vf_info *ivi)
2580 struct port_info *pi = netdev_priv(dev);
2581 struct adapter *adap = pi->adapter;
2583 if (vf >= adap->num_vfs)
2586 ivi->max_tx_rate = adap->vfinfo[vf].tx_rate;
2587 ivi->min_tx_rate = 0;
2588 ether_addr_copy(ivi->mac, adap->vfinfo[vf].vf_mac_addr);
2592 static int cxgb_get_phys_port_id(struct net_device *dev,
2593 struct netdev_phys_item_id *ppid)
2595 struct port_info *pi = netdev_priv(dev);
2596 unsigned int phy_port_id;
2598 phy_port_id = pi->adapter->adap_idx * 10 + pi->port_id;
2599 ppid->id_len = sizeof(phy_port_id);
2600 memcpy(ppid->id, &phy_port_id, ppid->id_len);
2604 static int cxgb_set_vf_rate(struct net_device *dev, int vf, int min_tx_rate,
2607 struct port_info *pi = netdev_priv(dev);
2608 struct adapter *adap = pi->adapter;
2609 struct fw_port_cmd port_cmd, port_rpl;
2610 u32 link_status, speed = 0;
2611 u32 fw_pfvf, fw_class;
2616 if (vf >= adap->num_vfs)
2620 dev_err(adap->pdev_dev,
2621 "Min tx rate (%d) (> 0) for VF %d is Invalid.\n",
2625 /* Retrieve link details for VF port */
2626 memset(&port_cmd, 0, sizeof(port_cmd));
2627 port_cmd.op_to_portid = cpu_to_be32(FW_CMD_OP_V(FW_PORT_CMD) |
2630 FW_PORT_CMD_PORTID_V(pi->port_id));
2631 port_cmd.action_to_len16 =
2632 cpu_to_be32(FW_PORT_CMD_ACTION_V(FW_PORT_ACTION_GET_PORT_INFO) |
2633 FW_LEN16(port_cmd));
2634 ret = t4_wr_mbox(adap, adap->mbox, &port_cmd, sizeof(port_cmd),
2636 if (ret != FW_SUCCESS) {
2637 dev_err(adap->pdev_dev,
2638 "Failed to get link status for VF %d\n", vf);
2641 link_status = be32_to_cpu(port_rpl.u.info.lstatus_to_modtype);
2642 link_ok = (link_status & FW_PORT_CMD_LSTATUS_F) != 0;
2644 dev_err(adap->pdev_dev, "Link down for VF %d\n", vf);
2647 /* Determine link speed */
2648 if (link_status & FW_PORT_CMD_LSPEED_V(FW_PORT_CAP_SPEED_100M))
2650 else if (link_status & FW_PORT_CMD_LSPEED_V(FW_PORT_CAP_SPEED_1G))
2652 else if (link_status & FW_PORT_CMD_LSPEED_V(FW_PORT_CAP_SPEED_10G))
2654 else if (link_status & FW_PORT_CMD_LSPEED_V(FW_PORT_CAP_SPEED_25G))
2656 else if (link_status & FW_PORT_CMD_LSPEED_V(FW_PORT_CAP_SPEED_40G))
2658 else if (link_status & FW_PORT_CMD_LSPEED_V(FW_PORT_CAP_SPEED_100G))
2661 if (max_tx_rate > speed) {
2662 dev_err(adap->pdev_dev,
2663 "Max tx rate %d for VF %d can't be > link-speed %u",
2664 max_tx_rate, vf, speed);
2667 pktsize = be16_to_cpu(port_rpl.u.info.mtu);
2668 /* subtract ethhdr size and 4 bytes crc since, f/w appends it */
2669 pktsize = pktsize - sizeof(struct ethhdr) - 4;
2670 /* subtract ipv4 hdr size, tcp hdr size to get typical IPv4 MSS size */
2671 pktsize = pktsize - sizeof(struct iphdr) - sizeof(struct tcphdr);
2672 /* configure Traffic Class for rate-limiting */
2673 ret = t4_sched_params(adap, SCHED_CLASS_TYPE_PACKET,
2674 SCHED_CLASS_LEVEL_CL_RL,
2675 SCHED_CLASS_MODE_CLASS,
2676 SCHED_CLASS_RATEUNIT_BITS,
2677 SCHED_CLASS_RATEMODE_ABS,
2678 pi->port_id, class_id, 0,
2679 max_tx_rate * 1000, 0, pktsize);
2681 dev_err(adap->pdev_dev, "Err %d for Traffic Class config\n",
2685 dev_info(adap->pdev_dev,
2686 "Class %d with MSS %u configured with rate %u\n",
2687 class_id, pktsize, max_tx_rate);
2689 /* bind VF to configured Traffic Class */
2690 fw_pfvf = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) |
2691 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_SCHEDCLASS_ETH));
2692 fw_class = class_id;
2693 ret = t4_set_params(adap, adap->mbox, adap->pf, vf + 1, 1, &fw_pfvf,
2696 dev_err(adap->pdev_dev,
2697 "Err %d in binding VF %d to Traffic Class %d\n",
2701 dev_info(adap->pdev_dev, "PF %d VF %d is bound to Class %d\n",
2702 adap->pf, vf, class_id);
2703 adap->vfinfo[vf].tx_rate = max_tx_rate;
2709 static int cxgb_set_mac_addr(struct net_device *dev, void *p)
2712 struct sockaddr *addr = p;
2713 struct port_info *pi = netdev_priv(dev);
2715 if (!is_valid_ether_addr(addr->sa_data))
2716 return -EADDRNOTAVAIL;
2718 ret = t4_change_mac(pi->adapter, pi->adapter->pf, pi->viid,
2719 pi->xact_addr_filt, addr->sa_data, true, true);
2723 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
2724 pi->xact_addr_filt = ret;
2728 #ifdef CONFIG_NET_POLL_CONTROLLER
2729 static void cxgb_netpoll(struct net_device *dev)
2731 struct port_info *pi = netdev_priv(dev);
2732 struct adapter *adap = pi->adapter;
2734 if (adap->flags & USING_MSIX) {
2736 struct sge_eth_rxq *rx = &adap->sge.ethrxq[pi->first_qset];
2738 for (i = pi->nqsets; i; i--, rx++)
2739 t4_sge_intr_msix(0, &rx->rspq);
2741 t4_intr_handler(adap)(0, adap);
2745 static int cxgb_set_tx_maxrate(struct net_device *dev, int index, u32 rate)
2747 struct port_info *pi = netdev_priv(dev);
2748 struct adapter *adap = pi->adapter;
2749 struct sched_class *e;
2750 struct ch_sched_params p;
2751 struct ch_sched_queue qe;
2755 if (!can_sched(dev))
2758 if (index < 0 || index > pi->nqsets - 1)
2761 if (!(adap->flags & FULL_INIT_DONE)) {
2762 dev_err(adap->pdev_dev,
2763 "Failed to rate limit on queue %d. Link Down?\n",
2768 /* Convert from Mbps to Kbps */
2769 req_rate = rate << 10;
2771 /* Max rate is 10 Gbps */
2772 if (req_rate >= SCHED_MAX_RATE_KBPS) {
2773 dev_err(adap->pdev_dev,
2774 "Invalid rate %u Mbps, Max rate is %u Gbps\n",
2775 rate, SCHED_MAX_RATE_KBPS);
2779 /* First unbind the queue from any existing class */
2780 memset(&qe, 0, sizeof(qe));
2782 qe.class = SCHED_CLS_NONE;
2784 err = cxgb4_sched_class_unbind(dev, (void *)(&qe), SCHED_QUEUE);
2786 dev_err(adap->pdev_dev,
2787 "Unbinding Queue %d on port %d fail. Err: %d\n",
2788 index, pi->port_id, err);
2792 /* Queue already unbound */
2796 /* Fetch any available unused or matching scheduling class */
2797 memset(&p, 0, sizeof(p));
2798 p.type = SCHED_CLASS_TYPE_PACKET;
2799 p.u.params.level = SCHED_CLASS_LEVEL_CL_RL;
2800 p.u.params.mode = SCHED_CLASS_MODE_CLASS;
2801 p.u.params.rateunit = SCHED_CLASS_RATEUNIT_BITS;
2802 p.u.params.ratemode = SCHED_CLASS_RATEMODE_ABS;
2803 p.u.params.channel = pi->tx_chan;
2804 p.u.params.class = SCHED_CLS_NONE;
2805 p.u.params.minrate = 0;
2806 p.u.params.maxrate = req_rate;
2807 p.u.params.weight = 0;
2808 p.u.params.pktsize = dev->mtu;
2810 e = cxgb4_sched_class_alloc(dev, &p);
2814 /* Bind the queue to a scheduling class */
2815 memset(&qe, 0, sizeof(qe));
2819 err = cxgb4_sched_class_bind(dev, (void *)(&qe), SCHED_QUEUE);
2821 dev_err(adap->pdev_dev,
2822 "Queue rate limiting failed. Err: %d\n", err);
2826 static int cxgb_setup_tc(struct net_device *dev, u32 handle, u32 chain_index,
2827 __be16 proto, struct tc_to_netdev *tc)
2829 struct port_info *pi = netdev2pinfo(dev);
2830 struct adapter *adap = netdev2adap(dev);
2835 if (!(adap->flags & FULL_INIT_DONE)) {
2836 dev_err(adap->pdev_dev,
2837 "Failed to setup tc on port %d. Link Down?\n",
2842 if (TC_H_MAJ(handle) == TC_H_MAJ(TC_H_INGRESS) &&
2843 tc->type == TC_SETUP_CLSU32) {
2844 switch (tc->cls_u32->command) {
2845 case TC_CLSU32_NEW_KNODE:
2846 case TC_CLSU32_REPLACE_KNODE:
2847 return cxgb4_config_knode(dev, proto, tc->cls_u32);
2848 case TC_CLSU32_DELETE_KNODE:
2849 return cxgb4_delete_knode(dev, proto, tc->cls_u32);
2858 static netdev_features_t cxgb_fix_features(struct net_device *dev,
2859 netdev_features_t features)
2861 /* Disable GRO, if RX_CSUM is disabled */
2862 if (!(features & NETIF_F_RXCSUM))
2863 features &= ~NETIF_F_GRO;
2868 static const struct net_device_ops cxgb4_netdev_ops = {
2869 .ndo_open = cxgb_open,
2870 .ndo_stop = cxgb_close,
2871 .ndo_start_xmit = t4_eth_xmit,
2872 .ndo_select_queue = cxgb_select_queue,
2873 .ndo_get_stats64 = cxgb_get_stats,
2874 .ndo_set_rx_mode = cxgb_set_rxmode,
2875 .ndo_set_mac_address = cxgb_set_mac_addr,
2876 .ndo_set_features = cxgb_set_features,
2877 .ndo_validate_addr = eth_validate_addr,
2878 .ndo_do_ioctl = cxgb_ioctl,
2879 .ndo_change_mtu = cxgb_change_mtu,
2880 #ifdef CONFIG_NET_POLL_CONTROLLER
2881 .ndo_poll_controller = cxgb_netpoll,
2883 #ifdef CONFIG_CHELSIO_T4_FCOE
2884 .ndo_fcoe_enable = cxgb_fcoe_enable,
2885 .ndo_fcoe_disable = cxgb_fcoe_disable,
2886 #endif /* CONFIG_CHELSIO_T4_FCOE */
2887 .ndo_set_tx_maxrate = cxgb_set_tx_maxrate,
2888 .ndo_setup_tc = cxgb_setup_tc,
2889 .ndo_fix_features = cxgb_fix_features,
2892 #ifdef CONFIG_PCI_IOV
2893 static const struct net_device_ops cxgb4_mgmt_netdev_ops = {
2894 .ndo_open = dummy_open,
2895 .ndo_set_vf_mac = cxgb_set_vf_mac,
2896 .ndo_get_vf_config = cxgb_get_vf_config,
2897 .ndo_set_vf_rate = cxgb_set_vf_rate,
2898 .ndo_get_phys_port_id = cxgb_get_phys_port_id,
2902 static void get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2904 struct adapter *adapter = netdev2adap(dev);
2906 strlcpy(info->driver, cxgb4_driver_name, sizeof(info->driver));
2907 strlcpy(info->version, cxgb4_driver_version,
2908 sizeof(info->version));
2909 strlcpy(info->bus_info, pci_name(adapter->pdev),
2910 sizeof(info->bus_info));
2913 static const struct ethtool_ops cxgb4_mgmt_ethtool_ops = {
2914 .get_drvinfo = get_drvinfo,
2917 void t4_fatal_err(struct adapter *adap)
2921 if (pci_channel_offline(adap->pdev))
2924 /* Disable the SGE since ULDs are going to free resources that
2925 * could be exposed to the adapter. RDMA MWs for example...
