2 * This file is part of the Chelsio T4 Ethernet driver for Linux.
4 * Copyright (c) 2003-2010 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 <asm/uaccess.h>
72 #include <../drivers/net/bonding/bonding.h>
77 #define DRV_VERSION "2.0.0-ko"
78 #define DRV_DESC "Chelsio T4/T5 Network Driver"
81 * Max interrupt hold-off timer value in us. Queues fall back to this value
82 * under extreme memory pressure so it's largish to give the system time to
85 #define MAX_SGE_TIMERVAL 200U
89 * Physical Function provisioning constants.
91 PFRES_NVI = 4, /* # of Virtual Interfaces */
92 PFRES_NETHCTRL = 128, /* # of EQs used for ETH or CTRL Qs */
93 PFRES_NIQFLINT = 128, /* # of ingress Qs/w Free List(s)/intr
95 PFRES_NEQ = 256, /* # of egress queues */
96 PFRES_NIQ = 0, /* # of ingress queues */
97 PFRES_TC = 0, /* PCI-E traffic class */
98 PFRES_NEXACTF = 128, /* # of exact MPS filters */
100 PFRES_R_CAPS = FW_CMD_CAP_PF,
101 PFRES_WX_CAPS = FW_CMD_CAP_PF,
103 #ifdef CONFIG_PCI_IOV
105 * Virtual Function provisioning constants. We need two extra Ingress
106 * Queues with Interrupt capability to serve as the VF's Firmware
107 * Event Queue and Forwarded Interrupt Queue (when using MSI mode) --
108 * neither will have Free Lists associated with them). For each
109 * Ethernet/Control Egress Queue and for each Free List, we need an
112 VFRES_NPORTS = 1, /* # of "ports" per VF */
113 VFRES_NQSETS = 2, /* # of "Queue Sets" per VF */
115 VFRES_NVI = VFRES_NPORTS, /* # of Virtual Interfaces */
116 VFRES_NETHCTRL = VFRES_NQSETS, /* # of EQs used for ETH or CTRL Qs */
117 VFRES_NIQFLINT = VFRES_NQSETS+2,/* # of ingress Qs/w Free List(s)/intr */
118 VFRES_NEQ = VFRES_NQSETS*2, /* # of egress queues */
119 VFRES_NIQ = 0, /* # of non-fl/int ingress queues */
120 VFRES_TC = 0, /* PCI-E traffic class */
121 VFRES_NEXACTF = 16, /* # of exact MPS filters */
123 VFRES_R_CAPS = FW_CMD_CAP_DMAQ|FW_CMD_CAP_VF|FW_CMD_CAP_PORT,
124 VFRES_WX_CAPS = FW_CMD_CAP_DMAQ|FW_CMD_CAP_VF,
129 * Provide a Port Access Rights Mask for the specified PF/VF. This is very
130 * static and likely not to be useful in the long run. We really need to
131 * implement some form of persistent configuration which the firmware
134 static unsigned int pfvfres_pmask(struct adapter *adapter,
135 unsigned int pf, unsigned int vf)
137 unsigned int portn, portvec;
140 * Give PF's access to all of the ports.
143 return FW_PFVF_CMD_PMASK_MASK;
146 * For VFs, we'll assign them access to the ports based purely on the
147 * PF. We assign active ports in order, wrapping around if there are
148 * fewer active ports than PFs: e.g. active port[pf % nports].
149 * Unfortunately the adapter's port_info structs haven't been
150 * initialized yet so we have to compute this.
152 if (adapter->params.nports == 0)
155 portn = pf % adapter->params.nports;
156 portvec = adapter->params.portvec;
159 * Isolate the lowest set bit in the port vector. If we're at
160 * the port number that we want, return that as the pmask.
161 * otherwise mask that bit out of the port vector and
162 * decrement our port number ...
164 unsigned int pmask = portvec ^ (portvec & (portvec-1));
174 MAX_TXQ_ENTRIES = 16384,
175 MAX_CTRL_TXQ_ENTRIES = 1024,
176 MAX_RSPQ_ENTRIES = 16384,
177 MAX_RX_BUFFERS = 16384,
178 MIN_TXQ_ENTRIES = 32,
179 MIN_CTRL_TXQ_ENTRIES = 32,
180 MIN_RSPQ_ENTRIES = 128,
184 /* Host shadow copy of ingress filter entry. This is in host native format
185 * and doesn't match the ordering or bit order, etc. of the hardware of the
186 * firmware command. The use of bit-field structure elements is purely to
187 * remind ourselves of the field size limitations and save memory in the case
188 * where the filter table is large.
190 struct filter_entry {
191 /* Administrative fields for filter.
193 u32 valid:1; /* filter allocated and valid */
194 u32 locked:1; /* filter is administratively locked */
196 u32 pending:1; /* filter action is pending firmware reply */
197 u32 smtidx:8; /* Source MAC Table index for smac */
198 struct l2t_entry *l2t; /* Layer Two Table entry for dmac */
200 /* The filter itself. Most of this is a straight copy of information
201 * provided by the extended ioctl(). Some fields are translated to
202 * internal forms -- for instance the Ingress Queue ID passed in from
203 * the ioctl() is translated into the Absolute Ingress Queue ID.
205 struct ch_filter_specification fs;
208 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
209 NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
210 NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
212 #define CH_DEVICE(devid, data) { PCI_VDEVICE(CHELSIO, devid), (data) }
214 static DEFINE_PCI_DEVICE_TABLE(cxgb4_pci_tbl) = {
215 CH_DEVICE(0xa000, 0), /* PE10K */
216 CH_DEVICE(0x4001, -1),
217 CH_DEVICE(0x4002, -1),
218 CH_DEVICE(0x4003, -1),
219 CH_DEVICE(0x4004, -1),
220 CH_DEVICE(0x4005, -1),
221 CH_DEVICE(0x4006, -1),
222 CH_DEVICE(0x4007, -1),
223 CH_DEVICE(0x4008, -1),
224 CH_DEVICE(0x4009, -1),
225 CH_DEVICE(0x400a, -1),
226 CH_DEVICE(0x4401, 4),
227 CH_DEVICE(0x4402, 4),
228 CH_DEVICE(0x4403, 4),
229 CH_DEVICE(0x4404, 4),
230 CH_DEVICE(0x4405, 4),
231 CH_DEVICE(0x4406, 4),
232 CH_DEVICE(0x4407, 4),
233 CH_DEVICE(0x4408, 4),
234 CH_DEVICE(0x4409, 4),
235 CH_DEVICE(0x440a, 4),
236 CH_DEVICE(0x440d, 4),
237 CH_DEVICE(0x440e, 4),
238 CH_DEVICE(0x5001, 4),
239 CH_DEVICE(0x5002, 4),
240 CH_DEVICE(0x5003, 4),
241 CH_DEVICE(0x5004, 4),
242 CH_DEVICE(0x5005, 4),
243 CH_DEVICE(0x5006, 4),
244 CH_DEVICE(0x5007, 4),
245 CH_DEVICE(0x5008, 4),
246 CH_DEVICE(0x5009, 4),
247 CH_DEVICE(0x500A, 4),
248 CH_DEVICE(0x500B, 4),
249 CH_DEVICE(0x500C, 4),
250 CH_DEVICE(0x500D, 4),
251 CH_DEVICE(0x500E, 4),
252 CH_DEVICE(0x500F, 4),
253 CH_DEVICE(0x5010, 4),
254 CH_DEVICE(0x5011, 4),
255 CH_DEVICE(0x5012, 4),
256 CH_DEVICE(0x5013, 4),
257 CH_DEVICE(0x5014, 4),
258 CH_DEVICE(0x5015, 4),
259 CH_DEVICE(0x5401, 4),
260 CH_DEVICE(0x5402, 4),
261 CH_DEVICE(0x5403, 4),
262 CH_DEVICE(0x5404, 4),
263 CH_DEVICE(0x5405, 4),
264 CH_DEVICE(0x5406, 4),
265 CH_DEVICE(0x5407, 4),
266 CH_DEVICE(0x5408, 4),
267 CH_DEVICE(0x5409, 4),
268 CH_DEVICE(0x540A, 4),
269 CH_DEVICE(0x540B, 4),
270 CH_DEVICE(0x540C, 4),
271 CH_DEVICE(0x540D, 4),
272 CH_DEVICE(0x540E, 4),
273 CH_DEVICE(0x540F, 4),
274 CH_DEVICE(0x5410, 4),
275 CH_DEVICE(0x5411, 4),
276 CH_DEVICE(0x5412, 4),
277 CH_DEVICE(0x5413, 4),
278 CH_DEVICE(0x5414, 4),
279 CH_DEVICE(0x5415, 4),
283 #define FW4_FNAME "cxgb4/t4fw.bin"
284 #define FW5_FNAME "cxgb4/t5fw.bin"
285 #define FW4_CFNAME "cxgb4/t4-config.txt"
286 #define FW5_CFNAME "cxgb4/t5-config.txt"
288 MODULE_DESCRIPTION(DRV_DESC);
289 MODULE_AUTHOR("Chelsio Communications");
290 MODULE_LICENSE("Dual BSD/GPL");
291 MODULE_VERSION(DRV_VERSION);
292 MODULE_DEVICE_TABLE(pci, cxgb4_pci_tbl);
293 MODULE_FIRMWARE(FW4_FNAME);
294 MODULE_FIRMWARE(FW5_FNAME);
297 * Normally we're willing to become the firmware's Master PF but will be happy
298 * if another PF has already become the Master and initialized the adapter.
299 * Setting "force_init" will cause this driver to forcibly establish itself as
300 * the Master PF and initialize the adapter.
302 static uint force_init;
304 module_param(force_init, uint, 0644);
305 MODULE_PARM_DESC(force_init, "Forcibly become Master PF and initialize adapter");
308 * Normally if the firmware we connect to has Configuration File support, we
309 * use that and only fall back to the old Driver-based initialization if the
310 * Configuration File fails for some reason. If force_old_init is set, then
311 * we'll always use the old Driver-based initialization sequence.
313 static uint force_old_init;
315 module_param(force_old_init, uint, 0644);
316 MODULE_PARM_DESC(force_old_init, "Force old initialization sequence");
318 static int dflt_msg_enable = DFLT_MSG_ENABLE;
320 module_param(dflt_msg_enable, int, 0644);
321 MODULE_PARM_DESC(dflt_msg_enable, "Chelsio T4 default message enable bitmap");
324 * The driver uses the best interrupt scheme available on a platform in the
325 * order MSI-X, MSI, legacy INTx interrupts. This parameter determines which
326 * of these schemes the driver may consider as follows:
328 * msi = 2: choose from among all three options
329 * msi = 1: only consider MSI and INTx interrupts
330 * msi = 0: force INTx interrupts
334 module_param(msi, int, 0644);
335 MODULE_PARM_DESC(msi, "whether to use INTx (0), MSI (1) or MSI-X (2)");
338 * Queue interrupt hold-off timer values. Queues default to the first of these
341 static unsigned int intr_holdoff[SGE_NTIMERS - 1] = { 5, 10, 20, 50, 100 };
343 module_param_array(intr_holdoff, uint, NULL, 0644);
344 MODULE_PARM_DESC(intr_holdoff, "values for queue interrupt hold-off timers "
345 "0..4 in microseconds");
347 static unsigned int intr_cnt[SGE_NCOUNTERS - 1] = { 4, 8, 16 };
349 module_param_array(intr_cnt, uint, NULL, 0644);
350 MODULE_PARM_DESC(intr_cnt,
351 "thresholds 1..3 for queue interrupt packet counters");
354 * Normally we tell the chip to deliver Ingress Packets into our DMA buffers
355 * offset by 2 bytes in order to have the IP headers line up on 4-byte
356 * boundaries. This is a requirement for many architectures which will throw
357 * a machine check fault if an attempt is made to access one of the 4-byte IP
358 * header fields on a non-4-byte boundary. And it's a major performance issue
359 * even on some architectures which allow it like some implementations of the
360 * x86 ISA. However, some architectures don't mind this and for some very
361 * edge-case performance sensitive applications (like forwarding large volumes
362 * of small packets), setting this DMA offset to 0 will decrease the number of
363 * PCI-E Bus transfers enough to measurably affect performance.
365 static int rx_dma_offset = 2;
369 #ifdef CONFIG_PCI_IOV
370 module_param(vf_acls, bool, 0644);
371 MODULE_PARM_DESC(vf_acls, "if set enable virtualization L2 ACL enforcement");
373 /* Configure the number of PCI-E Virtual Function which are to be instantiated
374 * on SR-IOV Capable Physical Functions.
376 static unsigned int num_vf[NUM_OF_PF_WITH_SRIOV];
378 module_param_array(num_vf, uint, NULL, 0644);
379 MODULE_PARM_DESC(num_vf, "number of VFs for each of PFs 0-3");
383 * The filter TCAM has a fixed portion and a variable portion. The fixed
384 * portion can match on source/destination IP IPv4/IPv6 addresses and TCP/UDP
385 * ports. The variable portion is 36 bits which can include things like Exact
386 * Match MAC Index (9 bits), Ether Type (16 bits), IP Protocol (8 bits),
387 * [Inner] VLAN Tag (17 bits), etc. which, if all were somehow selected, would
388 * far exceed the 36-bit budget for this "compressed" header portion of the
389 * filter. Thus, we have a scarce resource which must be carefully managed.
391 * By default we set this up to mostly match the set of filter matching
392 * capabilities of T3 but with accommodations for some of T4's more
393 * interesting features:
395 * { IP Fragment (1), MPS Match Type (3), IP Protocol (8),
396 * [Inner] VLAN (17), Port (3), FCoE (1) }
399 TP_VLAN_PRI_MAP_DEFAULT = HW_TPL_FR_MT_PR_IV_P_FC,
400 TP_VLAN_PRI_MAP_FIRST = FCOE_SHIFT,
401 TP_VLAN_PRI_MAP_LAST = FRAGMENTATION_SHIFT,
404 static unsigned int tp_vlan_pri_map = TP_VLAN_PRI_MAP_DEFAULT;
406 module_param(tp_vlan_pri_map, uint, 0644);
407 MODULE_PARM_DESC(tp_vlan_pri_map, "global compressed filter configuration");
409 static struct dentry *cxgb4_debugfs_root;
411 static LIST_HEAD(adapter_list);
412 static DEFINE_MUTEX(uld_mutex);
413 /* Adapter list to be accessed from atomic context */
414 static LIST_HEAD(adap_rcu_list);
415 static DEFINE_SPINLOCK(adap_rcu_lock);
416 static struct cxgb4_uld_info ulds[CXGB4_ULD_MAX];
417 static const char *uld_str[] = { "RDMA", "iSCSI" };
419 static void link_report(struct net_device *dev)
421 if (!netif_carrier_ok(dev))
422 netdev_info(dev, "link down\n");
424 static const char *fc[] = { "no", "Rx", "Tx", "Tx/Rx" };
426 const char *s = "10Mbps";
427 const struct port_info *p = netdev_priv(dev);
429 switch (p->link_cfg.speed) {
444 netdev_info(dev, "link up, %s, full-duplex, %s PAUSE\n", s,
449 void t4_os_link_changed(struct adapter *adapter, int port_id, int link_stat)
451 struct net_device *dev = adapter->port[port_id];
453 /* Skip changes from disabled ports. */
454 if (netif_running(dev) && link_stat != netif_carrier_ok(dev)) {
456 netif_carrier_on(dev);
458 netif_carrier_off(dev);
464 void t4_os_portmod_changed(const struct adapter *adap, int port_id)
466 static const char *mod_str[] = {
467 NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
470 const struct net_device *dev = adap->port[port_id];
471 const struct port_info *pi = netdev_priv(dev);
473 if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
474 netdev_info(dev, "port module unplugged\n");
475 else if (pi->mod_type < ARRAY_SIZE(mod_str))
476 netdev_info(dev, "%s module inserted\n", mod_str[pi->mod_type]);
480 * Configure the exact and hash address filters to handle a port's multicast
481 * and secondary unicast MAC addresses.
483 static int set_addr_filters(const struct net_device *dev, bool sleep)
491 const struct netdev_hw_addr *ha;
492 int uc_cnt = netdev_uc_count(dev);
493 int mc_cnt = netdev_mc_count(dev);
494 const struct port_info *pi = netdev_priv(dev);
495 unsigned int mb = pi->adapter->fn;
497 /* first do the secondary unicast addresses */
498 netdev_for_each_uc_addr(ha, dev) {
499 addr[naddr++] = ha->addr;
500 if (--uc_cnt == 0 || naddr >= ARRAY_SIZE(addr)) {
501 ret = t4_alloc_mac_filt(pi->adapter, mb, pi->viid, free,
502 naddr, addr, filt_idx, &uhash, sleep);
511 /* next set up the multicast addresses */
512 netdev_for_each_mc_addr(ha, dev) {
513 addr[naddr++] = ha->addr;
514 if (--mc_cnt == 0 || naddr >= ARRAY_SIZE(addr)) {
515 ret = t4_alloc_mac_filt(pi->adapter, mb, pi->viid, free,
516 naddr, addr, filt_idx, &mhash, sleep);
525 return t4_set_addr_hash(pi->adapter, mb, pi->viid, uhash != 0,
526 uhash | mhash, sleep);
529 int dbfifo_int_thresh = 10; /* 10 == 640 entry threshold */
530 module_param(dbfifo_int_thresh, int, 0644);
531 MODULE_PARM_DESC(dbfifo_int_thresh, "doorbell fifo interrupt threshold");
534 * usecs to sleep while draining the dbfifo
536 static int dbfifo_drain_delay = 1000;
537 module_param(dbfifo_drain_delay, int, 0644);
538 MODULE_PARM_DESC(dbfifo_drain_delay,
539 "usecs to sleep while draining the dbfifo");
542 * Set Rx properties of a port, such as promiscruity, address filters, and MTU.
543 * If @mtu is -1 it is left unchanged.
545 static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
548 struct port_info *pi = netdev_priv(dev);
550 ret = set_addr_filters(dev, sleep_ok);
552 ret = t4_set_rxmode(pi->adapter, pi->adapter->fn, pi->viid, mtu,
553 (dev->flags & IFF_PROMISC) ? 1 : 0,
554 (dev->flags & IFF_ALLMULTI) ? 1 : 0, 1, -1,
559 static struct workqueue_struct *workq;
562 * link_start - enable a port
563 * @dev: the port to enable
565 * Performs the MAC and PHY actions needed to enable a port.
567 static int link_start(struct net_device *dev)
570 struct port_info *pi = netdev_priv(dev);
571 unsigned int mb = pi->adapter->fn;
574 * We do not set address filters and promiscuity here, the stack does
575 * that step explicitly.
577 ret = t4_set_rxmode(pi->adapter, mb, pi->viid, dev->mtu, -1, -1, -1,
578 !!(dev->features & NETIF_F_HW_VLAN_CTAG_RX), true);
580 ret = t4_change_mac(pi->adapter, mb, pi->viid,
581 pi->xact_addr_filt, dev->dev_addr, true,
584 pi->xact_addr_filt = ret;
589 ret = t4_link_start(pi->adapter, mb, pi->tx_chan,
592 ret = t4_enable_vi(pi->adapter, mb, pi->viid, true, true);
596 /* Clear a filter and release any of its resources that we own. This also
597 * clears the filter's "pending" status.
599 static void clear_filter(struct adapter *adap, struct filter_entry *f)
601 /* If the new or old filter have loopback rewriteing rules then we'll
602 * need to free any existing Layer Two Table (L2T) entries of the old
603 * filter rule. The firmware will handle freeing up any Source MAC
604 * Table (SMT) entries used for rewriting Source MAC Addresses in
608 cxgb4_l2t_release(f->l2t);
610 /* The zeroing of the filter rule below clears the filter valid,
611 * pending, locked flags, l2t pointer, etc. so it's all we need for
614 memset(f, 0, sizeof(*f));
617 /* Handle a filter write/deletion reply.
619 static void filter_rpl(struct adapter *adap, const struct cpl_set_tcb_rpl *rpl)
621 unsigned int idx = GET_TID(rpl);
622 unsigned int nidx = idx - adap->tids.ftid_base;
624 struct filter_entry *f;
626 if (idx >= adap->tids.ftid_base && nidx <
627 (adap->tids.nftids + adap->tids.nsftids)) {
629 ret = GET_TCB_COOKIE(rpl->cookie);
630 f = &adap->tids.ftid_tab[idx];
632 if (ret == FW_FILTER_WR_FLT_DELETED) {
633 /* Clear the filter when we get confirmation from the
634 * hardware that the filter has been deleted.
636 clear_filter(adap, f);
637 } else if (ret == FW_FILTER_WR_SMT_TBL_FULL) {
638 dev_err(adap->pdev_dev, "filter %u setup failed due to full SMT\n",
640 clear_filter(adap, f);
641 } else if (ret == FW_FILTER_WR_FLT_ADDED) {
642 f->smtidx = (be64_to_cpu(rpl->oldval) >> 24) & 0xff;
643 f->pending = 0; /* asynchronous setup completed */
646 /* Something went wrong. Issue a warning about the
647 * problem and clear everything out.
649 dev_err(adap->pdev_dev, "filter %u setup failed with error %u\n",
651 clear_filter(adap, f);
656 /* Response queue handler for the FW event queue.
658 static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
659 const struct pkt_gl *gl)
661 u8 opcode = ((const struct rss_header *)rsp)->opcode;
663 rsp++; /* skip RSS header */
665 /* FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG.
667 if (unlikely(opcode == CPL_FW4_MSG &&
668 ((const struct cpl_fw4_msg *)rsp)->type == FW_TYPE_RSSCPL)) {
670 opcode = ((const struct rss_header *)rsp)->opcode;
672 if (opcode != CPL_SGE_EGR_UPDATE) {
673 dev_err(q->adap->pdev_dev, "unexpected FW4/CPL %#x on FW event queue\n"
679 if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
680 const struct cpl_sge_egr_update *p = (void *)rsp;
681 unsigned int qid = EGR_QID(ntohl(p->opcode_qid));
684 txq = q->adap->sge.egr_map[qid - q->adap->sge.egr_start];
686 if ((u8 *)txq < (u8 *)q->adap->sge.ofldtxq) {
687 struct sge_eth_txq *eq;
689 eq = container_of(txq, struct sge_eth_txq, q);
690 netif_tx_wake_queue(eq->txq);
692 struct sge_ofld_txq *oq;
694 oq = container_of(txq, struct sge_ofld_txq, q);
695 tasklet_schedule(&oq->qresume_tsk);
697 } else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
698 const struct cpl_fw6_msg *p = (void *)rsp;
701 t4_handle_fw_rpl(q->adap, p->data);
702 } else if (opcode == CPL_L2T_WRITE_RPL) {
703 const struct cpl_l2t_write_rpl *p = (void *)rsp;
705 do_l2t_write_rpl(q->adap, p);
706 } else if (opcode == CPL_SET_TCB_RPL) {
707 const struct cpl_set_tcb_rpl *p = (void *)rsp;
709 filter_rpl(q->adap, p);
711 dev_err(q->adap->pdev_dev,
712 "unexpected CPL %#x on FW event queue\n", opcode);
718 * uldrx_handler - response queue handler for ULD queues
719 * @q: the response queue that received the packet
720 * @rsp: the response queue descriptor holding the offload message
721 * @gl: the gather list of packet fragments
723 * Deliver an ingress offload packet to a ULD. All processing is done by
724 * the ULD, we just maintain statistics.
726 static int uldrx_handler(struct sge_rspq *q, const __be64 *rsp,
727 const struct pkt_gl *gl)
729 struct sge_ofld_rxq *rxq = container_of(q, struct sge_ofld_rxq, rspq);
731 /* FW can send CPLs encapsulated in a CPL_FW4_MSG.
733 if (((const struct rss_header *)rsp)->opcode == CPL_FW4_MSG &&
734 ((const struct cpl_fw4_msg *)(rsp + 1))->type == FW_TYPE_RSSCPL)
737 if (ulds[q->uld].rx_handler(q->adap->uld_handle[q->uld], rsp, gl)) {
743 else if (gl == CXGB4_MSG_AN)
750 static void disable_msi(struct adapter *adapter)
752 if (adapter->flags & USING_MSIX) {
753 pci_disable_msix(adapter->pdev);
754 adapter->flags &= ~USING_MSIX;
755 } else if (adapter->flags & USING_MSI) {
756 pci_disable_msi(adapter->pdev);
757 adapter->flags &= ~USING_MSI;
762 * Interrupt handler for non-data events used with MSI-X.