2927 t4_shutdown_adapter(adap);
2928 for_each_port(adap, port) {
2929 struct net_device *dev = adap->port[port];
2931 /* If we get here in very early initialization the network
2932 * devices may not have been set up yet.
2937 netif_tx_stop_all_queues(dev);
2938 netif_carrier_off(dev);
2940 dev_alert(adap->pdev_dev, "encountered fatal error, adapter stopped\n");
2943 static void setup_memwin(struct adapter *adap)
2945 u32 nic_win_base = t4_get_util_window(adap);
2947 t4_setup_memwin(adap, nic_win_base, MEMWIN_NIC);
2950 static void setup_memwin_rdma(struct adapter *adap)
2952 if (adap->vres.ocq.size) {
2956 start = t4_read_pcie_cfg4(adap, PCI_BASE_ADDRESS_2);
2957 start &= PCI_BASE_ADDRESS_MEM_MASK;
2958 start += OCQ_WIN_OFFSET(adap->pdev, &adap->vres);
2959 sz_kb = roundup_pow_of_two(adap->vres.ocq.size) >> 10;
2961 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN_A, 3),
2962 start | BIR_V(1) | WINDOW_V(ilog2(sz_kb)));
2964 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET_A, 3),
2965 adap->vres.ocq.start);
2967 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET_A, 3));
2971 static int adap_init1(struct adapter *adap, struct fw_caps_config_cmd *c)
2976 /* get device capabilities */
2977 memset(c, 0, sizeof(*c));
2978 c->op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
2979 FW_CMD_REQUEST_F | FW_CMD_READ_F);
2980 c->cfvalid_to_len16 = htonl(FW_LEN16(*c));
2981 ret = t4_wr_mbox(adap, adap->mbox, c, sizeof(*c), c);
2985 c->op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
2986 FW_CMD_REQUEST_F | FW_CMD_WRITE_F);
2987 ret = t4_wr_mbox(adap, adap->mbox, c, sizeof(*c), NULL);
2991 ret = t4_config_glbl_rss(adap, adap->pf,
2992 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL,
2993 FW_RSS_GLB_CONFIG_CMD_TNLMAPEN_F |
2994 FW_RSS_GLB_CONFIG_CMD_TNLALLLKP_F);
2998 ret = t4_cfg_pfvf(adap, adap->mbox, adap->pf, 0, adap->sge.egr_sz, 64,
2999 MAX_INGQ, 0, 0, 4, 0xf, 0xf, 16, FW_CMD_CAP_PF,
3006 /* tweak some settings */
3007 t4_write_reg(adap, TP_SHIFT_CNT_A, 0x64f8849);
3008 t4_write_reg(adap, ULP_RX_TDDP_PSZ_A, HPZ0_V(PAGE_SHIFT - 12));
3009 t4_write_reg(adap, TP_PIO_ADDR_A, TP_INGRESS_CONFIG_A);
3010 v = t4_read_reg(adap, TP_PIO_DATA_A);
3011 t4_write_reg(adap, TP_PIO_DATA_A, v & ~CSUM_HAS_PSEUDO_HDR_F);
3013 /* first 4 Tx modulation queues point to consecutive Tx channels */
3014 adap->params.tp.tx_modq_map = 0xE4;
3015 t4_write_reg(adap, TP_TX_MOD_QUEUE_REQ_MAP_A,
3016 TX_MOD_QUEUE_REQ_MAP_V(adap->params.tp.tx_modq_map));
3018 /* associate each Tx modulation queue with consecutive Tx channels */
3020 t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
3021 &v, 1, TP_TX_SCHED_HDR_A);
3022 t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
3023 &v, 1, TP_TX_SCHED_FIFO_A);
3024 t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
3025 &v, 1, TP_TX_SCHED_PCMD_A);
3027 #define T4_TX_MODQ_10G_WEIGHT_DEFAULT 16 /* in KB units */
3028 if (is_offload(adap)) {
3029 t4_write_reg(adap, TP_TX_MOD_QUEUE_WEIGHT0_A,
3030 TX_MODQ_WEIGHT0_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3031 TX_MODQ_WEIGHT1_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3032 TX_MODQ_WEIGHT2_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3033 TX_MODQ_WEIGHT3_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
3034 t4_write_reg(adap, TP_TX_MOD_CHANNEL_WEIGHT_A,
3035 TX_MODQ_WEIGHT0_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3036 TX_MODQ_WEIGHT1_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3037 TX_MODQ_WEIGHT2_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3038 TX_MODQ_WEIGHT3_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
3041 /* get basic stuff going */
3042 return t4_early_init(adap, adap->pf);
3046 * Max # of ATIDs. The absolute HW max is 16K but we keep it lower.
3048 #define MAX_ATIDS 8192U
3051 * Phase 0 of initialization: contact FW, obtain config, perform basic init.
3053 * If the firmware we're dealing with has Configuration File support, then
3054 * we use that to perform all configuration
3058 * Tweak configuration based on module parameters, etc. Most of these have
3059 * defaults assigned to them by Firmware Configuration Files (if we're using
3060 * them) but need to be explicitly set if we're using hard-coded
3061 * initialization. But even in the case of using Firmware Configuration
3062 * Files, we'd like to expose the ability to change these via module
3063 * parameters so these are essentially common tweaks/settings for
3064 * Configuration Files and hard-coded initialization ...
3066 static int adap_init0_tweaks(struct adapter *adapter)
3069 * Fix up various Host-Dependent Parameters like Page Size, Cache
3070 * Line Size, etc. The firmware default is for a 4KB Page Size and
3071 * 64B Cache Line Size ...
3073 t4_fixup_host_params(adapter, PAGE_SIZE, L1_CACHE_BYTES);
3076 * Process module parameters which affect early initialization.
3078 if (rx_dma_offset != 2 && rx_dma_offset != 0) {
3079 dev_err(&adapter->pdev->dev,
3080 "Ignoring illegal rx_dma_offset=%d, using 2\n",
3084 t4_set_reg_field(adapter, SGE_CONTROL_A,
3085 PKTSHIFT_V(PKTSHIFT_M),
3086 PKTSHIFT_V(rx_dma_offset));
3089 * Don't include the "IP Pseudo Header" in CPL_RX_PKT checksums: Linux
3090 * adds the pseudo header itself.
3092 t4_tp_wr_bits_indirect(adapter, TP_INGRESS_CONFIG_A,
3093 CSUM_HAS_PSEUDO_HDR_F, 0);
3098 /* 10Gb/s-BT PHY Support. chip-external 10Gb/s-BT PHYs are complex chips
3099 * unto themselves and they contain their own firmware to perform their
3102 static int phy_aq1202_version(const u8 *phy_fw_data,
3107 /* At offset 0x8 you're looking for the primary image's
3108 * starting offset which is 3 Bytes wide
3110 * At offset 0xa of the primary image, you look for the offset
3111 * of the DRAM segment which is 3 Bytes wide.
3113 * The FW version is at offset 0x27e of the DRAM and is 2 Bytes
3116 #define be16(__p) (((__p)[0] << 8) | (__p)[1])
3117 #define le16(__p) ((__p)[0] | ((__p)[1] << 8))
3118 #define le24(__p) (le16(__p) | ((__p)[2] << 16))
3120 offset = le24(phy_fw_data + 0x8) << 12;
3121 offset = le24(phy_fw_data + offset + 0xa);
3122 return be16(phy_fw_data + offset + 0x27e);
3129 static struct info_10gbt_phy_fw {
3130 unsigned int phy_fw_id; /* PCI Device ID */
3131 char *phy_fw_file; /* /lib/firmware/ PHY Firmware file */
3132 int (*phy_fw_version)(const u8 *phy_fw_data, size_t phy_fw_size);
3133 int phy_flash; /* Has FLASH for PHY Firmware */
3134 } phy_info_array[] = {
3136 PHY_AQ1202_DEVICEID,
3137 PHY_AQ1202_FIRMWARE,
3142 PHY_BCM84834_DEVICEID,
3143 PHY_BCM84834_FIRMWARE,
3150 static struct info_10gbt_phy_fw *find_phy_info(int devid)
3154 for (i = 0; i < ARRAY_SIZE(phy_info_array); i++) {
3155 if (phy_info_array[i].phy_fw_id == devid)
3156 return &phy_info_array[i];
3161 /* Handle updating of chip-external 10Gb/s-BT PHY firmware. This needs to
3162 * happen after the FW_RESET_CMD but before the FW_INITIALIZE_CMD. On error
3163 * we return a negative error number. If we transfer new firmware we return 1
3164 * (from t4_load_phy_fw()). If we don't do anything we return 0.
3166 static int adap_init0_phy(struct adapter *adap)
3168 const struct firmware *phyf;
3170 struct info_10gbt_phy_fw *phy_info;
3172 /* Use the device ID to determine which PHY file to flash.
3174 phy_info = find_phy_info(adap->pdev->device);
3176 dev_warn(adap->pdev_dev,
3177 "No PHY Firmware file found for this PHY\n");
3181 /* If we have a T4 PHY firmware file under /lib/firmware/cxgb4/, then
3182 * use that. The adapter firmware provides us with a memory buffer
3183 * where we can load a PHY firmware file from the host if we want to
3184 * override the PHY firmware File in flash.
3186 ret = request_firmware_direct(&phyf, phy_info->phy_fw_file,
3189 /* For adapters without FLASH attached to PHY for their
3190 * firmware, it's obviously a fatal error if we can't get the
3191 * firmware to the adapter. For adapters with PHY firmware
3192 * FLASH storage, it's worth a warning if we can't find the
3193 * PHY Firmware but we'll neuter the error ...
3195 dev_err(adap->pdev_dev, "unable to find PHY Firmware image "
3196 "/lib/firmware/%s, error %d\n",
3197 phy_info->phy_fw_file, -ret);
3198 if (phy_info->phy_flash) {
3199 int cur_phy_fw_ver = 0;
3201 t4_phy_fw_ver(adap, &cur_phy_fw_ver);
3202 dev_warn(adap->pdev_dev, "continuing with, on-adapter "
3203 "FLASH copy, version %#x\n", cur_phy_fw_ver);
3210 /* Load PHY Firmware onto adapter.
3212 ret = t4_load_phy_fw(adap, MEMWIN_NIC, &adap->win0_lock,
3213 phy_info->phy_fw_version,
3214 (u8 *)phyf->data, phyf->size);
3216 dev_err(adap->pdev_dev, "PHY Firmware transfer error %d\n",
3219 int new_phy_fw_ver = 0;
3221 if (phy_info->phy_fw_version)
3222 new_phy_fw_ver = phy_info->phy_fw_version(phyf->data,
3224 dev_info(adap->pdev_dev, "Successfully transferred PHY "
3225 "Firmware /lib/firmware/%s, version %#x\n",
3226 phy_info->phy_fw_file, new_phy_fw_ver);
3229 release_firmware(phyf);
3235 * Attempt to initialize the adapter via a Firmware Configuration File.