764 static irqreturn_t t4_nondata_intr(int irq, void *cookie)
766 struct adapter *adap = cookie;
768 u32 v = t4_read_reg(adap, MYPF_REG(PL_PF_INT_CAUSE));
771 t4_write_reg(adap, MYPF_REG(PL_PF_INT_CAUSE), v);
773 t4_slow_intr_handler(adap);
778 * Name the MSI-X interrupts.
780 static void name_msix_vecs(struct adapter *adap)
782 int i, j, msi_idx = 2, n = sizeof(adap->msix_info[0].desc);
784 /* non-data interrupts */
785 snprintf(adap->msix_info[0].desc, n, "%s", adap->port[0]->name);
788 snprintf(adap->msix_info[1].desc, n, "%s-FWeventq",
789 adap->port[0]->name);
791 /* Ethernet queues */
792 for_each_port(adap, j) {
793 struct net_device *d = adap->port[j];
794 const struct port_info *pi = netdev_priv(d);
796 for (i = 0; i < pi->nqsets; i++, msi_idx++)
797 snprintf(adap->msix_info[msi_idx].desc, n, "%s-Rx%d",
802 for_each_ofldrxq(&adap->sge, i)
803 snprintf(adap->msix_info[msi_idx++].desc, n, "%s-ofld%d",
804 adap->port[0]->name, i);
806 for_each_rdmarxq(&adap->sge, i)
807 snprintf(adap->msix_info[msi_idx++].desc, n, "%s-rdma%d",
808 adap->port[0]->name, i);
811 static int request_msix_queue_irqs(struct adapter *adap)
813 struct sge *s = &adap->sge;
814 int err, ethqidx, ofldqidx = 0, rdmaqidx = 0, msi_index = 2;
816 err = request_irq(adap->msix_info[1].vec, t4_sge_intr_msix, 0,
817 adap->msix_info[1].desc, &s->fw_evtq);
821 for_each_ethrxq(s, ethqidx) {
822 err = request_irq(adap->msix_info[msi_index].vec,
824 adap->msix_info[msi_index].desc,
825 &s->ethrxq[ethqidx].rspq);
830 for_each_ofldrxq(s, ofldqidx) {
831 err = request_irq(adap->msix_info[msi_index].vec,
833 adap->msix_info[msi_index].desc,
834 &s->ofldrxq[ofldqidx].rspq);
839 for_each_rdmarxq(s, rdmaqidx) {
840 err = request_irq(adap->msix_info[msi_index].vec,
842 adap->msix_info[msi_index].desc,
843 &s->rdmarxq[rdmaqidx].rspq);
851 while (--rdmaqidx >= 0)
852 free_irq(adap->msix_info[--msi_index].vec,
853 &s->rdmarxq[rdmaqidx].rspq);
854 while (--ofldqidx >= 0)
855 free_irq(adap->msix_info[--msi_index].vec,
856 &s->ofldrxq[ofldqidx].rspq);
857 while (--ethqidx >= 0)
858 free_irq(adap->msix_info[--msi_index].vec,
859 &s->ethrxq[ethqidx].rspq);
860 free_irq(adap->msix_info[1].vec, &s->fw_evtq);
864 static void free_msix_queue_irqs(struct adapter *adap)
866 int i, msi_index = 2;
867 struct sge *s = &adap->sge;
869 free_irq(adap->msix_info[1].vec, &s->fw_evtq);
870 for_each_ethrxq(s, i)
871 free_irq(adap->msix_info[msi_index++].vec, &s->ethrxq[i].rspq);
872 for_each_ofldrxq(s, i)
873 free_irq(adap->msix_info[msi_index++].vec, &s->ofldrxq[i].rspq);
874 for_each_rdmarxq(s, i)
875 free_irq(adap->msix_info[msi_index++].vec, &s->rdmarxq[i].rspq);
879 * write_rss - write the RSS table for a given port
881 * @queues: array of queue indices for RSS
883 * Sets up the portion of the HW RSS table for the port's VI to distribute
884 * packets to the Rx queues in @queues.
886 static int write_rss(const struct port_info *pi, const u16 *queues)
890 const struct sge_eth_rxq *q = &pi->adapter->sge.ethrxq[pi->first_qset];
892 rss = kmalloc(pi->rss_size * sizeof(u16), GFP_KERNEL);
896 /* map the queue indices to queue ids */
897 for (i = 0; i < pi->rss_size; i++, queues++)
898 rss[i] = q[*queues].rspq.abs_id;
900 err = t4_config_rss_range(pi->adapter, pi->adapter->fn, pi->viid, 0,
901 pi->rss_size, rss, pi->rss_size);
907 * setup_rss - configure RSS
910 * Sets up RSS for each port.
912 static int setup_rss(struct adapter *adap)
916 for_each_port(adap, i) {
917 const struct port_info *pi = adap2pinfo(adap, i);
919 err = write_rss(pi, pi->rss);
927 * Return the channel of the ingress queue with the given qid.
929 static unsigned int rxq_to_chan(const struct sge *p, unsigned int qid)
931 qid -= p->ingr_start;
932 return netdev2pinfo(p->ingr_map[qid]->netdev)->tx_chan;
936 * Wait until all NAPI handlers are descheduled.
938 static void quiesce_rx(struct adapter *adap)
942 for (i = 0; i < ARRAY_SIZE(adap->sge.ingr_map); i++) {
943 struct sge_rspq *q = adap->sge.ingr_map[i];
946 napi_disable(&q->napi);
951 * Enable NAPI scheduling and interrupt generation for all Rx queues.
953 static void enable_rx(struct adapter *adap)
957 for (i = 0; i < ARRAY_SIZE(adap->sge.ingr_map); i++) {
958 struct sge_rspq *q = adap->sge.ingr_map[i];
963 napi_enable(&q->napi);
964 /* 0-increment GTS to start the timer and enable interrupts */
965 t4_write_reg(adap, MYPF_REG(SGE_PF_GTS),
966 SEINTARM(q->intr_params) |
967 INGRESSQID(q->cntxt_id));
972 * setup_sge_queues - configure SGE Tx/Rx/response queues
975 * Determines how many sets of SGE queues to use and initializes them.
976 * We support multiple queue sets per port if we have MSI-X, otherwise
977 * just one queue set per port.
979 static int setup_sge_queues(struct adapter *adap)
981 int err, msi_idx, i, j;
982 struct sge *s = &adap->sge;
984 bitmap_zero(s->starving_fl, MAX_EGRQ);
985 bitmap_zero(s->txq_maperr, MAX_EGRQ);
987 if (adap->flags & USING_MSIX)
988 msi_idx = 1; /* vector 0 is for non-queue interrupts */
990 err = t4_sge_alloc_rxq(adap, &s->intrq, false, adap->port[0], 0,
994 msi_idx = -((int)s->intrq.abs_id + 1);
997 err = t4_sge_alloc_rxq(adap, &s->fw_evtq, true, adap->port[0],
998 msi_idx, NULL, fwevtq_handler);
1000 freeout: t4_free_sge_resources(adap);
1004 for_each_port(adap, i) {
1005 struct net_device *dev = adap->port[i];
1006 struct port_info *pi = netdev_priv(dev);
1007 struct sge_eth_rxq *q = &s->ethrxq[pi->first_qset];
1008 struct sge_eth_txq *t = &s->ethtxq[pi->first_qset];
1010 for (j = 0; j < pi->nqsets; j++, q++) {
1013 err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev,
1019 memset(&q->stats, 0, sizeof(q->stats));
1021 for (j = 0; j < pi->nqsets; j++, t++) {
1022 err = t4_sge_alloc_eth_txq(adap, t, dev,
1023 netdev_get_tx_queue(dev, j),
1024 s->fw_evtq.cntxt_id);
1030 j = s->ofldqsets / adap->params.nports; /* ofld queues per channel */
1031 for_each_ofldrxq(s, i) {
1032 struct sge_ofld_rxq *q = &s->ofldrxq[i];
1033 struct net_device *dev = adap->port[i / j];
1037 err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev, msi_idx,
1038 &q->fl, uldrx_handler);
1041 memset(&q->stats, 0, sizeof(q->stats));
1042 s->ofld_rxq[i] = q->rspq.abs_id;
1043 err = t4_sge_alloc_ofld_txq(adap, &s->ofldtxq[i], dev,
1044 s->fw_evtq.cntxt_id);
1049 for_each_rdmarxq(s, i) {
1050 struct sge_ofld_rxq *q = &s->rdmarxq[i];
1054 err = t4_sge_alloc_rxq(adap, &q->rspq, false, adap->port[i],
1055 msi_idx, &q->fl, uldrx_handler);
1058 memset(&q->stats, 0, sizeof(q->stats));
1059 s->rdma_rxq[i] = q->rspq.abs_id;
1062 for_each_port(adap, i) {
1064 * Note that ->rdmarxq[i].rspq.cntxt_id below is 0 if we don't
1065 * have RDMA queues, and that's the right value.
1067 err = t4_sge_alloc_ctrl_txq(adap, &s->ctrlq[i], adap->port[i],
1068 s->fw_evtq.cntxt_id,
1069 s->rdmarxq[i].rspq.cntxt_id);
1074 t4_write_reg(adap, MPS_TRC_RSS_CONTROL,
1075 RSSCONTROL(netdev2pinfo(adap->port[0])->tx_chan) |
1076 QUEUENUMBER(s->ethrxq[0].rspq.abs_id));
1081 * Allocate a chunk of memory using kmalloc or, if that fails, vmalloc.
1082 * The allocated memory is cleared.
1084 void *t4_alloc_mem(size_t size)
1086 void *p = kzalloc(size, GFP_KERNEL | __GFP_NOWARN);
1094 * Free memory allocated through alloc_mem().
1096 static void t4_free_mem(void *addr)
1098 if (is_vmalloc_addr(addr))
1104 /* Send a Work Request to write the filter at a specified index. We construct
1105 * a Firmware Filter Work Request to have the work done and put the indicated
1106 * filter into "pending" mode which will prevent any further actions against
1107 * it till we get a reply from the firmware on the completion status of the
1110 static int set_filter_wr(struct adapter *adapter, int fidx)
1112 struct filter_entry *f = &adapter->tids.ftid_tab[fidx];
1113 struct sk_buff *skb;
1114 struct fw_filter_wr *fwr;
1117 /* If the new filter requires loopback Destination MAC and/or VLAN
1118 * rewriting then we need to allocate a Layer 2 Table (L2T) entry for
1121 if (f->fs.newdmac || f->fs.newvlan) {
1122 /* allocate L2T entry for new filter */
1123 f->l2t = t4_l2t_alloc_switching(adapter->l2t);
1126 if (t4_l2t_set_switching(adapter, f->l2t, f->fs.vlan,
1127 f->fs.eport, f->fs.dmac)) {
1128 cxgb4_l2t_release(f->l2t);
1134 ftid = adapter->tids.ftid_base + fidx;
1136 skb = alloc_skb(sizeof(*fwr), GFP_KERNEL | __GFP_NOFAIL);
1137 fwr = (struct fw_filter_wr *)__skb_put(skb, sizeof(*fwr));
1138 memset(fwr, 0, sizeof(*fwr));
1140 /* It would be nice to put most of the following in t4_hw.c but most
1141 * of the work is translating the cxgbtool ch_filter_specification
1142 * into the Work Request and the definition of that structure is
1143 * currently in cxgbtool.h which isn't appropriate to pull into the
1144 * common code. We may eventually try to come up with a more neutral
1145 * filter specification structure but for now it's easiest to simply
1146 * put this fairly direct code in line ...