3237 static int adap_init0_config(struct adapter *adapter, int reset)
3239 struct fw_caps_config_cmd caps_cmd;
3240 const struct firmware *cf;
3241 unsigned long mtype = 0, maddr = 0;
3242 u32 finiver, finicsum, cfcsum;
3244 int config_issued = 0;
3245 char *fw_config_file, fw_config_file_path[256];
3246 char *config_name = NULL;
3249 * Reset device if necessary.
3252 ret = t4_fw_reset(adapter, adapter->mbox,
3253 PIORSTMODE_F | PIORST_F);
3258 /* If this is a 10Gb/s-BT adapter make sure the chip-external
3259 * 10Gb/s-BT PHYs have up-to-date firmware. Note that this step needs
3260 * to be performed after any global adapter RESET above since some
3261 * PHYs only have local RAM copies of the PHY firmware.
3263 if (is_10gbt_device(adapter->pdev->device)) {
3264 ret = adap_init0_phy(adapter);
3269 * If we have a T4 configuration file under /lib/firmware/cxgb4/,
3270 * then use that. Otherwise, use the configuration file stored
3271 * in the adapter flash ...
3273 switch (CHELSIO_CHIP_VERSION(adapter->params.chip)) {
3275 fw_config_file = FW4_CFNAME;
3278 fw_config_file = FW5_CFNAME;
3281 fw_config_file = FW6_CFNAME;
3284 dev_err(adapter->pdev_dev, "Device %d is not supported\n",
3285 adapter->pdev->device);
3290 ret = request_firmware(&cf, fw_config_file, adapter->pdev_dev);
3292 config_name = "On FLASH";
3293 mtype = FW_MEMTYPE_CF_FLASH;
3294 maddr = t4_flash_cfg_addr(adapter);
3296 u32 params[7], val[7];
3298 sprintf(fw_config_file_path,
3299 "/lib/firmware/%s", fw_config_file);
3300 config_name = fw_config_file_path;
3302 if (cf->size >= FLASH_CFG_MAX_SIZE)
3305 params[0] = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
3306 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_CF));
3307 ret = t4_query_params(adapter, adapter->mbox,
3308 adapter->pf, 0, 1, params, val);
3311 * For t4_memory_rw() below addresses and
3312 * sizes have to be in terms of multiples of 4
3313 * bytes. So, if the Configuration File isn't
3314 * a multiple of 4 bytes in length we'll have
3315 * to write that out separately since we can't
3316 * guarantee that the bytes following the
3317 * residual byte in the buffer returned by
3318 * request_firmware() are zeroed out ...
3320 size_t resid = cf->size & 0x3;
3321 size_t size = cf->size & ~0x3;
3322 __be32 *data = (__be32 *)cf->data;
3324 mtype = FW_PARAMS_PARAM_Y_G(val[0]);
3325 maddr = FW_PARAMS_PARAM_Z_G(val[0]) << 16;
3327 spin_lock(&adapter->win0_lock);
3328 ret = t4_memory_rw(adapter, 0, mtype, maddr,
3329 size, data, T4_MEMORY_WRITE);
3330 if (ret == 0 && resid != 0) {
3337 last.word = data[size >> 2];
3338 for (i = resid; i < 4; i++)
3340 ret = t4_memory_rw(adapter, 0, mtype,
3345 spin_unlock(&adapter->win0_lock);
3349 release_firmware(cf);
3355 * Issue a Capability Configuration command to the firmware to get it
3356 * to parse the Configuration File. We don't use t4_fw_config_file()
3357 * because we want the ability to modify various features after we've
3358 * processed the configuration file ...
3360 memset(&caps_cmd, 0, sizeof(caps_cmd));
3361 caps_cmd.op_to_write =
3362 htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3365 caps_cmd.cfvalid_to_len16 =
3366 htonl(FW_CAPS_CONFIG_CMD_CFVALID_F |
3367 FW_CAPS_CONFIG_CMD_MEMTYPE_CF_V(mtype) |
3368 FW_CAPS_CONFIG_CMD_MEMADDR64K_CF_V(maddr >> 16) |
3369 FW_LEN16(caps_cmd));
3370 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
3373 /* If the CAPS_CONFIG failed with an ENOENT (for a Firmware
3374 * Configuration File in FLASH), our last gasp effort is to use the
3375 * Firmware Configuration File which is embedded in the firmware. A
3376 * very few early versions of the firmware didn't have one embedded
3377 * but we can ignore those.
3379 if (ret == -ENOENT) {
3380 memset(&caps_cmd, 0, sizeof(caps_cmd));
3381 caps_cmd.op_to_write =
3382 htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3385 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
3386 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd,
3387 sizeof(caps_cmd), &caps_cmd);
3388 config_name = "Firmware Default";
3395 finiver = ntohl(caps_cmd.finiver);
3396 finicsum = ntohl(caps_cmd.finicsum);
3397 cfcsum = ntohl(caps_cmd.cfcsum);
3398 if (finicsum != cfcsum)
3399 dev_warn(adapter->pdev_dev, "Configuration File checksum "\
3400 "mismatch: [fini] csum=%#x, computed csum=%#x\n",
3404 * And now tell the firmware to use the configuration we just loaded.
3406 caps_cmd.op_to_write =
3407 htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3410 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
3411 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
3417 * Tweak configuration based on system architecture, module
3420 ret = adap_init0_tweaks(adapter);
3425 * And finally tell the firmware to initialize itself using the
3426 * parameters from the Configuration File.
3428 ret = t4_fw_initialize(adapter, adapter->mbox);
3432 /* Emit Firmware Configuration File information and return
3435 dev_info(adapter->pdev_dev, "Successfully configured using Firmware "\
3436 "Configuration File \"%s\", version %#x, computed checksum %#x\n",
3437 config_name, finiver, cfcsum);
3441 * Something bad happened. Return the error ... (If the "error"
3442 * is that there's no Configuration File on the adapter we don't
3443 * want to issue a warning since this is fairly common.)
3446 if (config_issued && ret != -ENOENT)
3447 dev_warn(adapter->pdev_dev, "\"%s\" configuration file error %d\n",
3452 static struct fw_info fw_info_array[] = {
3455 .fs_name = FW4_CFNAME,
3456 .fw_mod_name = FW4_FNAME,
3458 .chip = FW_HDR_CHIP_T4,
3459 .fw_ver = __cpu_to_be32(FW_VERSION(T4)),
3460 .intfver_nic = FW_INTFVER(T4, NIC),
3461 .intfver_vnic = FW_INTFVER(T4, VNIC),
3462 .intfver_ri = FW_INTFVER(T4, RI),
3463 .intfver_iscsi = FW_INTFVER(T4, ISCSI),
3464 .intfver_fcoe = FW_INTFVER(T4, FCOE),
3468 .fs_name = FW5_CFNAME,
3469 .fw_mod_name = FW5_FNAME,
3471 .chip = FW_HDR_CHIP_T5,
3472 .fw_ver = __cpu_to_be32(FW_VERSION(T5)),
3473 .intfver_nic = FW_INTFVER(T5, NIC),
3474 .intfver_vnic = FW_INTFVER(T5, VNIC),
3475 .intfver_ri = FW_INTFVER(T5, RI),
3476 .intfver_iscsi = FW_INTFVER(T5, ISCSI),
3477 .intfver_fcoe = FW_INTFVER(T5, FCOE),
3481 .fs_name = FW6_CFNAME,
3482 .fw_mod_name = FW6_FNAME,
3484 .chip = FW_HDR_CHIP_T6,
3485 .fw_ver = __cpu_to_be32(FW_VERSION(T6)),
3486 .intfver_nic = FW_INTFVER(T6, NIC),
3487 .intfver_vnic = FW_INTFVER(T6, VNIC),
3488 .intfver_ofld = FW_INTFVER(T6, OFLD),
3489 .intfver_ri = FW_INTFVER(T6, RI),
3490 .intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
3491 .intfver_iscsi = FW_INTFVER(T6, ISCSI),
3492 .intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
3493 .intfver_fcoe = FW_INTFVER(T6, FCOE),
3499 static struct fw_info *find_fw_info(int chip)
3503 for (i = 0; i < ARRAY_SIZE(fw_info_array); i++) {
3504 if (fw_info_array[i].chip == chip)
3505 return &fw_info_array[i];
3511 * Phase 0 of initialization: contact FW, obtain config, perform basic init.
3513 static int adap_init0(struct adapter *adap)
3517 enum dev_state state;
3518 u32 params[7], val[7];
3519 struct fw_caps_config_cmd caps_cmd;
3522 /* Grab Firmware Device Log parameters as early as possible so we have
3523 * access to it for debugging, etc.
3525 ret = t4_init_devlog_params(adap);
3529 /* Contact FW, advertising Master capability */
3530 ret = t4_fw_hello(adap, adap->mbox, adap->mbox,
3531 is_kdump_kernel() ? MASTER_MUST : MASTER_MAY, &state);
3533 dev_err(adap->pdev_dev, "could not connect to FW, error %d\n",
3537 if (ret == adap->mbox)
3538 adap->flags |= MASTER_PF;
3541 * If we're the Master PF Driver and the device is uninitialized,
3542 * then let's consider upgrading the firmware ... (We always want
3543 * to check the firmware version number in order to A. get it for
3544 * later reporting and B. to warn if the currently loaded firmware
3545 * is excessively mismatched relative to the driver.)
3547 t4_get_fw_version(adap, &adap->params.fw_vers);
3548 t4_get_bs_version(adap, &adap->params.bs_vers);
3549 t4_get_tp_version(adap, &adap->params.tp_vers);
3550 t4_get_exprom_version(adap, &adap->params.er_vers);
3552 ret = t4_check_fw_version(adap);
3553 /* If firmware is too old (not supported by driver) force an update. */
3555 state = DEV_STATE_UNINIT;
3556 if ((adap->flags & MASTER_PF) && state != DEV_STATE_INIT) {
3557 struct fw_info *fw_info;
3558 struct fw_hdr *card_fw;
3559 const struct firmware *fw;
3560 const u8 *fw_data = NULL;
3561 unsigned int fw_size = 0;
3563 /* This is the firmware whose headers the driver was compiled
3566 fw_info = find_fw_info(CHELSIO_CHIP_VERSION(adap->params.chip));
3567 if (fw_info == NULL) {
3568 dev_err(adap->pdev_dev,
3569 "unable to get firmware info for chip %d.\n",
3570 CHELSIO_CHIP_VERSION(adap->params.chip));
3574 /* allocate memory to read the header of the firmware on the
3577 card_fw = kvzalloc(sizeof(*card_fw), GFP_KERNEL);
3579 /* Get FW from from /lib/firmware/ */
3580 ret = request_firmware(&fw, fw_info->fw_mod_name,
3583 dev_err(adap->pdev_dev,
3584 "unable to load firmware image %s, error %d\n",
3585 fw_info->fw_mod_name, ret);
3591 /* upgrade FW logic */
3592 ret = t4_prep_fw(adap, fw_info, fw_data, fw_size, card_fw,
3596 release_firmware(fw);
3604 * Grab VPD parameters. This should be done after we establish a
3605 * connection to the firmware since some of the VPD parameters
3606 * (notably the Core Clock frequency) are retrieved via requests to
3607 * the firmware. On the other hand, we need these fairly early on
3608 * so we do this right after getting ahold of the firmware.
3610 ret = t4_get_vpd_params(adap, &adap->params.vpd);
3615 * Find out what ports are available to us. Note that we need to do
3616 * this before calling adap_init0_no_config() since it needs nports
3620 FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
3621 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_PORTVEC);
3622 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, &v, &port_vec);
3626 adap->params.nports = hweight32(port_vec);
3627 adap->params.portvec = port_vec;
3629 /* If the firmware is initialized already, emit a simply note to that
3630 * effect. Otherwise, it's time to try initializing the adapter.
3632 if (state == DEV_STATE_INIT) {
3633 dev_info(adap->pdev_dev, "Coming up as %s: "\
3634 "Adapter already initialized\n",
3635 adap->flags & MASTER_PF ? "MASTER" : "SLAVE");
3637 dev_info(adap->pdev_dev, "Coming up as MASTER: "\
3638 "Initializing adapter\n");
3640 /* Find out whether we're dealing with a version of the
3641 * firmware which has configuration file support.
3643 params[0] = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
3644 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_CF));
3645 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1,
3648 /* If the firmware doesn't support Configuration Files,
3652 dev_err(adap->pdev_dev, "firmware doesn't support "
3653 "Firmware Configuration Files\n");
3657 /* The firmware provides us with a memory buffer where we can
3658 * load a Configuration File from the host if we want to
3659 * override the Configuration File in flash.
3661 ret = adap_init0_config(adap, reset);
3662 if (ret == -ENOENT) {
3663 dev_err(adap->pdev_dev, "no Configuration File "
3664 "present on adapter.\n");
3668 dev_err(adap->pdev_dev, "could not initialize "
3669 "adapter, error %d\n", -ret);
3674 /* Give the SGE code a chance to pull in anything that it needs ...
3675 * Note that this must be called after we retrieve our VPD parameters
3676 * in order to know how to convert core ticks to seconds, etc.
3678 ret = t4_sge_init(adap);
3682 if (is_bypass_device(adap->pdev->device))
3683 adap->params.bypass = 1;
3686 * Grab some of our basic fundamental operating parameters.
3688 #define FW_PARAM_DEV(param) \
3689 (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) | \
3690 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_##param))
3692 #define FW_PARAM_PFVF(param) \
3693 FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) | \
3694 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_##param)| \
3695 FW_PARAMS_PARAM_Y_V(0) | \
3696 FW_PARAMS_PARAM_Z_V(0)
3698 params[0] = FW_PARAM_PFVF(EQ_START);
3699 params[1] = FW_PARAM_PFVF(L2T_START);
3700 params[2] = FW_PARAM_PFVF(L2T_END);
3701 params[3] = FW_PARAM_PFVF(FILTER_START);
3702 params[4] = FW_PARAM_PFVF(FILTER_END);
3703 params[5] = FW_PARAM_PFVF(IQFLINT_START);
3704 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6, params, val);
3707 adap->sge.egr_start = val[0];
3708 adap->l2t_start = val[1];
3709 adap->l2t_end = val[2];
3710 adap->tids.ftid_base = val[3];
3711 adap->tids.nftids = val[4] - val[3] + 1;
3712 adap->sge.ingr_start = val[5];
3714 /* qids (ingress/egress) returned from firmware can be anywhere
3715 * in the range from EQ(IQFLINT)_START to EQ(IQFLINT)_END.
3716 * Hence driver needs to allocate memory for this range to
3717 * store the queue info. Get the highest IQFLINT/EQ index returned
3718 * in FW_EQ_*_CMD.alloc command.
3720 params[0] = FW_PARAM_PFVF(EQ_END);
3721 params[1] = FW_PARAM_PFVF(IQFLINT_END);
3722 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
3725 adap->sge.egr_sz = val[0] - adap->sge.egr_start + 1;
3726 adap->sge.ingr_sz = val[1] - adap->sge.ingr_start + 1;
3728 adap->sge.egr_map = kcalloc(adap->sge.egr_sz,
3729 sizeof(*adap->sge.egr_map), GFP_KERNEL);
3730 if (!adap->sge.egr_map) {
3735 adap->sge.ingr_map = kcalloc(adap->sge.ingr_sz,
3736 sizeof(*adap->sge.ingr_map), GFP_KERNEL);
3737 if (!adap->sge.ingr_map) {
3742 /* Allocate the memory for the vaious egress queue bitmaps
3743 * ie starving_fl, txq_maperr and blocked_fl.
3745 adap->sge.starving_fl = kcalloc(BITS_TO_LONGS(adap->sge.egr_sz),
3746 sizeof(long), GFP_KERNEL);
3747 if (!adap->sge.starving_fl) {
3752 adap->sge.txq_maperr = kcalloc(BITS_TO_LONGS(adap->sge.egr_sz),
3753 sizeof(long), GFP_KERNEL);
3754 if (!adap->sge.txq_maperr) {
3759 #ifdef CONFIG_DEBUG_FS
3760 adap->sge.blocked_fl = kcalloc(BITS_TO_LONGS(adap->sge.egr_sz),
3761 sizeof(long), GFP_KERNEL);
3762 if (!adap->sge.blocked_fl) {
3768 params[0] = FW_PARAM_PFVF(CLIP_START);
3769 params[1] = FW_PARAM_PFVF(CLIP_END);
3770 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
3773 adap->clipt_start = val[0];
3774 adap->clipt_end = val[1];
3776 /* We don't yet have a PARAMs calls to retrieve the number of Traffic
3777 * Classes supported by the hardware/firmware so we hard code it here
3780 adap->params.nsched_cls = is_t4(adap->params.chip) ? 15 : 16;
3782 /* query params related to active filter region */
3783 params[0] = FW_PARAM_PFVF(ACTIVE_FILTER_START);
3784 params[1] = FW_PARAM_PFVF(ACTIVE_FILTER_END);
3785 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
3786 /* If Active filter size is set we enable establishing
3787 * offload connection through firmware work request
3789 if ((val[0] != val[1]) && (ret >= 0)) {
3790 adap->flags |= FW_OFLD_CONN;
3791 adap->tids.aftid_base = val[0];
3792 adap->tids.aftid_end = val[1];
3795 /* If we're running on newer firmware, let it know that we're
3796 * prepared to deal with encapsulated CPL messages. Older
3797 * firmware won't understand this and we'll just get
3798 * unencapsulated messages ...
3800 params[0] = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
3802 (void)t4_set_params(adap, adap->mbox, adap->pf, 0, 1, params, val);
3805 * Find out whether we're allowed to use the T5+ ULPTX MEMWRITE DSGL
3806 * capability. Earlier versions of the firmware didn't have the
3807 * ULPTX_MEMWRITE_DSGL so we'll interpret a query failure as no
3808 * permission to use ULPTX MEMWRITE DSGL.
3810 if (is_t4(adap->params.chip)) {
3811 adap->params.ulptx_memwrite_dsgl = false;
3813 params[0] = FW_PARAM_DEV(ULPTX_MEMWRITE_DSGL);
3814 ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
3816 adap->params.ulptx_memwrite_dsgl = (ret == 0 && val[0] != 0);
3819 /* See if FW supports FW_RI_FR_NSMR_TPTE_WR work request */
3820 params[0] = FW_PARAM_DEV(RI_FR_NSMR_TPTE_WR);
3821 ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
3823 adap->params.fr_nsmr_tpte_wr_support = (ret == 0 && val[0] != 0);
3826 * Get device capabilities so we can determine what resources we need
3829 memset(&caps_cmd, 0, sizeof(caps_cmd));
3830 caps_cmd.op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3831 FW_CMD_REQUEST_F | FW_CMD_READ_F);
3832 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
3833 ret = t4_wr_mbox(adap, adap->mbox, &caps_cmd, sizeof(caps_cmd),
3838 if (caps_cmd.ofldcaps) {
3839 /* query offload-related parameters */
3840 params[0] = FW_PARAM_DEV(NTID);
3841 params[1] = FW_PARAM_PFVF(SERVER_START);
3842 params[2] = FW_PARAM_PFVF(SERVER_END);
3843 params[3] = FW_PARAM_PFVF(TDDP_START);
3844 params[4] = FW_PARAM_PFVF(TDDP_END);
3845 params[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
3846 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6,
3850 adap->tids.ntids = val[0];
3851 adap->tids.natids = min(adap->tids.ntids / 2, MAX_ATIDS);
3852 adap->tids.stid_base = val[1];
3853 adap->tids.nstids = val[2] - val[1] + 1;
3855 * Setup server filter region. Divide the available filter
3856 * region into two parts. Regular filters get 1/3rd and server
3857 * filters get 2/3rd part. This is only enabled if workarond
3859 * 1. For regular filters.
3860 * 2. Server filter: This are special filters which are used
3861 * to redirect SYN packets to offload queue.
3863 if (adap->flags & FW_OFLD_CONN && !is_bypass(adap)) {
3864 adap->tids.sftid_base = adap->tids.ftid_base +
3865 DIV_ROUND_UP(adap->tids.nftids, 3);
3866 adap->tids.nsftids = adap->tids.nftids -
3867 DIV_ROUND_UP(adap->tids.nftids, 3);
3868 adap->tids.nftids = adap->tids.sftid_base -
3869 adap->tids.ftid_base;
3871 adap->vres.ddp.start = val[3];
3872 adap->vres.ddp.size = val[4] - val[3] + 1;
3873 adap->params.ofldq_wr_cred = val[5];
3875 adap->params.offload = 1;
3876 adap->num_ofld_uld += 1;
3878 if (caps_cmd.rdmacaps) {
3879 params[0] = FW_PARAM_PFVF(STAG_START);
3880 params[1] = FW_PARAM_PFVF(STAG_END);
3881 params[2] = FW_PARAM_PFVF(RQ_START);
3882 params[3] = FW_PARAM_PFVF(RQ_END);
3883 params[4] = FW_PARAM_PFVF(PBL_START);
3884 params[5] = FW_PARAM_PFVF(PBL_END);
3885 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6,
3889 adap->vres.stag.start = val[0];
3890 adap->vres.stag.size = val[1] - val[0] + 1;
3891 adap->vres.rq.start = val[2];
3892 adap->vres.rq.size = val[3] - val[2] + 1;
3893 adap->vres.pbl.start = val[4];
3894 adap->vres.pbl.size = val[5] - val[4] + 1;
3896 params[0] = FW_PARAM_PFVF(SQRQ_START);
3897 params[1] = FW_PARAM_PFVF(SQRQ_END);
3898 params[2] = FW_PARAM_PFVF(CQ_START);
3899 params[3] = FW_PARAM_PFVF(CQ_END);
3900 params[4] = FW_PARAM_PFVF(OCQ_START);
3901 params[5] = FW_PARAM_PFVF(OCQ_END);
3902 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6, params,
3906 adap->vres.qp.start = val[0];
3907 adap->vres.qp.size = val[1] - val[0] + 1;
3908 adap->vres.cq.start = val[2];
3909 adap->vres.cq.size = val[3] - val[2] + 1;
3910 adap->vres.ocq.start = val[4];
3911 adap->vres.ocq.size = val[5] - val[4] + 1;
3913 params[0] = FW_PARAM_DEV(MAXORDIRD_QP);
3914 params[1] = FW_PARAM_DEV(MAXIRD_ADAPTER);
3915 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params,
3918 adap->params.max_ordird_qp = 8;
3919 adap->params.max_ird_adapter = 32 * adap->tids.ntids;
3922 adap->params.max_ordird_qp = val[0];
3923 adap->params.max_ird_adapter = val[1];
3925 dev_info(adap->pdev_dev,
3926 "max_ordird_qp %d max_ird_adapter %d\n",
3927 adap->params.max_ordird_qp,
3928 adap->params.max_ird_adapter);
3929 adap->num_ofld_uld += 2;
3931 if (caps_cmd.iscsicaps) {
3932 params[0] = FW_PARAM_PFVF(ISCSI_START);
3933 params[1] = FW_PARAM_PFVF(ISCSI_END);
3934 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
3938 adap->vres.iscsi.start = val[0];
3939 adap->vres.iscsi.size = val[1] - val[0] + 1;
3940 /* LIO target and cxgb4i initiaitor */
3941 adap->num_ofld_uld += 2;
3943 if (caps_cmd.cryptocaps) {
3944 /* Should query params here...TODO */
3945 params[0] = FW_PARAM_PFVF(NCRYPTO_LOOKASIDE);
3946 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
3952 adap->vres.ncrypto_fc = val[0];
3954 adap->params.crypto |= ULP_CRYPTO_LOOKASIDE;
3957 #undef FW_PARAM_PFVF
3960 /* The MTU/MSS Table is initialized by now, so load their values. If
3961 * we're initializing the adapter, then we'll make any modifications
3962 * we want to the MTU/MSS Table and also initialize the congestion
3965 t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
3966 if (state != DEV_STATE_INIT) {
3969 /* The default MTU Table contains values 1492 and 1500.