1148 fwr->op_pkd = htonl(FW_WR_OP(FW_FILTER_WR));
1149 fwr->len16_pkd = htonl(FW_WR_LEN16(sizeof(*fwr)/16));
1151 htonl(V_FW_FILTER_WR_TID(ftid) |
1152 V_FW_FILTER_WR_RQTYPE(f->fs.type) |
1153 V_FW_FILTER_WR_NOREPLY(0) |
1154 V_FW_FILTER_WR_IQ(f->fs.iq));
1155 fwr->del_filter_to_l2tix =
1156 htonl(V_FW_FILTER_WR_RPTTID(f->fs.rpttid) |
1157 V_FW_FILTER_WR_DROP(f->fs.action == FILTER_DROP) |
1158 V_FW_FILTER_WR_DIRSTEER(f->fs.dirsteer) |
1159 V_FW_FILTER_WR_MASKHASH(f->fs.maskhash) |
1160 V_FW_FILTER_WR_DIRSTEERHASH(f->fs.dirsteerhash) |
1161 V_FW_FILTER_WR_LPBK(f->fs.action == FILTER_SWITCH) |
1162 V_FW_FILTER_WR_DMAC(f->fs.newdmac) |
1163 V_FW_FILTER_WR_SMAC(f->fs.newsmac) |
1164 V_FW_FILTER_WR_INSVLAN(f->fs.newvlan == VLAN_INSERT ||
1165 f->fs.newvlan == VLAN_REWRITE) |
1166 V_FW_FILTER_WR_RMVLAN(f->fs.newvlan == VLAN_REMOVE ||
1167 f->fs.newvlan == VLAN_REWRITE) |
1168 V_FW_FILTER_WR_HITCNTS(f->fs.hitcnts) |
1169 V_FW_FILTER_WR_TXCHAN(f->fs.eport) |
1170 V_FW_FILTER_WR_PRIO(f->fs.prio) |
1171 V_FW_FILTER_WR_L2TIX(f->l2t ? f->l2t->idx : 0));
1172 fwr->ethtype = htons(f->fs.val.ethtype);
1173 fwr->ethtypem = htons(f->fs.mask.ethtype);
1174 fwr->frag_to_ovlan_vldm =
1175 (V_FW_FILTER_WR_FRAG(f->fs.val.frag) |
1176 V_FW_FILTER_WR_FRAGM(f->fs.mask.frag) |
1177 V_FW_FILTER_WR_IVLAN_VLD(f->fs.val.ivlan_vld) |
1178 V_FW_FILTER_WR_OVLAN_VLD(f->fs.val.ovlan_vld) |
1179 V_FW_FILTER_WR_IVLAN_VLDM(f->fs.mask.ivlan_vld) |
1180 V_FW_FILTER_WR_OVLAN_VLDM(f->fs.mask.ovlan_vld));
1182 fwr->rx_chan_rx_rpl_iq =
1183 htons(V_FW_FILTER_WR_RX_CHAN(0) |
1184 V_FW_FILTER_WR_RX_RPL_IQ(adapter->sge.fw_evtq.abs_id));
1185 fwr->maci_to_matchtypem =
1186 htonl(V_FW_FILTER_WR_MACI(f->fs.val.macidx) |
1187 V_FW_FILTER_WR_MACIM(f->fs.mask.macidx) |
1188 V_FW_FILTER_WR_FCOE(f->fs.val.fcoe) |
1189 V_FW_FILTER_WR_FCOEM(f->fs.mask.fcoe) |
1190 V_FW_FILTER_WR_PORT(f->fs.val.iport) |
1191 V_FW_FILTER_WR_PORTM(f->fs.mask.iport) |
1192 V_FW_FILTER_WR_MATCHTYPE(f->fs.val.matchtype) |
1193 V_FW_FILTER_WR_MATCHTYPEM(f->fs.mask.matchtype));
1194 fwr->ptcl = f->fs.val.proto;
1195 fwr->ptclm = f->fs.mask.proto;
1196 fwr->ttyp = f->fs.val.tos;
1197 fwr->ttypm = f->fs.mask.tos;
1198 fwr->ivlan = htons(f->fs.val.ivlan);
1199 fwr->ivlanm = htons(f->fs.mask.ivlan);
1200 fwr->ovlan = htons(f->fs.val.ovlan);
1201 fwr->ovlanm = htons(f->fs.mask.ovlan);
1202 memcpy(fwr->lip, f->fs.val.lip, sizeof(fwr->lip));
1203 memcpy(fwr->lipm, f->fs.mask.lip, sizeof(fwr->lipm));
1204 memcpy(fwr->fip, f->fs.val.fip, sizeof(fwr->fip));
1205 memcpy(fwr->fipm, f->fs.mask.fip, sizeof(fwr->fipm));
1206 fwr->lp = htons(f->fs.val.lport);
1207 fwr->lpm = htons(f->fs.mask.lport);
1208 fwr->fp = htons(f->fs.val.fport);
1209 fwr->fpm = htons(f->fs.mask.fport);
1211 memcpy(fwr->sma, f->fs.smac, sizeof(fwr->sma));
1213 /* Mark the filter as "pending" and ship off the Filter Work Request.
1214 * When we get the Work Request Reply we'll clear the pending status.
1217 set_wr_txq(skb, CPL_PRIORITY_CONTROL, f->fs.val.iport & 0x3);
1218 t4_ofld_send(adapter, skb);
1222 /* Delete the filter at a specified index.
1224 static int del_filter_wr(struct adapter *adapter, int fidx)
1226 struct filter_entry *f = &adapter->tids.ftid_tab[fidx];
1227 struct sk_buff *skb;
1228 struct fw_filter_wr *fwr;
1229 unsigned int len, ftid;
1232 ftid = adapter->tids.ftid_base + fidx;
1234 skb = alloc_skb(len, GFP_KERNEL | __GFP_NOFAIL);
1235 fwr = (struct fw_filter_wr *)__skb_put(skb, len);
1236 t4_mk_filtdelwr(ftid, fwr, adapter->sge.fw_evtq.abs_id);
1238 /* Mark the filter as "pending" and ship off the Filter Work Request.
1239 * When we get the Work Request Reply we'll clear the pending status.
1242 t4_mgmt_tx(adapter, skb);
1246 static inline int is_offload(const struct adapter *adap)
1248 return adap->params.offload;
1252 * Implementation of ethtool operations.
1255 static u32 get_msglevel(struct net_device *dev)
1257 return netdev2adap(dev)->msg_enable;
1260 static void set_msglevel(struct net_device *dev, u32 val)
1262 netdev2adap(dev)->msg_enable = val;
1265 static char stats_strings[][ETH_GSTRING_LEN] = {
1268 "TxBroadcastFrames ",
1269 "TxMulticastFrames ",
1275 "TxFrames128To255 ",
1276 "TxFrames256To511 ",
1277 "TxFrames512To1023 ",
1278 "TxFrames1024To1518 ",
1279 "TxFrames1519ToMax ",
1294 "RxBroadcastFrames ",
1295 "RxMulticastFrames ",
1307 "RxFrames128To255 ",
1308 "RxFrames256To511 ",
1309 "RxFrames512To1023 ",
1310 "RxFrames1024To1518 ",
1311 "RxFrames1519ToMax ",
1323 "RxBG0FramesDropped ",
1324 "RxBG1FramesDropped ",
1325 "RxBG2FramesDropped ",
1326 "RxBG3FramesDropped ",
1327 "RxBG0FramesTrunc ",
1328 "RxBG1FramesTrunc ",
1329 "RxBG2FramesTrunc ",
1330 "RxBG3FramesTrunc ",
1339 "WriteCoalSuccess ",
1343 static int get_sset_count(struct net_device *dev, int sset)
1347 return ARRAY_SIZE(stats_strings);
1353 #define T4_REGMAP_SIZE (160 * 1024)
1354 #define T5_REGMAP_SIZE (332 * 1024)
1356 static int get_regs_len(struct net_device *dev)
1358 struct adapter *adap = netdev2adap(dev);
1359 if (is_t4(adap->params.chip))
1360 return T4_REGMAP_SIZE;
1362 return T5_REGMAP_SIZE;
1365 static int get_eeprom_len(struct net_device *dev)
1370 static void get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1372 struct adapter *adapter = netdev2adap(dev);
1374 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
1375 strlcpy(info->version, DRV_VERSION, sizeof(info->version));
1376 strlcpy(info->bus_info, pci_name(adapter->pdev),
1377 sizeof(info->bus_info));
1379 if (adapter->params.fw_vers)
1380 snprintf(info->fw_version, sizeof(info->fw_version),
1381 "%u.%u.%u.%u, TP %u.%u.%u.%u",
1382 FW_HDR_FW_VER_MAJOR_GET(adapter->params.fw_vers),
1383 FW_HDR_FW_VER_MINOR_GET(adapter->params.fw_vers),
1384 FW_HDR_FW_VER_MICRO_GET(adapter->params.fw_vers),
1385 FW_HDR_FW_VER_BUILD_GET(adapter->params.fw_vers),
1386 FW_HDR_FW_VER_MAJOR_GET(adapter->params.tp_vers),
1387 FW_HDR_FW_VER_MINOR_GET(adapter->params.tp_vers),
1388 FW_HDR_FW_VER_MICRO_GET(adapter->params.tp_vers),
1389 FW_HDR_FW_VER_BUILD_GET(adapter->params.tp_vers));
1392 static void get_strings(struct net_device *dev, u32 stringset, u8 *data)
1394 if (stringset == ETH_SS_STATS)
1395 memcpy(data, stats_strings, sizeof(stats_strings));
1399 * port stats maintained per queue of the port. They should be in the same
1400 * order as in stats_strings above.