3970 * However, for TCP, it's better to have two values which are
3971 * a multiple of 8 +/- 4 bytes apart near this popular MTU.
3972 * This allows us to have a TCP Data Payload which is a
3973 * multiple of 8 regardless of what combination of TCP Options
3974 * are in use (always a multiple of 4 bytes) which is
3975 * important for performance reasons. For instance, if no
3976 * options are in use, then we have a 20-byte IP header and a
3977 * 20-byte TCP header. In this case, a 1500-byte MSS would
3978 * result in a TCP Data Payload of 1500 - 40 == 1460 bytes
3979 * which is not a multiple of 8. So using an MSS of 1488 in
3980 * this case results in a TCP Data Payload of 1448 bytes which
3981 * is a multiple of 8. On the other hand, if 12-byte TCP Time
3982 * Stamps have been negotiated, then an MTU of 1500 bytes
3983 * results in a TCP Data Payload of 1448 bytes which, as
3984 * above, is a multiple of 8 bytes ...
3986 for (i = 0; i < NMTUS; i++)
3987 if (adap->params.mtus[i] == 1492) {
3988 adap->params.mtus[i] = 1488;
3992 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
3993 adap->params.b_wnd);
3995 t4_init_sge_params(adap);
3996 adap->flags |= FW_OK;
3997 t4_init_tp_params(adap);
4001 * Something bad happened. If a command timed out or failed with EIO
4002 * FW does not operate within its spec or something catastrophic
4003 * happened to HW/FW, stop issuing commands.
4006 kfree(adap->sge.egr_map);
4007 kfree(adap->sge.ingr_map);
4008 kfree(adap->sge.starving_fl);
4009 kfree(adap->sge.txq_maperr);
4010 #ifdef CONFIG_DEBUG_FS
4011 kfree(adap->sge.blocked_fl);
4013 if (ret != -ETIMEDOUT && ret != -EIO)
4014 t4_fw_bye(adap, adap->mbox);
4020 static pci_ers_result_t eeh_err_detected(struct pci_dev *pdev,
4021 pci_channel_state_t state)
4024 struct adapter *adap = pci_get_drvdata(pdev);
4030 adap->flags &= ~FW_OK;
4031 notify_ulds(adap, CXGB4_STATE_START_RECOVERY);
4032 spin_lock(&adap->stats_lock);
4033 for_each_port(adap, i) {
4034 struct net_device *dev = adap->port[i];
4036 netif_device_detach(dev);
4037 netif_carrier_off(dev);
4040 spin_unlock(&adap->stats_lock);
4041 disable_interrupts(adap);
4042 if (adap->flags & FULL_INIT_DONE)
4045 if ((adap->flags & DEV_ENABLED)) {
4046 pci_disable_device(pdev);
4047 adap->flags &= ~DEV_ENABLED;
4049 out: return state == pci_channel_io_perm_failure ?
4050 PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
4053 static pci_ers_result_t eeh_slot_reset(struct pci_dev *pdev)
4056 struct fw_caps_config_cmd c;
4057 struct adapter *adap = pci_get_drvdata(pdev);
4060 pci_restore_state(pdev);
4061 pci_save_state(pdev);
4062 return PCI_ERS_RESULT_RECOVERED;
4065 if (!(adap->flags & DEV_ENABLED)) {
4066 if (pci_enable_device(pdev)) {
4067 dev_err(&pdev->dev, "Cannot reenable PCI "
4068 "device after reset\n");
4069 return PCI_ERS_RESULT_DISCONNECT;
4071 adap->flags |= DEV_ENABLED;
4074 pci_set_master(pdev);
4075 pci_restore_state(pdev);
4076 pci_save_state(pdev);
4077 pci_cleanup_aer_uncorrect_error_status(pdev);
4079 if (t4_wait_dev_ready(adap->regs) < 0)
4080 return PCI_ERS_RESULT_DISCONNECT;
4081 if (t4_fw_hello(adap, adap->mbox, adap->pf, MASTER_MUST, NULL) < 0)
4082 return PCI_ERS_RESULT_DISCONNECT;
4083 adap->flags |= FW_OK;
4084 if (adap_init1(adap, &c))
4085 return PCI_ERS_RESULT_DISCONNECT;
4087 for_each_port(adap, i) {
4088 struct port_info *p = adap2pinfo(adap, i);
4090 ret = t4_alloc_vi(adap, adap->mbox, p->tx_chan, adap->pf, 0, 1,
4093 return PCI_ERS_RESULT_DISCONNECT;
4095 p->xact_addr_filt = -1;
4098 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
4099 adap->params.b_wnd);
4102 return PCI_ERS_RESULT_DISCONNECT;
4103 return PCI_ERS_RESULT_RECOVERED;
4106 static void eeh_resume(struct pci_dev *pdev)
4109 struct adapter *adap = pci_get_drvdata(pdev);
4115 for_each_port(adap, i) {
4116 struct net_device *dev = adap->port[i];
4118 if (netif_running(dev)) {
4120 cxgb_set_rxmode(dev);
4122 netif_device_attach(dev);
4128 static const struct pci_error_handlers cxgb4_eeh = {
4129 .error_detected = eeh_err_detected,
4130 .slot_reset = eeh_slot_reset,
4131 .resume = eeh_resume,
4134 /* Return true if the Link Configuration supports "High Speeds" (those greater
4137 static inline bool is_x_10g_port(const struct link_config *lc)
4139 unsigned int speeds, high_speeds;
4141 speeds = FW_PORT_CAP_SPEED_V(FW_PORT_CAP_SPEED_G(lc->supported));
4142 high_speeds = speeds & ~(FW_PORT_CAP_SPEED_100M | FW_PORT_CAP_SPEED_1G);
4144 return high_speeds != 0;
4148 * Perform default configuration of DMA queues depending on the number and type
4149 * of ports we found and the number of available CPUs. Most settings can be
4150 * modified by the admin prior to actual use.
4152 static void cfg_queues(struct adapter *adap)
4154 struct sge *s = &adap->sge;
4155 int i = 0, n10g = 0, qidx = 0;
4156 #ifndef CONFIG_CHELSIO_T4_DCB
4160 /* Reduce memory usage in kdump environment, disable all offload.
4162 if (is_kdump_kernel() || (is_uld(adap) && t4_uld_mem_alloc(adap))) {
4163 adap->params.offload = 0;
4164 adap->params.crypto = 0;
4167 n10g += is_x_10g_port(&adap2pinfo(adap, i)->link_cfg);
4168 #ifdef CONFIG_CHELSIO_T4_DCB
4169 /* For Data Center Bridging support we need to be able to support up
4170 * to 8 Traffic Priorities; each of which will be assigned to its
4171 * own TX Queue in order to prevent Head-Of-Line Blocking.
4173 if (adap->params.nports * 8 > MAX_ETH_QSETS) {
4174 dev_err(adap->pdev_dev, "MAX_ETH_QSETS=%d < %d!\n",
4175 MAX_ETH_QSETS, adap->params.nports * 8);
4179 for_each_port(adap, i) {
4180 struct port_info *pi = adap2pinfo(adap, i);
4182 pi->first_qset = qidx;
4183 pi->nqsets = is_kdump_kernel() ? 1 : 8;
4186 #else /* !CONFIG_CHELSIO_T4_DCB */
4188 * We default to 1 queue per non-10G port and up to # of cores queues
4192 q10g = (MAX_ETH_QSETS - (adap->params.nports - n10g)) / n10g;
4193 if (q10g > netif_get_num_default_rss_queues())
4194 q10g = netif_get_num_default_rss_queues();
4196 if (is_kdump_kernel())
4199 for_each_port(adap, i) {
4200 struct port_info *pi = adap2pinfo(adap, i);
4202 pi->first_qset = qidx;
4203 pi->nqsets = is_x_10g_port(&pi->link_cfg) ? q10g : 1;
4206 #endif /* !CONFIG_CHELSIO_T4_DCB */
4209 s->max_ethqsets = qidx; /* MSI-X may lower it later */
4213 * For offload we use 1 queue/channel if all ports are up to 1G,
4214 * otherwise we divide all available queues amongst the channels
4215 * capped by the number of available cores.
4218 i = min_t(int, MAX_OFLD_QSETS, num_online_cpus());
4219 s->ofldqsets = roundup(i, adap->params.nports);
4221 s->ofldqsets = adap->params.nports;
4225 for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
4226 struct sge_eth_rxq *r = &s->ethrxq[i];
4228 init_rspq(adap, &r->rspq, 5, 10, 1024, 64);
4232 for (i = 0; i < ARRAY_SIZE(s->ethtxq); i++)
4233 s->ethtxq[i].q.size = 1024;
4235 for (i = 0; i < ARRAY_SIZE(s->ctrlq); i++)
4236 s->ctrlq[i].q.size = 512;
4238 init_rspq(adap, &s->fw_evtq, 0, 1, 1024, 64);
4239 init_rspq(adap, &s->intrq, 0, 1, 512, 64);
4243 * Reduce the number of Ethernet queues across all ports to at most n.
4244 * n provides at least one queue per port.
4246 static void reduce_ethqs(struct adapter *adap, int n)
4249 struct port_info *pi;
4251 while (n < adap->sge.ethqsets)
4252 for_each_port(adap, i) {
4253 pi = adap2pinfo(adap, i);
4254 if (pi->nqsets > 1) {
4256 adap->sge.ethqsets--;
4257 if (adap->sge.ethqsets <= n)
4263 for_each_port(adap, i) {
4264 pi = adap2pinfo(adap, i);
4270 static int get_msix_info(struct adapter *adap)
4272 struct uld_msix_info *msix_info;
4273 unsigned int max_ingq = 0;
4275 if (is_offload(adap))
4276 max_ingq += MAX_OFLD_QSETS * adap->num_ofld_uld;
4277 if (is_pci_uld(adap))
4278 max_ingq += MAX_OFLD_QSETS * adap->num_uld;
4283 msix_info = kcalloc(max_ingq, sizeof(*msix_info), GFP_KERNEL);
4287 adap->msix_bmap_ulds.msix_bmap = kcalloc(BITS_TO_LONGS(max_ingq),
4288 sizeof(long), GFP_KERNEL);
4289 if (!adap->msix_bmap_ulds.msix_bmap) {
4293 spin_lock_init(&adap->msix_bmap_ulds.lock);
4294 adap->msix_info_ulds = msix_info;
4299 static void free_msix_info(struct adapter *adap)
4301 if (!(adap->num_uld && adap->num_ofld_uld))
4304 kfree(adap->msix_info_ulds);
4305 kfree(adap->msix_bmap_ulds.msix_bmap);
4308 /* 2 MSI-X vectors needed for the FW queue and non-data interrupts */
4309 #define EXTRA_VECS 2
4311 static int enable_msix(struct adapter *adap)
4313 int ofld_need = 0, uld_need = 0;
4314 int i, j, want, need, allocated;
4315 struct sge *s = &adap->sge;
4316 unsigned int nchan = adap->params.nports;
4317 struct msix_entry *entries;
4318 int max_ingq = MAX_INGQ;
4320 if (is_pci_uld(adap))
4321 max_ingq += (MAX_OFLD_QSETS * adap->num_uld);
4322 if (is_offload(adap))
4323 max_ingq += (MAX_OFLD_QSETS * adap->num_ofld_uld);
4324 entries = kmalloc(sizeof(*entries) * (max_ingq + 1),
4330 if (get_msix_info(adap)) {
4331 adap->params.offload = 0;
4332 adap->params.crypto = 0;
4335 for (i = 0; i < max_ingq + 1; ++i)
4336 entries[i].entry = i;
4338 want = s->max_ethqsets + EXTRA_VECS;
4339 if (is_offload(adap)) {
4340 want += adap->num_ofld_uld * s->ofldqsets;
4341 ofld_need = adap->num_ofld_uld * nchan;
4343 if (is_pci_uld(adap)) {
4344 want += adap->num_uld * s->ofldqsets;
4345 uld_need = adap->num_uld * nchan;
4347 #ifdef CONFIG_CHELSIO_T4_DCB
4348 /* For Data Center Bridging we need 8 Ethernet TX Priority Queues for
4351 need = 8 * adap->params.nports + EXTRA_VECS + ofld_need + uld_need;
4353 need = adap->params.nports + EXTRA_VECS + ofld_need + uld_need;
4355 allocated = pci_enable_msix_range(adap->pdev, entries, need, want);
4356 if (allocated < 0) {
4357 dev_info(adap->pdev_dev, "not enough MSI-X vectors left,"
4358 " not using MSI-X\n");
4363 /* Distribute available vectors to the various queue groups.