1402 struct queue_port_stats {
1412 static void collect_sge_port_stats(const struct adapter *adap,
1413 const struct port_info *p, struct queue_port_stats *s)
1416 const struct sge_eth_txq *tx = &adap->sge.ethtxq[p->first_qset];
1417 const struct sge_eth_rxq *rx = &adap->sge.ethrxq[p->first_qset];
1419 memset(s, 0, sizeof(*s));
1420 for (i = 0; i < p->nqsets; i++, rx++, tx++) {
1422 s->tx_csum += tx->tx_cso;
1423 s->rx_csum += rx->stats.rx_cso;
1424 s->vlan_ex += rx->stats.vlan_ex;
1425 s->vlan_ins += tx->vlan_ins;
1426 s->gro_pkts += rx->stats.lro_pkts;
1427 s->gro_merged += rx->stats.lro_merged;
1431 static void get_stats(struct net_device *dev, struct ethtool_stats *stats,
1434 struct port_info *pi = netdev_priv(dev);
1435 struct adapter *adapter = pi->adapter;
1438 t4_get_port_stats(adapter, pi->tx_chan, (struct port_stats *)data);
1440 data += sizeof(struct port_stats) / sizeof(u64);
1441 collect_sge_port_stats(adapter, pi, (struct queue_port_stats *)data);
1442 data += sizeof(struct queue_port_stats) / sizeof(u64);
1443 if (!is_t4(adapter->params.chip)) {
1444 t4_write_reg(adapter, SGE_STAT_CFG, STATSOURCE_T5(7));
1445 val1 = t4_read_reg(adapter, SGE_STAT_TOTAL);
1446 val2 = t4_read_reg(adapter, SGE_STAT_MATCH);
1447 *data = val1 - val2;
1452 memset(data, 0, 2 * sizeof(u64));
1458 * Return a version number to identify the type of adapter. The scheme is:
1459 * - bits 0..9: chip version
1460 * - bits 10..15: chip revision
1461 * - bits 16..23: register dump version
1463 static inline unsigned int mk_adap_vers(const struct adapter *ap)
1465 return CHELSIO_CHIP_VERSION(ap->params.chip) |
1466 (CHELSIO_CHIP_RELEASE(ap->params.chip) << 10) | (1 << 16);
1469 static void reg_block_dump(struct adapter *ap, void *buf, unsigned int start,
1472 u32 *p = buf + start;
1474 for ( ; start <= end; start += sizeof(u32))
1475 *p++ = t4_read_reg(ap, start);
1478 static void get_regs(struct net_device *dev, struct ethtool_regs *regs,
1481 static const unsigned int t4_reg_ranges[] = {
1701 static const unsigned int t5_reg_ranges[] = {
2129 struct adapter *ap = netdev2adap(dev);
2130 static const unsigned int *reg_ranges;
2131 int arr_size = 0, buf_size = 0;
2133 if (is_t4(ap->params.chip)) {
2134 reg_ranges = &t4_reg_ranges[0];
2135 arr_size = ARRAY_SIZE(t4_reg_ranges);
2136 buf_size = T4_REGMAP_SIZE;
2138 reg_ranges = &t5_reg_ranges[0];
2139 arr_size = ARRAY_SIZE(t5_reg_ranges);
2140 buf_size = T5_REGMAP_SIZE;
2143 regs->version = mk_adap_vers(ap);
2145 memset(buf, 0, buf_size);
2146 for (i = 0; i < arr_size; i += 2)
2147 reg_block_dump(ap, buf, reg_ranges[i], reg_ranges[i + 1]);
2150 static int restart_autoneg(struct net_device *dev)
2152 struct port_info *p = netdev_priv(dev);
2154 if (!netif_running(dev))
2156 if (p->link_cfg.autoneg != AUTONEG_ENABLE)
2158 t4_restart_aneg(p->adapter, p->adapter->fn, p->tx_chan);
2162 static int identify_port(struct net_device *dev,
2163 enum ethtool_phys_id_state state)
2166 struct adapter *adap = netdev2adap(dev);
2168 if (state == ETHTOOL_ID_ACTIVE)
2170 else if (state == ETHTOOL_ID_INACTIVE)
2175 return t4_identify_port(adap, adap->fn, netdev2pinfo(dev)->viid, val);
2178 static unsigned int from_fw_linkcaps(unsigned int type, unsigned int caps)
2182 if (type == FW_PORT_TYPE_BT_SGMII || type == FW_PORT_TYPE_BT_XFI ||
2183 type == FW_PORT_TYPE_BT_XAUI) {
2185 if (caps & FW_PORT_CAP_SPEED_100M)
2186 v |= SUPPORTED_100baseT_Full;
2187 if (caps & FW_PORT_CAP_SPEED_1G)
2188 v |= SUPPORTED_1000baseT_Full;
2189 if (caps & FW_PORT_CAP_SPEED_10G)
2190 v |= SUPPORTED_10000baseT_Full;
2191 } else if (type == FW_PORT_TYPE_KX4 || type == FW_PORT_TYPE_KX) {
2192 v |= SUPPORTED_Backplane;
2193 if (caps & FW_PORT_CAP_SPEED_1G)
2194 v |= SUPPORTED_1000baseKX_Full;
2195 if (caps & FW_PORT_CAP_SPEED_10G)
2196 v |= SUPPORTED_10000baseKX4_Full;
2197 } else if (type == FW_PORT_TYPE_KR)
2198 v |= SUPPORTED_Backplane | SUPPORTED_10000baseKR_Full;
2199 else if (type == FW_PORT_TYPE_BP_AP)
2200 v |= SUPPORTED_Backplane | SUPPORTED_10000baseR_FEC |
2201 SUPPORTED_10000baseKR_Full | SUPPORTED_1000baseKX_Full;
2202 else if (type == FW_PORT_TYPE_BP4_AP)
2203 v |= SUPPORTED_Backplane | SUPPORTED_10000baseR_FEC |
2204 SUPPORTED_10000baseKR_Full | SUPPORTED_1000baseKX_Full |
2205 SUPPORTED_10000baseKX4_Full;
2206 else if (type == FW_PORT_TYPE_FIBER_XFI ||
2207 type == FW_PORT_TYPE_FIBER_XAUI || type == FW_PORT_TYPE_SFP)
2208 v |= SUPPORTED_FIBRE;
2209 else if (type == FW_PORT_TYPE_BP40_BA)
2210 v |= SUPPORTED_40000baseSR4_Full;
2212 if (caps & FW_PORT_CAP_ANEG)
2213 v |= SUPPORTED_Autoneg;
2217 static unsigned int to_fw_linkcaps(unsigned int caps)
2221 if (caps & ADVERTISED_100baseT_Full)
2222 v |= FW_PORT_CAP_SPEED_100M;
2223 if (caps & ADVERTISED_1000baseT_Full)
2224 v |= FW_PORT_CAP_SPEED_1G;
2225 if (caps & ADVERTISED_10000baseT_Full)
2226 v |= FW_PORT_CAP_SPEED_10G;
2227 if (caps & ADVERTISED_40000baseSR4_Full)
2228 v |= FW_PORT_CAP_SPEED_40G;
2232 static int get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2234 const struct port_info *p = netdev_priv(dev);
2236 if (p->port_type == FW_PORT_TYPE_BT_SGMII ||
2237 p->port_type == FW_PORT_TYPE_BT_XFI ||
2238 p->port_type == FW_PORT_TYPE_BT_XAUI)
2239 cmd->port = PORT_TP;
2240 else if (p->port_type == FW_PORT_TYPE_FIBER_XFI ||
2241 p->port_type == FW_PORT_TYPE_FIBER_XAUI)
2242 cmd->port = PORT_FIBRE;
2243 else if (p->port_type == FW_PORT_TYPE_SFP) {
2244 if (p->mod_type == FW_PORT_MOD_TYPE_TWINAX_PASSIVE ||
2245 p->mod_type == FW_PORT_MOD_TYPE_TWINAX_ACTIVE)
2246 cmd->port = PORT_DA;
2248 cmd->port = PORT_FIBRE;
2250 cmd->port = PORT_OTHER;
2252 if (p->mdio_addr >= 0) {
2253 cmd->phy_address = p->mdio_addr;
2254 cmd->transceiver = XCVR_EXTERNAL;
2255 cmd->mdio_support = p->port_type == FW_PORT_TYPE_BT_SGMII ?
2256 MDIO_SUPPORTS_C22 : MDIO_SUPPORTS_C45;
2258 cmd->phy_address = 0; /* not really, but no better option */
2259 cmd->transceiver = XCVR_INTERNAL;
2260 cmd->mdio_support = 0;
2263 cmd->supported = from_fw_linkcaps(p->port_type, p->link_cfg.supported);
2264 cmd->advertising = from_fw_linkcaps(p->port_type,
2265 p->link_cfg.advertising);
2266 ethtool_cmd_speed_set(cmd,
2267 netif_carrier_ok(dev) ? p->link_cfg.speed : 0);
2268 cmd->duplex = DUPLEX_FULL;
2269 cmd->autoneg = p->link_cfg.autoneg;
2275 static unsigned int speed_to_caps(int speed)
2278 return FW_PORT_CAP_SPEED_100M;
2280 return FW_PORT_CAP_SPEED_1G;
2282 return FW_PORT_CAP_SPEED_10G;
2284 return FW_PORT_CAP_SPEED_40G;
2288 static int set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
2291 struct port_info *p = netdev_priv(dev);
2292 struct link_config *lc = &p->link_cfg;
2293 u32 speed = ethtool_cmd_speed(cmd);
2295 if (cmd->duplex != DUPLEX_FULL) /* only full-duplex supported */
2298 if (!(lc->supported & FW_PORT_CAP_ANEG)) {
2300 * PHY offers a single speed. See if that's what's
2303 if (cmd->autoneg == AUTONEG_DISABLE &&
2304 (lc->supported & speed_to_caps(speed)))
2309 if (cmd->autoneg == AUTONEG_DISABLE) {
2310 cap = speed_to_caps(speed);
2312 if (!(lc->supported & cap) ||
2317 lc->requested_speed = cap;
2318 lc->advertising = 0;
2320 cap = to_fw_linkcaps(cmd->advertising);
2321 if (!(lc->supported & cap))
2323 lc->requested_speed = 0;
2324 lc->advertising = cap | FW_PORT_CAP_ANEG;
2326 lc->autoneg = cmd->autoneg;
2328 if (netif_running(dev))
2329 return t4_link_start(p->adapter, p->adapter->fn, p->tx_chan,
2334 static void get_pauseparam(struct net_device *dev,
2335 struct ethtool_pauseparam *epause)
2337 struct port_info *p = netdev_priv(dev);
2339 epause->autoneg = (p->link_cfg.requested_fc & PAUSE_AUTONEG) != 0;
2340 epause->rx_pause = (p->link_cfg.fc & PAUSE_RX) != 0;
2341 epause->tx_pause = (p->link_cfg.fc & PAUSE_TX) != 0;
2344 static int set_pauseparam(struct net_device *dev,
2345 struct ethtool_pauseparam *epause)
2347 struct port_info *p = netdev_priv(dev);
2348 struct link_config *lc = &p->link_cfg;
2350 if (epause->autoneg == AUTONEG_DISABLE)
2351 lc->requested_fc = 0;
2352 else if (lc->supported & FW_PORT_CAP_ANEG)
2353 lc->requested_fc = PAUSE_AUTONEG;
2357 if (epause->rx_pause)
2358 lc->requested_fc |= PAUSE_RX;
2359 if (epause->tx_pause)
2360 lc->requested_fc |= PAUSE_TX;
2361 if (netif_running(dev))
2362 return t4_link_start(p->adapter, p->adapter->fn, p->tx_chan,
2367 static void get_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
2369 const struct port_info *pi = netdev_priv(dev);
2370 const struct sge *s = &pi->adapter->sge;
2372 e->rx_max_pending = MAX_RX_BUFFERS;
2373 e->rx_mini_max_pending = MAX_RSPQ_ENTRIES;
2374 e->rx_jumbo_max_pending = 0;
2375 e->tx_max_pending = MAX_TXQ_ENTRIES;
2377 e->rx_pending = s->ethrxq[pi->first_qset].fl.size - 8;
2378 e->rx_mini_pending = s->ethrxq[pi->first_qset].rspq.size;
2379 e->rx_jumbo_pending = 0;
2380 e->tx_pending = s->ethtxq[pi->first_qset].q.size;
2383 static int set_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
2386 const struct port_info *pi = netdev_priv(dev);
2387 struct adapter *adapter = pi->adapter;
2388 struct sge *s = &adapter->sge;
2390 if (e->rx_pending > MAX_RX_BUFFERS || e->rx_jumbo_pending ||
2391 e->tx_pending > MAX_TXQ_ENTRIES ||
2392 e->rx_mini_pending > MAX_RSPQ_ENTRIES ||
2393 e->rx_mini_pending < MIN_RSPQ_ENTRIES ||
2394 e->rx_pending < MIN_FL_ENTRIES || e->tx_pending < MIN_TXQ_ENTRIES)
2397 if (adapter->flags & FULL_INIT_DONE)
2400 for (i = 0; i < pi->nqsets; ++i) {
2401 s->ethtxq[pi->first_qset + i].q.size = e->tx_pending;
2402 s->ethrxq[pi->first_qset + i].fl.size = e->rx_pending + 8;
2403 s->ethrxq[pi->first_qset + i].rspq.size = e->rx_mini_pending;
2408 static int closest_timer(const struct sge *s, int time)
2410 int i, delta, match = 0, min_delta = INT_MAX;
2412 for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
2413 delta = time - s->timer_val[i];
2416 if (delta < min_delta) {
2424 static int closest_thres(const struct sge *s, int thres)
2426 int i, delta, match = 0, min_delta = INT_MAX;
2428 for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
2429 delta = thres - s->counter_val[i];
2432 if (delta < min_delta) {
2441 * Return a queue's interrupt hold-off time in us. 0 means no timer.
2443 static unsigned int qtimer_val(const struct adapter *adap,
2444 const struct sge_rspq *q)
2446 unsigned int idx = q->intr_params >> 1;
2448 return idx < SGE_NTIMERS ? adap->sge.timer_val[idx] : 0;
2452 * set_rxq_intr_params - set a queue's interrupt holdoff parameters
2453 * @adap: the adapter
2455 * @us: the hold-off time in us, or 0 to disable timer
2456 * @cnt: the hold-off packet count, or 0 to disable counter
2458 * Sets an Rx queue's interrupt hold-off time and packet count. At least
2459 * one of the two needs to be enabled for the queue to generate interrupts.
2461 static int set_rxq_intr_params(struct adapter *adap, struct sge_rspq *q,
2462 unsigned int us, unsigned int cnt)
2464 if ((us | cnt) == 0)
2471 new_idx = closest_thres(&adap->sge, cnt);
2472 if (q->desc && q->pktcnt_idx != new_idx) {
2473 /* the queue has already been created, update it */
2474 v = FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
2475 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
2476 FW_PARAMS_PARAM_YZ(q->cntxt_id);
2477 err = t4_set_params(adap, adap->fn, adap->fn, 0, 1, &v,
2482 q->pktcnt_idx = new_idx;
2485 us = us == 0 ? 6 : closest_timer(&adap->sge, us);
2486 q->intr_params = QINTR_TIMER_IDX(us) | (cnt > 0 ? QINTR_CNT_EN : 0);
2490 static int set_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
2492 const struct port_info *pi = netdev_priv(dev);
2493 struct adapter *adap = pi->adapter;
2498 for (i = pi->first_qset; i < pi->first_qset + pi->nqsets; i++) {
2499 q = &adap->sge.ethrxq[i].rspq;
2500 r = set_rxq_intr_params(adap, q, c->rx_coalesce_usecs,
2501 c->rx_max_coalesced_frames);
2503 dev_err(&dev->dev, "failed to set coalesce %d\n", r);
2510 static int get_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
2512 const struct port_info *pi = netdev_priv(dev);
2513 const struct adapter *adap = pi->adapter;
2514 const struct sge_rspq *rq = &adap->sge.ethrxq[pi->first_qset].rspq;
2516 c->rx_coalesce_usecs = qtimer_val(adap, rq);
2517 c->rx_max_coalesced_frames = (rq->intr_params & QINTR_CNT_EN) ?
2518 adap->sge.counter_val[rq->pktcnt_idx] : 0;
2523 * eeprom_ptov - translate a physical EEPROM address to virtual
2524 * @phys_addr: the physical EEPROM address
2525 * @fn: the PCI function number
2526 * @sz: size of function-specific area
2528 * Translate a physical EEPROM address to virtual. The first 1K is
2529 * accessed through virtual addresses starting at 31K, the rest is
2530 * accessed through virtual addresses starting at 0.
2532 * The mapping is as follows:
2533 * [0..1K) -> [31K..32K)
2534 * [1K..1K+A) -> [31K-A..31K)
2535 * [1K+A..ES) -> [0..ES-A-1K)
2537 * where A = @fn * @sz, and ES = EEPROM size.
2539 static int eeprom_ptov(unsigned int phys_addr, unsigned int fn, unsigned int sz)
2542 if (phys_addr < 1024)
2543 return phys_addr + (31 << 10);
2544 if (phys_addr < 1024 + fn)
2545 return 31744 - fn + phys_addr - 1024;
2546 if (phys_addr < EEPROMSIZE)
2547 return phys_addr - 1024 - fn;
2552 * The next two routines implement eeprom read/write from physical addresses.
2554 static int eeprom_rd_phys(struct adapter *adap, unsigned int phys_addr, u32 *v)
2556 int vaddr = eeprom_ptov(phys_addr, adap->fn, EEPROMPFSIZE);
2559 vaddr = pci_read_vpd(adap->pdev, vaddr, sizeof(u32), v);
2560 return vaddr < 0 ? vaddr : 0;
2563 static int eeprom_wr_phys(struct adapter *adap, unsigned int phys_addr, u32 v)
2565 int vaddr = eeprom_ptov(phys_addr, adap->fn, EEPROMPFSIZE);
2568 vaddr = pci_write_vpd(adap->pdev, vaddr, sizeof(u32), &v);
2569 return vaddr < 0 ? vaddr : 0;
2572 #define EEPROM_MAGIC 0x38E2F10C
2574 static int get_eeprom(struct net_device *dev, struct ethtool_eeprom *e,
2578 struct adapter *adapter = netdev2adap(dev);
2580 u8 *buf = kmalloc(EEPROMSIZE, GFP_KERNEL);
2584 e->magic = EEPROM_MAGIC;
2585 for (i = e->offset & ~3; !err && i < e->offset + e->len; i += 4)
2586 err = eeprom_rd_phys(adapter, i, (u32 *)&buf[i]);
2589 memcpy(data, buf + e->offset, e->len);
2594 static int set_eeprom(struct net_device *dev, struct ethtool_eeprom *eeprom,
2599 u32 aligned_offset, aligned_len, *p;
2600 struct adapter *adapter = netdev2adap(dev);
2602 if (eeprom->magic != EEPROM_MAGIC)
2605 aligned_offset = eeprom->offset & ~3;
2606 aligned_len = (eeprom->len + (eeprom->offset & 3) + 3) & ~3;
2608 if (adapter->fn > 0) {
2609 u32 start = 1024 + adapter->fn * EEPROMPFSIZE;
2611 if (aligned_offset < start ||
2612 aligned_offset + aligned_len > start + EEPROMPFSIZE)
2616 if (aligned_offset != eeprom->offset || aligned_len != eeprom->len) {
2618 * RMW possibly needed for first or last words.
2620 buf = kmalloc(aligned_len, GFP_KERNEL);
2623 err = eeprom_rd_phys(adapter, aligned_offset, (u32 *)buf);
2624 if (!err && aligned_len > 4)
2625 err = eeprom_rd_phys(adapter,
2626 aligned_offset + aligned_len - 4,
2627 (u32 *)&buf[aligned_len - 4]);
2630 memcpy(buf + (eeprom->offset & 3), data, eeprom->len);
2634 err = t4_seeprom_wp(adapter, false);
2638 for (p = (u32 *)buf; !err && aligned_len; aligned_len -= 4, p++) {
2639 err = eeprom_wr_phys(adapter, aligned_offset, *p);
2640 aligned_offset += 4;
2644 err = t4_seeprom_wp(adapter, true);
2651 static int set_flash(struct net_device *netdev, struct ethtool_flash *ef)
2654 const struct firmware *fw;
2655 struct adapter *adap = netdev2adap(netdev);
2657 ef->data[sizeof(ef->data) - 1] = '\0';
2658 ret = request_firmware(&fw, ef->data, adap->pdev_dev);
2662 ret = t4_load_fw(adap, fw->data, fw->size);
2663 release_firmware(fw);
2665 dev_info(adap->pdev_dev, "loaded firmware %s\n", ef->data);
2669 #define WOL_SUPPORTED (WAKE_BCAST | WAKE_MAGIC)
2670 #define BCAST_CRC 0xa0ccc1a6
2672 static void get_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
2674 wol->supported = WAKE_BCAST | WAKE_MAGIC;
2675 wol->wolopts = netdev2adap(dev)->wol;
2676 memset(&wol->sopass, 0, sizeof(wol->sopass));
2679 static int set_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
2682 struct port_info *pi = netdev_priv(dev);
2684 if (wol->wolopts & ~WOL_SUPPORTED)
2686 t4_wol_magic_enable(pi->adapter, pi->tx_chan,
2687 (wol->wolopts & WAKE_MAGIC) ? dev->dev_addr : NULL);
2688 if (wol->wolopts & WAKE_BCAST) {
2689 err = t4_wol_pat_enable(pi->adapter, pi->tx_chan, 0xfe, ~0ULL,
2692 err = t4_wol_pat_enable(pi->adapter, pi->tx_chan, 1,
2693 ~6ULL, ~0ULL, BCAST_CRC, true);
2695 t4_wol_pat_enable(pi->adapter, pi->tx_chan, 0, 0, 0, 0, false);
2699 static int cxgb_set_features(struct net_device *dev, netdev_features_t features)
2701 const struct port_info *pi = netdev_priv(dev);
2702 netdev_features_t changed = dev->features ^ features;
2705 if (!(changed & NETIF_F_HW_VLAN_CTAG_RX))
2708 err = t4_set_rxmode(pi->adapter, pi->adapter->fn, pi->viid, -1,
2710 !!(features & NETIF_F_HW_VLAN_CTAG_RX), true);
2712 dev->features = features ^ NETIF_F_HW_VLAN_CTAG_RX;
2716 static u32 get_rss_table_size(struct net_device *dev)
2718 const struct port_info *pi = netdev_priv(dev);
2720 return pi->rss_size;
2723 static int get_rss_table(struct net_device *dev, u32 *p)
2725 const struct port_info *pi = netdev_priv(dev);
2726 unsigned int n = pi->rss_size;
2733 static int set_rss_table(struct net_device *dev, const u32 *p)
2736 struct port_info *pi = netdev_priv(dev);
2738 for (i = 0; i < pi->rss_size; i++)
2740 if (pi->adapter->flags & FULL_INIT_DONE)
2741 return write_rss(pi, pi->rss);
2745 static int get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
2748 const struct port_info *pi = netdev_priv(dev);
2750 switch (info->cmd) {
2751 case ETHTOOL_GRXFH: {
2752 unsigned int v = pi->rss_mode;
2755 switch (info->flow_type) {
2757 if (v & FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
2758 info->data = RXH_IP_SRC | RXH_IP_DST |
2759 RXH_L4_B_0_1 | RXH_L4_B_2_3;
2760 else if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
2761 info->data = RXH_IP_SRC | RXH_IP_DST;
2764 if ((v & FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN) &&
2765 (v & FW_RSS_VI_CONFIG_CMD_UDPEN))
2766 info->data = RXH_IP_SRC | RXH_IP_DST |
2767 RXH_L4_B_0_1 | RXH_L4_B_2_3;
2768 else if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
2769 info->data = RXH_IP_SRC | RXH_IP_DST;
2772 case AH_ESP_V4_FLOW:
2774 if (v & FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
2775 info->data = RXH_IP_SRC | RXH_IP_DST;
2778 if (v & FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
2779 info->data = RXH_IP_SRC | RXH_IP_DST |
2780 RXH_L4_B_0_1 | RXH_L4_B_2_3;
2781 else if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
2782 info->data = RXH_IP_SRC | RXH_IP_DST;
2785 if ((v & FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN) &&
2786 (v & FW_RSS_VI_CONFIG_CMD_UDPEN))
2787 info->data = RXH_IP_SRC | RXH_IP_DST |
2788 RXH_L4_B_0_1 | RXH_L4_B_2_3;
2789 else if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
2790 info->data = RXH_IP_SRC | RXH_IP_DST;
2793 case AH_ESP_V6_FLOW:
2795 if (v & FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
2796 info->data = RXH_IP_SRC | RXH_IP_DST;
2801 case ETHTOOL_GRXRINGS:
2802 info->data = pi->nqsets;
2808 static const struct ethtool_ops cxgb_ethtool_ops = {
2809 .get_settings = get_settings,
2810 .set_settings = set_settings,
2811 .get_drvinfo = get_drvinfo,
2812 .get_msglevel = get_msglevel,
2813 .set_msglevel = set_msglevel,
2814 .get_ringparam = get_sge_param,
2815 .set_ringparam = set_sge_param,
2816 .get_coalesce = get_coalesce,
2817 .set_coalesce = set_coalesce,
2818 .get_eeprom_len = get_eeprom_len,
2819 .get_eeprom = get_eeprom,
2820 .set_eeprom = set_eeprom,
2821 .get_pauseparam = get_pauseparam,
2822 .set_pauseparam = set_pauseparam,
2823 .get_link = ethtool_op_get_link,
2824 .get_strings = get_strings,
2825 .set_phys_id = identify_port,
2826 .nway_reset = restart_autoneg,
2827 .get_sset_count = get_sset_count,
2828 .get_ethtool_stats = get_stats,
2829 .get_regs_len = get_regs_len,
2830 .get_regs = get_regs,
2833 .get_rxnfc = get_rxnfc,
2834 .get_rxfh_indir_size = get_rss_table_size,
2835 .get_rxfh_indir = get_rss_table,
2836 .set_rxfh_indir = set_rss_table,
2837 .flash_device = set_flash,
2843 static ssize_t mem_read(struct file *file, char __user *buf, size_t count,
2847 loff_t avail = file_inode(file)->i_size;
2848 unsigned int mem = (uintptr_t)file->private_data & 3;
2849 struct adapter *adap = file->private_data - mem;
2855 if (count > avail - pos)
2856 count = avail - pos;
2863 if ((mem == MEM_MC) || (mem == MEM_MC1))
2864 ret = t4_mc_read(adap, mem % MEM_MC, pos, data, NULL);
2866 ret = t4_edc_read(adap, mem, pos, data, NULL);
2870 ofst = pos % sizeof(data);
2871 len = min(count, sizeof(data) - ofst);
2872 if (copy_to_user(buf, (u8 *)data + ofst, len))
2879 count = pos - *ppos;
2884 static const struct file_operations mem_debugfs_fops = {
2885 .owner = THIS_MODULE,
2886 .open = simple_open,
2888 .llseek = default_llseek,
2891 static void add_debugfs_mem(struct adapter *adap, const char *name,
2892 unsigned int idx, unsigned int size_mb)
2896 de = debugfs_create_file(name, S_IRUSR, adap->debugfs_root,
2897 (void *)adap + idx, &mem_debugfs_fops);
2898 if (de && de->d_inode)
2899 de->d_inode->i_size = size_mb << 20;
2902 static int setup_debugfs(struct adapter *adap)
2907 if (IS_ERR_OR_NULL(adap->debugfs_root))
2910 i = t4_read_reg(adap, MA_TARGET_MEM_ENABLE);
2911 if (i & EDRAM0_ENABLE) {
2912 size = t4_read_reg(adap, MA_EDRAM0_BAR);
2913 add_debugfs_mem(adap, "edc0", MEM_EDC0, EDRAM_SIZE_GET(size));
2915 if (i & EDRAM1_ENABLE) {
2916 size = t4_read_reg(adap, MA_EDRAM1_BAR);
2917 add_debugfs_mem(adap, "edc1", MEM_EDC1, EDRAM_SIZE_GET(size));
2919 if (is_t4(adap->params.chip)) {
2920 size = t4_read_reg(adap, MA_EXT_MEMORY_BAR);
2921 if (i & EXT_MEM_ENABLE)
2922 add_debugfs_mem(adap, "mc", MEM_MC,
2923 EXT_MEM_SIZE_GET(size));
2925 if (i & EXT_MEM_ENABLE) {
2926 size = t4_read_reg(adap, MA_EXT_MEMORY_BAR);
2927 add_debugfs_mem(adap, "mc0", MEM_MC0,
2928 EXT_MEM_SIZE_GET(size));
2930 if (i & EXT_MEM1_ENABLE) {
2931 size = t4_read_reg(adap, MA_EXT_MEMORY1_BAR);
2932 add_debugfs_mem(adap, "mc1", MEM_MC1,
2933 EXT_MEM_SIZE_GET(size));
2937 debugfs_create_file("l2t", S_IRUSR, adap->debugfs_root, adap,
2943 * upper-layer driver support
2947 * Allocate an active-open TID and set it to the supplied value.