4364 * Every group gets its minimum requirement and NIC gets top
4365 * priority for leftovers.
4367 i = allocated - EXTRA_VECS - ofld_need - uld_need;
4368 if (i < s->max_ethqsets) {
4369 s->max_ethqsets = i;
4370 if (i < s->ethqsets)
4371 reduce_ethqs(adap, i);
4374 if (allocated < want)
4375 s->nqs_per_uld = nchan;
4377 s->nqs_per_uld = s->ofldqsets;
4380 for (i = 0; i < (s->max_ethqsets + EXTRA_VECS); ++i)
4381 adap->msix_info[i].vec = entries[i].vector;
4383 for (j = 0 ; i < allocated; ++i, j++) {
4384 adap->msix_info_ulds[j].vec = entries[i].vector;
4385 adap->msix_info_ulds[j].idx = i;
4387 adap->msix_bmap_ulds.mapsize = j;
4389 dev_info(adap->pdev_dev, "%d MSI-X vectors allocated, "
4390 "nic %d per uld %d\n",
4391 allocated, s->max_ethqsets, s->nqs_per_uld);
4399 static int init_rss(struct adapter *adap)
4404 err = t4_init_rss_mode(adap, adap->mbox);
4408 for_each_port(adap, i) {
4409 struct port_info *pi = adap2pinfo(adap, i);
4411 pi->rss = kcalloc(pi->rss_size, sizeof(u16), GFP_KERNEL);
4418 static int cxgb4_get_pcie_dev_link_caps(struct adapter *adap,
4419 enum pci_bus_speed *speed,
4420 enum pcie_link_width *width)
4422 u32 lnkcap1, lnkcap2;
4425 #define PCIE_MLW_CAP_SHIFT 4 /* start of MLW mask in link capabilities */
4427 *speed = PCI_SPEED_UNKNOWN;
4428 *width = PCIE_LNK_WIDTH_UNKNOWN;
4430 err1 = pcie_capability_read_dword(adap->pdev, PCI_EXP_LNKCAP,
4432 err2 = pcie_capability_read_dword(adap->pdev, PCI_EXP_LNKCAP2,
4434 if (!err2 && lnkcap2) { /* PCIe r3.0-compliant */
4435 if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_8_0GB)
4436 *speed = PCIE_SPEED_8_0GT;
4437 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_5_0GB)
4438 *speed = PCIE_SPEED_5_0GT;
4439 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_2_5GB)
4440 *speed = PCIE_SPEED_2_5GT;
4443 *width = (lnkcap1 & PCI_EXP_LNKCAP_MLW) >> PCIE_MLW_CAP_SHIFT;
4444 if (!lnkcap2) { /* pre-r3.0 */
4445 if (lnkcap1 & PCI_EXP_LNKCAP_SLS_5_0GB)
4446 *speed = PCIE_SPEED_5_0GT;
4447 else if (lnkcap1 & PCI_EXP_LNKCAP_SLS_2_5GB)
4448 *speed = PCIE_SPEED_2_5GT;
4452 if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN)
4453 return err1 ? err1 : err2 ? err2 : -EINVAL;
4457 static void cxgb4_check_pcie_caps(struct adapter *adap)
4459 enum pcie_link_width width, width_cap;
4460 enum pci_bus_speed speed, speed_cap;
4462 #define PCIE_SPEED_STR(speed) \
4463 (speed == PCIE_SPEED_8_0GT ? "8.0GT/s" : \
4464 speed == PCIE_SPEED_5_0GT ? "5.0GT/s" : \
4465 speed == PCIE_SPEED_2_5GT ? "2.5GT/s" : \
4468 if (cxgb4_get_pcie_dev_link_caps(adap, &speed_cap, &width_cap)) {
4469 dev_warn(adap->pdev_dev,
4470 "Unable to determine PCIe device BW capabilities\n");
4474 if (pcie_get_minimum_link(adap->pdev, &speed, &width) ||
4475 speed == PCI_SPEED_UNKNOWN || width == PCIE_LNK_WIDTH_UNKNOWN) {
4476 dev_warn(adap->pdev_dev,
4477 "Unable to determine PCI Express bandwidth.\n");
4481 dev_info(adap->pdev_dev, "PCIe link speed is %s, device supports %s\n",
4482 PCIE_SPEED_STR(speed), PCIE_SPEED_STR(speed_cap));
4483 dev_info(adap->pdev_dev, "PCIe link width is x%d, device supports x%d\n",
4485 if (speed < speed_cap || width < width_cap)
4486 dev_info(adap->pdev_dev,
4487 "A slot with more lanes and/or higher speed is "
4488 "suggested for optimal performance.\n");
4491 /* Dump basic information about the adapter */
4492 static void print_adapter_info(struct adapter *adapter)
4494 /* Device information */
4495 dev_info(adapter->pdev_dev, "Chelsio %s rev %d\n",
4496 adapter->params.vpd.id,
4497 CHELSIO_CHIP_RELEASE(adapter->params.chip));
4498 dev_info(adapter->pdev_dev, "S/N: %s, P/N: %s\n",
4499 adapter->params.vpd.sn, adapter->params.vpd.pn);
4501 /* Firmware Version */
4502 if (!adapter->params.fw_vers)
4503 dev_warn(adapter->pdev_dev, "No firmware loaded\n");
4505 dev_info(adapter->pdev_dev, "Firmware version: %u.%u.%u.%u\n",
4506 FW_HDR_FW_VER_MAJOR_G(adapter->params.fw_vers),
4507 FW_HDR_FW_VER_MINOR_G(adapter->params.fw_vers),
4508 FW_HDR_FW_VER_MICRO_G(adapter->params.fw_vers),
4509 FW_HDR_FW_VER_BUILD_G(adapter->params.fw_vers));
4511 /* Bootstrap Firmware Version. (Some adapters don't have Bootstrap
4512 * Firmware, so dev_info() is more appropriate here.)
4514 if (!adapter->params.bs_vers)
4515 dev_info(adapter->pdev_dev, "No bootstrap loaded\n");
4517 dev_info(adapter->pdev_dev, "Bootstrap version: %u.%u.%u.%u\n",
4518 FW_HDR_FW_VER_MAJOR_G(adapter->params.bs_vers),
4519 FW_HDR_FW_VER_MINOR_G(adapter->params.bs_vers),
4520 FW_HDR_FW_VER_MICRO_G(adapter->params.bs_vers),
4521 FW_HDR_FW_VER_BUILD_G(adapter->params.bs_vers));
4523 /* TP Microcode Version */
4524 if (!adapter->params.tp_vers)
4525 dev_warn(adapter->pdev_dev, "No TP Microcode loaded\n");
4527 dev_info(adapter->pdev_dev,
4528 "TP Microcode version: %u.%u.%u.%u\n",
4529 FW_HDR_FW_VER_MAJOR_G(adapter->params.tp_vers),
4530 FW_HDR_FW_VER_MINOR_G(adapter->params.tp_vers),
4531 FW_HDR_FW_VER_MICRO_G(adapter->params.tp_vers),
4532 FW_HDR_FW_VER_BUILD_G(adapter->params.tp_vers));
4534 /* Expansion ROM version */
4535 if (!adapter->params.er_vers)
4536 dev_info(adapter->pdev_dev, "No Expansion ROM loaded\n");
4538 dev_info(adapter->pdev_dev,
4539 "Expansion ROM version: %u.%u.%u.%u\n",
4540 FW_HDR_FW_VER_MAJOR_G(adapter->params.er_vers),
4541 FW_HDR_FW_VER_MINOR_G(adapter->params.er_vers),
4542 FW_HDR_FW_VER_MICRO_G(adapter->params.er_vers),
4543 FW_HDR_FW_VER_BUILD_G(adapter->params.er_vers));
4545 /* Software/Hardware configuration */
4546 dev_info(adapter->pdev_dev, "Configuration: %sNIC %s, %s capable\n",
4547 is_offload(adapter) ? "R" : "",
4548 ((adapter->flags & USING_MSIX) ? "MSI-X" :
4549 (adapter->flags & USING_MSI) ? "MSI" : ""),
4550 is_offload(adapter) ? "Offload" : "non-Offload");
4553 static void print_port_info(const struct net_device *dev)
4557 const char *spd = "";
4558 const struct port_info *pi = netdev_priv(dev);
4559 const struct adapter *adap = pi->adapter;
4561 if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_2_5GB)
4563 else if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_5_0GB)
4565 else if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_8_0GB)
4568 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_100M)
4569 bufp += sprintf(bufp, "100M/");
4570 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_1G)
4571 bufp += sprintf(bufp, "1G/");
4572 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_10G)
4573 bufp += sprintf(bufp, "10G/");
4574 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_25G)
4575 bufp += sprintf(bufp, "25G/");
4576 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_40G)
4577 bufp += sprintf(bufp, "40G/");
4578 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_100G)
4579 bufp += sprintf(bufp, "100G/");
4582 sprintf(bufp, "BASE-%s", t4_get_port_type_description(pi->port_type));
4584 netdev_info(dev, "%s: Chelsio %s (%s) %s\n",
4585 dev->name, adap->params.vpd.id, adap->name, buf);
4588 static void enable_pcie_relaxed_ordering(struct pci_dev *dev)
4590 pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_RELAX_EN);
4594 * Free the following resources:
4595 * - memory used for tables
4598 * - resources FW is holding for us
4600 static void free_some_resources(struct adapter *adapter)
4604 kvfree(adapter->l2t);
4605 t4_cleanup_sched(adapter);
4606 kvfree(adapter->tids.tid_tab);
4607 cxgb4_cleanup_tc_u32(adapter);
4608 kfree(adapter->sge.egr_map);
4609 kfree(adapter->sge.ingr_map);
4610 kfree(adapter->sge.starving_fl);
4611 kfree(adapter->sge.txq_maperr);
4612 #ifdef CONFIG_DEBUG_FS
4613 kfree(adapter->sge.blocked_fl);
4615 disable_msi(adapter);
4617 for_each_port(adapter, i)
4618 if (adapter->port[i]) {
4619 struct port_info *pi = adap2pinfo(adapter, i);
4622 t4_free_vi(adapter, adapter->mbox, adapter->pf,
4624 kfree(adap2pinfo(adapter, i)->rss);
4625 free_netdev(adapter->port[i]);
4627 if (adapter->flags & FW_OK)
4628 t4_fw_bye(adapter, adapter->pf);
4631 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
4632 #define VLAN_FEAT (NETIF_F_SG | NETIF_F_IP_CSUM | TSO_FLAGS | \
4633 NETIF_F_IPV6_CSUM | NETIF_F_HIGHDMA)
4634 #define SEGMENT_SIZE 128
4636 static int get_chip_type(struct pci_dev *pdev, u32 pl_rev)
4640 /* Retrieve adapter's device ID */
4641 pci_read_config_word(pdev, PCI_DEVICE_ID, &device_id);
4643 switch (device_id >> 12) {
4645 return CHELSIO_CHIP_CODE(CHELSIO_T4, pl_rev);
4647 return CHELSIO_CHIP_CODE(CHELSIO_T5, pl_rev);
4649 return CHELSIO_CHIP_CODE(CHELSIO_T6, pl_rev);
4651 dev_err(&pdev->dev, "Device %d is not supported\n",
4657 #ifdef CONFIG_PCI_IOV
4658 static void dummy_setup(struct net_device *dev)
4660 dev->type = ARPHRD_NONE;
4662 dev->hard_header_len = 0;
4664 dev->tx_queue_len = 0;
4665 dev->flags |= IFF_NOARP;
4666 dev->priv_flags |= IFF_NO_QUEUE;
4668 /* Initialize the device structure. */
4669 dev->netdev_ops = &cxgb4_mgmt_netdev_ops;
4670 dev->ethtool_ops = &cxgb4_mgmt_ethtool_ops;
4671 dev->needs_free_netdev = true;
4674 static int config_mgmt_dev(struct pci_dev *pdev)
4676 struct adapter *adap = pci_get_drvdata(pdev);
4677 struct net_device *netdev;
4678 struct port_info *pi;
4679 char name[IFNAMSIZ];
4682 snprintf(name, IFNAMSIZ, "mgmtpf%d%d", adap->adap_idx, adap->pf);
4683 netdev = alloc_netdev(sizeof(struct port_info), name, NET_NAME_UNKNOWN,
4688 pi = netdev_priv(netdev);
4690 pi->port_id = adap->pf % adap->params.nports;
4691 SET_NETDEV_DEV(netdev, &pdev->dev);
4693 adap->port[0] = netdev;
4695 err = register_netdev(adap->port[0]);
4697 pr_info("Unable to register VF mgmt netdev %s\n", name);
4698 free_netdev(adap->port[0]);
4699 adap->port[0] = NULL;
4705 static int cxgb4_iov_configure(struct pci_dev *pdev, int num_vfs)
4707 struct adapter *adap = pci_get_drvdata(pdev);
4709 int current_vfs = pci_num_vf(pdev);
4712 pcie_fw = readl(adap->regs + PCIE_FW_A);
4713 /* Check if cxgb4 is the MASTER and fw is initialized */
4714 if (!(pcie_fw & PCIE_FW_INIT_F) ||
4715 !(pcie_fw & PCIE_FW_MASTER_VLD_F) ||
4716 PCIE_FW_MASTER_G(pcie_fw) != 4) {
4717 dev_warn(&pdev->dev,
4718 "cxgb4 driver needs to be MASTER to support SRIOV\n");
4722 /* If any of the VF's is already assigned to Guest OS, then
4723 * SRIOV for the same cannot be modified
4725 if (current_vfs && pci_vfs_assigned(pdev)) {
4727 "Cannot modify SR-IOV while VFs are assigned\n");
4728 num_vfs = current_vfs;
4732 /* Disable SRIOV when zero is passed.
4733 * One needs to disable SRIOV before modifying it, else
4734 * stack throws the below warning:
4735 * " 'n' VFs already enabled. Disable before enabling 'm' VFs."
4738 pci_disable_sriov(pdev);
4739 if (adap->port[0]) {
4740 unregister_netdev(adap->port[0]);
4741 adap->port[0] = NULL;
4743 /* free VF resources */
4744 kfree(adap->vfinfo);
4745 adap->vfinfo = NULL;
4750 if (num_vfs != current_vfs) {
4751 err = pci_enable_sriov(pdev, num_vfs);
4755 adap->num_vfs = num_vfs;
4756 err = config_mgmt_dev(pdev);
4761 adap->vfinfo = kcalloc(adap->num_vfs,
4762 sizeof(struct vf_info), GFP_KERNEL);
4764 fill_vf_station_mac_addr(adap);
4769 static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
4771 int func, i, err, s_qpp, qpp, num_seg;
4772 struct port_info *pi;
4773 bool highdma = false;
4774 struct adapter *adapter = NULL;
4775 struct net_device *netdev;
4778 enum chip_type chip;
4779 static int adap_idx = 1;
4780 #ifdef CONFIG_PCI_IOV
4784 printk_once(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
4786 err = pci_request_regions(pdev, KBUILD_MODNAME);
4788 /* Just info, some other driver may have claimed the device. */
4789 dev_info(&pdev->dev, "cannot obtain PCI resources\n");
4793 err = pci_enable_device(pdev);
4795 dev_err(&pdev->dev, "cannot enable PCI device\n");
4796 goto out_release_regions;
4799 regs = pci_ioremap_bar(pdev, 0);
4801 dev_err(&pdev->dev, "cannot map device registers\n");
4803 goto out_disable_device;
4806 err = t4_wait_dev_ready(regs);
4808 goto out_unmap_bar0;
4810 /* We control everything through one PF */
4811 whoami = readl(regs + PL_WHOAMI_A);
4812 pl_rev = REV_G(readl(regs + PL_REV_A));
4813 chip = get_chip_type(pdev, pl_rev);
4814 func = CHELSIO_CHIP_VERSION(chip) <= CHELSIO_T5 ?
4815 SOURCEPF_G(whoami) : T6_SOURCEPF_G(whoami);
4816 if (func != ent->driver_data) {
4817 #ifndef CONFIG_PCI_IOV
4820 pci_disable_device(pdev);
4821 pci_save_state(pdev); /* to restore SR-IOV later */
4825 if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
4827 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
4829 dev_err(&pdev->dev, "unable to obtain 64-bit DMA for "
4830 "coherent allocations\n");
4831 goto out_unmap_bar0;
4834 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
4836 dev_err(&pdev->dev, "no usable DMA configuration\n");
4837 goto out_unmap_bar0;
4841 pci_enable_pcie_error_reporting(pdev);
4842 enable_pcie_relaxed_ordering(pdev);
4843 pci_set_master(pdev);
4844 pci_save_state(pdev);
4846 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
4849 goto out_unmap_bar0;
4853 adapter->workq = create_singlethread_workqueue("cxgb4");
4854 if (!adapter->workq) {
4856 goto out_free_adapter;
4859 adapter->mbox_log = kzalloc(sizeof(*adapter->mbox_log) +
4860 (sizeof(struct mbox_cmd) *
4861 T4_OS_LOG_MBOX_CMDS),
4863 if (!adapter->mbox_log) {
4865 goto out_free_adapter;
4867 adapter->mbox_log->size = T4_OS_LOG_MBOX_CMDS;
4869 /* PCI device has been enabled */
4870 adapter->flags |= DEV_ENABLED;
4872 adapter->regs = regs;
4873 adapter->pdev = pdev;
4874 adapter->pdev_dev = &pdev->dev;
4875 adapter->name = pci_name(pdev);
4876 adapter->mbox = func;
4878 adapter->msg_enable = DFLT_MSG_ENABLE;
4879 memset(adapter->chan_map, 0xff, sizeof(adapter->chan_map));
4881 spin_lock_init(&adapter->stats_lock);
4882 spin_lock_init(&adapter->tid_release_lock);
4883 spin_lock_init(&adapter->win0_lock);
4884 spin_lock_init(&adapter->mbox_lock);
4886 INIT_LIST_HEAD(&adapter->mlist.list);
4888 INIT_WORK(&adapter->tid_release_task, process_tid_release_list);
4889 INIT_WORK(&adapter->db_full_task, process_db_full);
4890 INIT_WORK(&adapter->db_drop_task, process_db_drop);
4892 err = t4_prep_adapter(adapter);
4894 goto out_free_adapter;
4897 if (!is_t4(adapter->params.chip)) {
4898 s_qpp = (QUEUESPERPAGEPF0_S +
4899 (QUEUESPERPAGEPF1_S - QUEUESPERPAGEPF0_S) *
4901 qpp = 1 << QUEUESPERPAGEPF0_G(t4_read_reg(adapter,
4902 SGE_EGRESS_QUEUES_PER_PAGE_PF_A) >> s_qpp);
4903 num_seg = PAGE_SIZE / SEGMENT_SIZE;
4905 /* Each segment size is 128B. Write coalescing is enabled only
4906 * when SGE_EGRESS_QUEUES_PER_PAGE_PF reg value for the
4907 * queue is less no of segments that can be accommodated in
4910 if (qpp > num_seg) {
4912 "Incorrect number of egress queues per page\n");
4914 goto out_free_adapter;
4916 adapter->bar2 = ioremap_wc(pci_resource_start(pdev, 2),
4917 pci_resource_len(pdev, 2));
4918 if (!adapter->bar2) {
4919 dev_err(&pdev->dev, "cannot map device bar2 region\n");
4921 goto out_free_adapter;
4925 setup_memwin(adapter);
4926 err = adap_init0(adapter);
4927 #ifdef CONFIG_DEBUG_FS
4928 bitmap_zero(adapter->sge.blocked_fl, adapter->sge.egr_sz);
4930 setup_memwin_rdma(adapter);
4934 /* configure SGE_STAT_CFG_A to read WC stats */
4935 if (!is_t4(adapter->params.chip))
4936 t4_write_reg(adapter, SGE_STAT_CFG_A, STATSOURCE_T5_V(7) |
4937 (is_t5(adapter->params.chip) ? STATMODE_V(0) :
4940 for_each_port(adapter, i) {
4941 netdev = alloc_etherdev_mq(sizeof(struct port_info),
4948 SET_NETDEV_DEV(netdev, &pdev->dev);
4950 adapter->port[i] = netdev;
4951 pi = netdev_priv(netdev);
4952 pi->adapter = adapter;
4953 pi->xact_addr_filt = -1;
4955 netdev->irq = pdev->irq;
4957 netdev->hw_features = NETIF_F_SG | TSO_FLAGS |
4958 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
4959 NETIF_F_RXCSUM | NETIF_F_RXHASH |
4960 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
4963 netdev->hw_features |= NETIF_F_HIGHDMA;
4964 netdev->features |= netdev->hw_features;
4965 netdev->vlan_features = netdev->features & VLAN_FEAT;
4967 netdev->priv_flags |= IFF_UNICAST_FLT;
4969 /* MTU range: 81 - 9600 */
4970 netdev->min_mtu = 81;
4971 netdev->max_mtu = MAX_MTU;
4973 netdev->netdev_ops = &cxgb4_netdev_ops;
4974 #ifdef CONFIG_CHELSIO_T4_DCB
4975 netdev->dcbnl_ops = &cxgb4_dcb_ops;
4976 cxgb4_dcb_state_init(netdev);
4978 cxgb4_set_ethtool_ops(netdev);
4981 pci_set_drvdata(pdev, adapter);
4983 if (adapter->flags & FW_OK) {
4984 err = t4_port_init(adapter, func, func, 0);
4987 } else if (adapter->params.nports == 1) {
4988 /* If we don't have a connection to the firmware -- possibly
4989 * because of an error -- grab the raw VPD parameters so we
4990 * can set the proper MAC Address on the debug network
4991 * interface that we've created.