2949 int cxgb4_alloc_atid(struct tid_info *t, void *data)
2953 spin_lock_bh(&t->atid_lock);
2955 union aopen_entry *p = t->afree;
2957 atid = (p - t->atid_tab) + t->atid_base;
2962 spin_unlock_bh(&t->atid_lock);
2965 EXPORT_SYMBOL(cxgb4_alloc_atid);
2968 * Release an active-open TID.
2970 void cxgb4_free_atid(struct tid_info *t, unsigned int atid)
2972 union aopen_entry *p = &t->atid_tab[atid - t->atid_base];
2974 spin_lock_bh(&t->atid_lock);
2978 spin_unlock_bh(&t->atid_lock);
2980 EXPORT_SYMBOL(cxgb4_free_atid);
2983 * Allocate a server TID and set it to the supplied value.
2985 int cxgb4_alloc_stid(struct tid_info *t, int family, void *data)
2989 spin_lock_bh(&t->stid_lock);
2990 if (family == PF_INET) {
2991 stid = find_first_zero_bit(t->stid_bmap, t->nstids);
2992 if (stid < t->nstids)
2993 __set_bit(stid, t->stid_bmap);
2997 stid = bitmap_find_free_region(t->stid_bmap, t->nstids, 2);
3002 t->stid_tab[stid].data = data;
3003 stid += t->stid_base;
3004 /* IPv6 requires max of 520 bits or 16 cells in TCAM
3005 * This is equivalent to 4 TIDs. With CLIP enabled it
3008 if (family == PF_INET)
3011 t->stids_in_use += 4;
3013 spin_unlock_bh(&t->stid_lock);
3016 EXPORT_SYMBOL(cxgb4_alloc_stid);
3018 /* Allocate a server filter TID and set it to the supplied value.
3020 int cxgb4_alloc_sftid(struct tid_info *t, int family, void *data)
3024 spin_lock_bh(&t->stid_lock);
3025 if (family == PF_INET) {
3026 stid = find_next_zero_bit(t->stid_bmap,
3027 t->nstids + t->nsftids, t->nstids);
3028 if (stid < (t->nstids + t->nsftids))
3029 __set_bit(stid, t->stid_bmap);
3036 t->stid_tab[stid].data = data;
3038 stid += t->sftid_base;
3041 spin_unlock_bh(&t->stid_lock);
3044 EXPORT_SYMBOL(cxgb4_alloc_sftid);
3046 /* Release a server TID.
3048 void cxgb4_free_stid(struct tid_info *t, unsigned int stid, int family)
3050 /* Is it a server filter TID? */
3051 if (t->nsftids && (stid >= t->sftid_base)) {
3052 stid -= t->sftid_base;
3055 stid -= t->stid_base;
3058 spin_lock_bh(&t->stid_lock);
3059 if (family == PF_INET)
3060 __clear_bit(stid, t->stid_bmap);
3062 bitmap_release_region(t->stid_bmap, stid, 2);
3063 t->stid_tab[stid].data = NULL;
3064 if (family == PF_INET)
3067 t->stids_in_use -= 4;
3068 spin_unlock_bh(&t->stid_lock);
3070 EXPORT_SYMBOL(cxgb4_free_stid);
3073 * Populate a TID_RELEASE WR. Caller must properly size the skb.
3075 static void mk_tid_release(struct sk_buff *skb, unsigned int chan,
3078 struct cpl_tid_release *req;
3080 set_wr_txq(skb, CPL_PRIORITY_SETUP, chan);
3081 req = (struct cpl_tid_release *)__skb_put(skb, sizeof(*req));
3082 INIT_TP_WR(req, tid);
3083 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, tid));
3087 * Queue a TID release request and if necessary schedule a work queue to
3090 static void cxgb4_queue_tid_release(struct tid_info *t, unsigned int chan,
3093 void **p = &t->tid_tab[tid];
3094 struct adapter *adap = container_of(t, struct adapter, tids);
3096 spin_lock_bh(&adap->tid_release_lock);
3097 *p = adap->tid_release_head;
3098 /* Low 2 bits encode the Tx channel number */
3099 adap->tid_release_head = (void **)((uintptr_t)p | chan);
3100 if (!adap->tid_release_task_busy) {
3101 adap->tid_release_task_busy = true;
3102 queue_work(workq, &adap->tid_release_task);
3104 spin_unlock_bh(&adap->tid_release_lock);
3108 * Process the list of pending TID release requests.
3110 static void process_tid_release_list(struct work_struct *work)
3112 struct sk_buff *skb;
3113 struct adapter *adap;
3115 adap = container_of(work, struct adapter, tid_release_task);
3117 spin_lock_bh(&adap->tid_release_lock);
3118 while (adap->tid_release_head) {
3119 void **p = adap->tid_release_head;
3120 unsigned int chan = (uintptr_t)p & 3;
3121 p = (void *)p - chan;
3123 adap->tid_release_head = *p;
3125 spin_unlock_bh(&adap->tid_release_lock);
3127 while (!(skb = alloc_skb(sizeof(struct cpl_tid_release),
3129 schedule_timeout_uninterruptible(1);
3131 mk_tid_release(skb, chan, p - adap->tids.tid_tab);
3132 t4_ofld_send(adap, skb);
3133 spin_lock_bh(&adap->tid_release_lock);
3135 adap->tid_release_task_busy = false;
3136 spin_unlock_bh(&adap->tid_release_lock);
3140 * Release a TID and inform HW. If we are unable to allocate the release
3141 * message we defer to a work queue.
3143 void cxgb4_remove_tid(struct tid_info *t, unsigned int chan, unsigned int tid)
3146 struct sk_buff *skb;
3147 struct adapter *adap = container_of(t, struct adapter, tids);
3149 old = t->tid_tab[tid];
3150 skb = alloc_skb(sizeof(struct cpl_tid_release), GFP_ATOMIC);
3152 t->tid_tab[tid] = NULL;
3153 mk_tid_release(skb, chan, tid);
3154 t4_ofld_send(adap, skb);
3156 cxgb4_queue_tid_release(t, chan, tid);
3158 atomic_dec(&t->tids_in_use);
3160 EXPORT_SYMBOL(cxgb4_remove_tid);
3163 * Allocate and initialize the TID tables. Returns 0 on success.
3165 static int tid_init(struct tid_info *t)
3168 unsigned int stid_bmap_size;
3169 unsigned int natids = t->natids;
3170 struct adapter *adap = container_of(t, struct adapter, tids);
3172 stid_bmap_size = BITS_TO_LONGS(t->nstids + t->nsftids);
3173 size = t->ntids * sizeof(*t->tid_tab) +
3174 natids * sizeof(*t->atid_tab) +
3175 t->nstids * sizeof(*t->stid_tab) +
3176 t->nsftids * sizeof(*t->stid_tab) +
3177 stid_bmap_size * sizeof(long) +
3178 t->nftids * sizeof(*t->ftid_tab) +
3179 t->nsftids * sizeof(*t->ftid_tab);
3181 t->tid_tab = t4_alloc_mem(size);
3185 t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
3186 t->stid_tab = (struct serv_entry *)&t->atid_tab[natids];
3187 t->stid_bmap = (unsigned long *)&t->stid_tab[t->nstids + t->nsftids];
3188 t->ftid_tab = (struct filter_entry *)&t->stid_bmap[stid_bmap_size];
3189 spin_lock_init(&t->stid_lock);
3190 spin_lock_init(&t->atid_lock);
3192 t->stids_in_use = 0;
3194 t->atids_in_use = 0;
3195 atomic_set(&t->tids_in_use, 0);
3197 /* Setup the free list for atid_tab and clear the stid bitmap. */
3200 t->atid_tab[natids - 1].next = &t->atid_tab[natids];
3201 t->afree = t->atid_tab;
3203 bitmap_zero(t->stid_bmap, t->nstids + t->nsftids);
3204 /* Reserve stid 0 for T4/T5 adapters */
3205 if (!t->stid_base &&
3206 (is_t4(adap->params.chip) || is_t5(adap->params.chip)))
3207 __set_bit(0, t->stid_bmap);
3212 static int cxgb4_clip_get(const struct net_device *dev,
3213 const struct in6_addr *lip)
3215 struct adapter *adap;
3216 struct fw_clip_cmd c;
3218 adap = netdev2adap(dev);
3219 memset(&c, 0, sizeof(c));
3220 c.op_to_write = htonl(FW_CMD_OP(FW_CLIP_CMD) |
3221 FW_CMD_REQUEST | FW_CMD_WRITE);
3222 c.alloc_to_len16 = htonl(F_FW_CLIP_CMD_ALLOC | FW_LEN16(c));
3223 *(__be64 *)&c.ip_hi = *(__be64 *)(lip->s6_addr);
3224 *(__be64 *)&c.ip_lo = *(__be64 *)(lip->s6_addr + 8);
3225 return t4_wr_mbox_meat(adap, adap->mbox, &c, sizeof(c), &c, false);
3228 static int cxgb4_clip_release(const struct net_device *dev,
3229 const struct in6_addr *lip)
3231 struct adapter *adap;
3232 struct fw_clip_cmd c;
3234 adap = netdev2adap(dev);
3235 memset(&c, 0, sizeof(c));
3236 c.op_to_write = htonl(FW_CMD_OP(FW_CLIP_CMD) |
3237 FW_CMD_REQUEST | FW_CMD_READ);
3238 c.alloc_to_len16 = htonl(F_FW_CLIP_CMD_FREE | FW_LEN16(c));
3239 *(__be64 *)&c.ip_hi = *(__be64 *)(lip->s6_addr);
3240 *(__be64 *)&c.ip_lo = *(__be64 *)(lip->s6_addr + 8);
3241 return t4_wr_mbox_meat(adap, adap->mbox, &c, sizeof(c), &c, false);
3245 * cxgb4_create_server - create an IP server
3247 * @stid: the server TID
3248 * @sip: local IP address to bind server to
3249 * @sport: the server's TCP port
3250 * @queue: queue to direct messages from this server to
3252 * Create an IP server for the given port and address.
3253 * Returns <0 on error and one of the %NET_XMIT_* values on success.
3255 int cxgb4_create_server(const struct net_device *dev, unsigned int stid,
3256 __be32 sip, __be16 sport, __be16 vlan,
3260 struct sk_buff *skb;
3261 struct adapter *adap;
3262 struct cpl_pass_open_req *req;
3265 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
3269 adap = netdev2adap(dev);
3270 req = (struct cpl_pass_open_req *)__skb_put(skb, sizeof(*req));
3272 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, stid));
3273 req->local_port = sport;
3274 req->peer_port = htons(0);
3275 req->local_ip = sip;
3276 req->peer_ip = htonl(0);
3277 chan = rxq_to_chan(&adap->sge, queue);
3278 req->opt0 = cpu_to_be64(TX_CHAN(chan));
3279 req->opt1 = cpu_to_be64(CONN_POLICY_ASK |
3280 SYN_RSS_ENABLE | SYN_RSS_QUEUE(queue));
3281 ret = t4_mgmt_tx(adap, skb);
3282 return net_xmit_eval(ret);
3284 EXPORT_SYMBOL(cxgb4_create_server);
3286 /* cxgb4_create_server6 - create an IPv6 server
3288 * @stid: the server TID
3289 * @sip: local IPv6 address to bind server to
3290 * @sport: the server's TCP port
3291 * @queue: queue to direct messages from this server to
3293 * Create an IPv6 server for the given port and address.