4993 u8 hw_addr[ETH_ALEN];
4994 u8 *na = adapter->params.vpd.na;
4996 err = t4_get_raw_vpd_params(adapter, &adapter->params.vpd);
4998 for (i = 0; i < ETH_ALEN; i++)
4999 hw_addr[i] = (hex2val(na[2 * i + 0]) * 16 +
5000 hex2val(na[2 * i + 1]));
5001 t4_set_hw_addr(adapter, 0, hw_addr);
5005 /* Configure queues and allocate tables now, they can be needed as
5006 * soon as the first register_netdev completes.
5008 cfg_queues(adapter);
5010 adapter->l2t = t4_init_l2t(adapter->l2t_start, adapter->l2t_end);
5011 if (!adapter->l2t) {
5012 /* We tolerate a lack of L2T, giving up some functionality */
5013 dev_warn(&pdev->dev, "could not allocate L2T, continuing\n");
5014 adapter->params.offload = 0;
5017 #if IS_ENABLED(CONFIG_IPV6)
5018 if ((CHELSIO_CHIP_VERSION(adapter->params.chip) <= CHELSIO_T5) &&
5019 (!(t4_read_reg(adapter, LE_DB_CONFIG_A) & ASLIPCOMPEN_F))) {
5020 /* CLIP functionality is not present in hardware,
5021 * hence disable all offload features
5023 dev_warn(&pdev->dev,
5024 "CLIP not enabled in hardware, continuing\n");
5025 adapter->params.offload = 0;
5027 adapter->clipt = t4_init_clip_tbl(adapter->clipt_start,
5028 adapter->clipt_end);
5029 if (!adapter->clipt) {
5030 /* We tolerate a lack of clip_table, giving up
5031 * some functionality
5033 dev_warn(&pdev->dev,
5034 "could not allocate Clip table, continuing\n");
5035 adapter->params.offload = 0;
5040 for_each_port(adapter, i) {
5041 pi = adap2pinfo(adapter, i);
5042 pi->sched_tbl = t4_init_sched(adapter->params.nsched_cls);
5044 dev_warn(&pdev->dev,
5045 "could not activate scheduling on port %d\n",
5049 if (tid_init(&adapter->tids) < 0) {
5050 dev_warn(&pdev->dev, "could not allocate TID table, "
5052 adapter->params.offload = 0;
5054 adapter->tc_u32 = cxgb4_init_tc_u32(adapter);
5055 if (!adapter->tc_u32)
5056 dev_warn(&pdev->dev,
5057 "could not offload tc u32, continuing\n");
5060 if (is_offload(adapter)) {
5061 if (t4_read_reg(adapter, LE_DB_CONFIG_A) & HASHEN_F) {
5062 u32 hash_base, hash_reg;
5064 if (chip <= CHELSIO_T5) {
5065 hash_reg = LE_DB_TID_HASHBASE_A;
5066 hash_base = t4_read_reg(adapter, hash_reg);
5067 adapter->tids.hash_base = hash_base / 4;
5069 hash_reg = T6_LE_DB_HASH_TID_BASE_A;
5070 hash_base = t4_read_reg(adapter, hash_reg);
5071 adapter->tids.hash_base = hash_base;
5076 /* See what interrupts we'll be using */
5077 if (msi > 1 && enable_msix(adapter) == 0)
5078 adapter->flags |= USING_MSIX;
5079 else if (msi > 0 && pci_enable_msi(pdev) == 0) {
5080 adapter->flags |= USING_MSI;
5082 free_msix_info(adapter);
5085 /* check for PCI Express bandwidth capabiltites */
5086 cxgb4_check_pcie_caps(adapter);
5088 err = init_rss(adapter);
5093 * The card is now ready to go. If any errors occur during device
5094 * registration we do not fail the whole card but rather proceed only
5095 * with the ports we manage to register successfully. However we must
5096 * register at least one net device.
5098 for_each_port(adapter, i) {
5099 pi = adap2pinfo(adapter, i);
5100 adapter->port[i]->dev_port = pi->lport;
5101 netif_set_real_num_tx_queues(adapter->port[i], pi->nqsets);
5102 netif_set_real_num_rx_queues(adapter->port[i], pi->nqsets);
5104 netif_carrier_off(adapter->port[i]);
5106 err = register_netdev(adapter->port[i]);
5109 adapter->chan_map[pi->tx_chan] = i;
5110 print_port_info(adapter->port[i]);
5113 dev_err(&pdev->dev, "could not register any net devices\n");
5117 dev_warn(&pdev->dev, "only %d net devices registered\n", i);
5121 if (cxgb4_debugfs_root) {
5122 adapter->debugfs_root = debugfs_create_dir(pci_name(pdev),
5123 cxgb4_debugfs_root);
5124 setup_debugfs(adapter);
5127 /* PCIe EEH recovery on powerpc platforms needs fundamental reset */
5128 pdev->needs_freset = 1;
5130 if (is_uld(adapter)) {
5131 mutex_lock(&uld_mutex);
5132 list_add_tail(&adapter->list_node, &adapter_list);
5133 mutex_unlock(&uld_mutex);
5136 print_adapter_info(adapter);
5137 setup_fw_sge_queues(adapter);
5141 #ifdef CONFIG_PCI_IOV
5142 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
5145 goto free_pci_region;
5148 adapter->pdev = pdev;
5149 adapter->pdev_dev = &pdev->dev;
5150 adapter->name = pci_name(pdev);
5151 adapter->mbox = func;
5153 adapter->regs = regs;
5154 adapter->adap_idx = adap_idx;
5155 adapter->mbox_log = kzalloc(sizeof(*adapter->mbox_log) +
5156 (sizeof(struct mbox_cmd) *
5157 T4_OS_LOG_MBOX_CMDS),
5159 if (!adapter->mbox_log) {
5163 spin_lock_init(&adapter->mbox_lock);
5164 INIT_LIST_HEAD(&adapter->mlist.list);
5166 v = FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
5167 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_PORTVEC);
5168 err = t4_query_params(adapter, adapter->mbox, adapter->pf, 0, 1,
5171 dev_err(adapter->pdev_dev, "Could not fetch port params\n");
5175 adapter->params.nports = hweight32(port_vec);
5176 pci_set_drvdata(pdev, adapter);
5180 kfree(adapter->mbox_log);
5185 pci_disable_sriov(pdev);
5186 pci_release_regions(pdev);
5193 free_some_resources(adapter);
5194 if (adapter->flags & USING_MSIX)
5195 free_msix_info(adapter);
5196 if (adapter->num_uld || adapter->num_ofld_uld)
5197 t4_uld_mem_free(adapter);
5199 if (!is_t4(adapter->params.chip))
5200 iounmap(adapter->bar2);
5203 destroy_workqueue(adapter->workq);
5205 kfree(adapter->mbox_log);
5210 pci_disable_pcie_error_reporting(pdev);
5211 pci_disable_device(pdev);
5212 out_release_regions:
5213 pci_release_regions(pdev);
5217 static void remove_one(struct pci_dev *pdev)
5219 struct adapter *adapter = pci_get_drvdata(pdev);
5222 pci_release_regions(pdev);
5226 if (adapter->pf == 4) {
5229 /* Tear down per-adapter Work Queue first since it can contain
5230 * references to our adapter data structure.
5232 destroy_workqueue(adapter->workq);
5234 if (is_uld(adapter))
5235 detach_ulds(adapter);
5237 disable_interrupts(adapter);
5239 for_each_port(adapter, i)
5240 if (adapter->port[i]->reg_state == NETREG_REGISTERED)
5241 unregister_netdev(adapter->port[i]);
5243 debugfs_remove_recursive(adapter->debugfs_root);
5245 /* If we allocated filters, free up state associated with any
5248 clear_all_filters(adapter);
5250 if (adapter->flags & FULL_INIT_DONE)
5253 if (adapter->flags & USING_MSIX)
5254 free_msix_info(adapter);
5255 if (adapter->num_uld || adapter->num_ofld_uld)
5256 t4_uld_mem_free(adapter);
5257 free_some_resources(adapter);
5258 #if IS_ENABLED(CONFIG_IPV6)
5259 t4_cleanup_clip_tbl(adapter);
5261 iounmap(adapter->regs);
5262 if (!is_t4(adapter->params.chip))
5263 iounmap(adapter->bar2);
5264 pci_disable_pcie_error_reporting(pdev);
5265 if ((adapter->flags & DEV_ENABLED)) {
5266 pci_disable_device(pdev);
5267 adapter->flags &= ~DEV_ENABLED;
5269 pci_release_regions(pdev);
5270 kfree(adapter->mbox_log);
5274 #ifdef CONFIG_PCI_IOV
5276 if (adapter->port[0])
5277 unregister_netdev(adapter->port[0]);
5278 iounmap(adapter->regs);
5279 kfree(adapter->vfinfo);
5280 kfree(adapter->mbox_log);
5282 pci_disable_sriov(pdev);
5283 pci_release_regions(pdev);
5288 /* "Shutdown" quiesces the device, stopping Ingress Packet and Interrupt
5289 * delivery. This is essentially a stripped down version of the PCI remove()
5290 * function where we do the minimal amount of work necessary to shutdown any
5293 static void shutdown_one(struct pci_dev *pdev)
5295 struct adapter *adapter = pci_get_drvdata(pdev);
5297 /* As with remove_one() above (see extended comment), we only want do
5298 * do cleanup on PCI Devices which went all the way through init_one()
5302 pci_release_regions(pdev);
5306 if (adapter->pf == 4) {
5309 for_each_port(adapter, i)
5310 if (adapter->port[i]->reg_state == NETREG_REGISTERED)
5311 cxgb_close(adapter->port[i]);
5313 t4_uld_clean_up(adapter);
5314 disable_interrupts(adapter);
5315 disable_msi(adapter);
5317 t4_sge_stop(adapter);
5318 if (adapter->flags & FW_OK)
5319 t4_fw_bye(adapter, adapter->mbox);
5321 #ifdef CONFIG_PCI_IOV
5323 if (adapter->port[0])
5324 unregister_netdev(adapter->port[0]);
5325 iounmap(adapter->regs);
5326 kfree(adapter->vfinfo);
5327 kfree(adapter->mbox_log);
5329 pci_disable_sriov(pdev);
5330 pci_release_regions(pdev);
5335 static struct pci_driver cxgb4_driver = {
5336 .name = KBUILD_MODNAME,
5337 .id_table = cxgb4_pci_tbl,
5339 .remove = remove_one,
5340 .shutdown = shutdown_one,
5341 #ifdef CONFIG_PCI_IOV
5342 .sriov_configure = cxgb4_iov_configure,
5344 .err_handler = &cxgb4_eeh,
5347 static int __init cxgb4_init_module(void)
5351 /* Debugfs support is optional, just warn if this fails */
5352 cxgb4_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
5353 if (!cxgb4_debugfs_root)
5354 pr_warn("could not create debugfs entry, continuing\n");
5356 ret = pci_register_driver(&cxgb4_driver);
5358 debugfs_remove(cxgb4_debugfs_root);
5360 #if IS_ENABLED(CONFIG_IPV6)
5361 if (!inet6addr_registered) {
5362 register_inet6addr_notifier(&cxgb4_inet6addr_notifier);
5363 inet6addr_registered = true;
5370 static void __exit cxgb4_cleanup_module(void)
5372 #if IS_ENABLED(CONFIG_IPV6)
5373 if (inet6addr_registered) {
5374 unregister_inet6addr_notifier(&cxgb4_inet6addr_notifier);
5375 inet6addr_registered = false;
5378 pci_unregister_driver(&cxgb4_driver);
5379 debugfs_remove(cxgb4_debugfs_root); /* NULL ok */
5382 module_init(cxgb4_init_module);
5383 module_exit(cxgb4_cleanup_module);