3294 * Returns <0 on error and one of the %NET_XMIT_* values on success.
3296 int cxgb4_create_server6(const struct net_device *dev, unsigned int stid,
3297 const struct in6_addr *sip, __be16 sport,
3301 struct sk_buff *skb;
3302 struct adapter *adap;
3303 struct cpl_pass_open_req6 *req;
3306 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
3310 adap = netdev2adap(dev);
3311 req = (struct cpl_pass_open_req6 *)__skb_put(skb, sizeof(*req));
3313 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ6, stid));
3314 req->local_port = sport;
3315 req->peer_port = htons(0);
3316 req->local_ip_hi = *(__be64 *)(sip->s6_addr);
3317 req->local_ip_lo = *(__be64 *)(sip->s6_addr + 8);
3318 req->peer_ip_hi = cpu_to_be64(0);
3319 req->peer_ip_lo = cpu_to_be64(0);
3320 chan = rxq_to_chan(&adap->sge, queue);
3321 req->opt0 = cpu_to_be64(TX_CHAN(chan));
3322 req->opt1 = cpu_to_be64(CONN_POLICY_ASK |
3323 SYN_RSS_ENABLE | SYN_RSS_QUEUE(queue));
3324 ret = t4_mgmt_tx(adap, skb);
3325 return net_xmit_eval(ret);
3327 EXPORT_SYMBOL(cxgb4_create_server6);
3329 int cxgb4_remove_server(const struct net_device *dev, unsigned int stid,
3330 unsigned int queue, bool ipv6)
3332 struct sk_buff *skb;
3333 struct adapter *adap;
3334 struct cpl_close_listsvr_req *req;
3337 adap = netdev2adap(dev);
3339 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
3343 req = (struct cpl_close_listsvr_req *)__skb_put(skb, sizeof(*req));
3345 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, stid));
3346 req->reply_ctrl = htons(NO_REPLY(0) | (ipv6 ? LISTSVR_IPV6(1) :
3347 LISTSVR_IPV6(0)) | QUEUENO(queue));
3348 ret = t4_mgmt_tx(adap, skb);
3349 return net_xmit_eval(ret);
3351 EXPORT_SYMBOL(cxgb4_remove_server);
3354 * cxgb4_best_mtu - find the entry in the MTU table closest to an MTU
3355 * @mtus: the HW MTU table
3356 * @mtu: the target MTU
3357 * @idx: index of selected entry in the MTU table
3359 * Returns the index and the value in the HW MTU table that is closest to
3360 * but does not exceed @mtu, unless @mtu is smaller than any value in the
3361 * table, in which case that smallest available value is selected.
3363 unsigned int cxgb4_best_mtu(const unsigned short *mtus, unsigned short mtu,
3368 while (i < NMTUS - 1 && mtus[i + 1] <= mtu)
3374 EXPORT_SYMBOL(cxgb4_best_mtu);
3377 * cxgb4_port_chan - get the HW channel of a port
3378 * @dev: the net device for the port
3380 * Return the HW Tx channel of the given port.
3382 unsigned int cxgb4_port_chan(const struct net_device *dev)
3384 return netdev2pinfo(dev)->tx_chan;
3386 EXPORT_SYMBOL(cxgb4_port_chan);
3388 unsigned int cxgb4_dbfifo_count(const struct net_device *dev, int lpfifo)
3390 struct adapter *adap = netdev2adap(dev);
3391 u32 v1, v2, lp_count, hp_count;
3393 v1 = t4_read_reg(adap, A_SGE_DBFIFO_STATUS);
3394 v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2);
3395 if (is_t4(adap->params.chip)) {
3396 lp_count = G_LP_COUNT(v1);
3397 hp_count = G_HP_COUNT(v1);
3399 lp_count = G_LP_COUNT_T5(v1);
3400 hp_count = G_HP_COUNT_T5(v2);
3402 return lpfifo ? lp_count : hp_count;
3404 EXPORT_SYMBOL(cxgb4_dbfifo_count);
3407 * cxgb4_port_viid - get the VI id of a port
3408 * @dev: the net device for the port
3410 * Return the VI id of the given port.
3412 unsigned int cxgb4_port_viid(const struct net_device *dev)
3414 return netdev2pinfo(dev)->viid;
3416 EXPORT_SYMBOL(cxgb4_port_viid);
3419 * cxgb4_port_idx - get the index of a port
3420 * @dev: the net device for the port
3422 * Return the index of the given port.
3424 unsigned int cxgb4_port_idx(const struct net_device *dev)
3426 return netdev2pinfo(dev)->port_id;
3428 EXPORT_SYMBOL(cxgb4_port_idx);
3430 void cxgb4_get_tcp_stats(struct pci_dev *pdev, struct tp_tcp_stats *v4,
3431 struct tp_tcp_stats *v6)
3433 struct adapter *adap = pci_get_drvdata(pdev);
3435 spin_lock(&adap->stats_lock);
3436 t4_tp_get_tcp_stats(adap, v4, v6);
3437 spin_unlock(&adap->stats_lock);
3439 EXPORT_SYMBOL(cxgb4_get_tcp_stats);
3441 void cxgb4_iscsi_init(struct net_device *dev, unsigned int tag_mask,
3442 const unsigned int *pgsz_order)
3444 struct adapter *adap = netdev2adap(dev);
3446 t4_write_reg(adap, ULP_RX_ISCSI_TAGMASK, tag_mask);
3447 t4_write_reg(adap, ULP_RX_ISCSI_PSZ, HPZ0(pgsz_order[0]) |
3448 HPZ1(pgsz_order[1]) | HPZ2(pgsz_order[2]) |
3449 HPZ3(pgsz_order[3]));
3451 EXPORT_SYMBOL(cxgb4_iscsi_init);
3453 int cxgb4_flush_eq_cache(struct net_device *dev)
3455 struct adapter *adap = netdev2adap(dev);
3458 ret = t4_fwaddrspace_write(adap, adap->mbox,
3459 0xe1000000 + A_SGE_CTXT_CMD, 0x20000000);
3462 EXPORT_SYMBOL(cxgb4_flush_eq_cache);
3464 static int read_eq_indices(struct adapter *adap, u16 qid, u16 *pidx, u16 *cidx)
3466 u32 addr = t4_read_reg(adap, A_SGE_DBQ_CTXT_BADDR) + 24 * qid + 8;
3470 ret = t4_mem_win_read_len(adap, addr, (__be32 *)&indices, 8);
3472 *cidx = (be64_to_cpu(indices) >> 25) & 0xffff;
3473 *pidx = (be64_to_cpu(indices) >> 9) & 0xffff;
3478 int cxgb4_sync_txq_pidx(struct net_device *dev, u16 qid, u16 pidx,
3481 struct adapter *adap = netdev2adap(dev);
3482 u16 hw_pidx, hw_cidx;
3485 ret = read_eq_indices(adap, qid, &hw_pidx, &hw_cidx);
3489 if (pidx != hw_pidx) {
3492 if (pidx >= hw_pidx)
3493 delta = pidx - hw_pidx;
3495 delta = size - hw_pidx + pidx;
3497 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL),
3498 QID(qid) | PIDX(delta));
3503 EXPORT_SYMBOL(cxgb4_sync_txq_pidx);
3505 void cxgb4_disable_db_coalescing(struct net_device *dev)
3507 struct adapter *adap;
3509 adap = netdev2adap(dev);
3510 t4_set_reg_field(adap, A_SGE_DOORBELL_CONTROL, F_NOCOALESCE,
3513 EXPORT_SYMBOL(cxgb4_disable_db_coalescing);
3515 void cxgb4_enable_db_coalescing(struct net_device *dev)
3517 struct adapter *adap;
3519 adap = netdev2adap(dev);
3520 t4_set_reg_field(adap, A_SGE_DOORBELL_CONTROL, F_NOCOALESCE, 0);
3522 EXPORT_SYMBOL(cxgb4_enable_db_coalescing);
3524 static struct pci_driver cxgb4_driver;
3526 static void check_neigh_update(struct neighbour *neigh)
3528 const struct device *parent;
3529 const struct net_device *netdev = neigh->dev;
3531 if (netdev->priv_flags & IFF_802_1Q_VLAN)
3532 netdev = vlan_dev_real_dev(netdev);
3533 parent = netdev->dev.parent;
3534 if (parent && parent->driver == &cxgb4_driver.driver)
3535 t4_l2t_update(dev_get_drvdata(parent), neigh);
3538 static int netevent_cb(struct notifier_block *nb, unsigned long event,
3542 case NETEVENT_NEIGH_UPDATE:
3543 check_neigh_update(data);
3545 case NETEVENT_REDIRECT:
3552 static bool netevent_registered;
3553 static struct notifier_block cxgb4_netevent_nb = {
3554 .notifier_call = netevent_cb
3557 static void drain_db_fifo(struct adapter *adap, int usecs)
3559 u32 v1, v2, lp_count, hp_count;
3562 v1 = t4_read_reg(adap, A_SGE_DBFIFO_STATUS);
3563 v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2);
3564 if (is_t4(adap->params.chip)) {
3565 lp_count = G_LP_COUNT(v1);
3566 hp_count = G_HP_COUNT(v1);
3568 lp_count = G_LP_COUNT_T5(v1);
3569 hp_count = G_HP_COUNT_T5(v2);
3572 if (lp_count == 0 && hp_count == 0)
3574 set_current_state(TASK_UNINTERRUPTIBLE);
3575 schedule_timeout(usecs_to_jiffies(usecs));
3579 static void disable_txq_db(struct sge_txq *q)
3581 spin_lock_irq(&q->db_lock);
3583 spin_unlock_irq(&q->db_lock);
3586 static void enable_txq_db(struct sge_txq *q)
3588 spin_lock_irq(&q->db_lock);
3590 spin_unlock_irq(&q->db_lock);
3593 static void disable_dbs(struct adapter *adap)
3597 for_each_ethrxq(&adap->sge, i)
3598 disable_txq_db(&adap->sge.ethtxq[i].q);
3599 for_each_ofldrxq(&adap->sge, i)
3600 disable_txq_db(&adap->sge.ofldtxq[i].q);
3601 for_each_port(adap, i)
3602 disable_txq_db(&adap->sge.ctrlq[i].q);
3605 static void enable_dbs(struct adapter *adap)
3609 for_each_ethrxq(&adap->sge, i)
3610 enable_txq_db(&adap->sge.ethtxq[i].q);
3611 for_each_ofldrxq(&adap->sge, i)
3612 enable_txq_db(&adap->sge.ofldtxq[i].q);
3613 for_each_port(adap, i)
3614 enable_txq_db(&adap->sge.ctrlq[i].q);
3617 static void sync_txq_pidx(struct adapter *adap, struct sge_txq *q)
3619 u16 hw_pidx, hw_cidx;
3622 spin_lock_bh(&q->db_lock);
3623 ret = read_eq_indices(adap, (u16)q->cntxt_id, &hw_pidx, &hw_cidx);
3626 if (q->db_pidx != hw_pidx) {
3629 if (q->db_pidx >= hw_pidx)
3630 delta = q->db_pidx - hw_pidx;
3632 delta = q->size - hw_pidx + q->db_pidx;
3634 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL),
3635 QID(q->cntxt_id) | PIDX(delta));
3639 spin_unlock_bh(&q->db_lock);
3641 CH_WARN(adap, "DB drop recovery failed.\n");
3643 static void recover_all_queues(struct adapter *adap)
3647 for_each_ethrxq(&adap->sge, i)
3648 sync_txq_pidx(adap, &adap->sge.ethtxq[i].q);
3649 for_each_ofldrxq(&adap->sge, i)
3650 sync_txq_pidx(adap, &adap->sge.ofldtxq[i].q);
3651 for_each_port(adap, i)
3652 sync_txq_pidx(adap, &adap->sge.ctrlq[i].q);
3655 static void notify_rdma_uld(struct adapter *adap, enum cxgb4_control cmd)
3657 mutex_lock(&uld_mutex);
3658 if (adap->uld_handle[CXGB4_ULD_RDMA])
3659 ulds[CXGB4_ULD_RDMA].control(adap->uld_handle[CXGB4_ULD_RDMA],
3661 mutex_unlock(&uld_mutex);
3664 static void process_db_full(struct work_struct *work)
3666 struct adapter *adap;
3668 adap = container_of(work, struct adapter, db_full_task);
3670 notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
3671 drain_db_fifo(adap, dbfifo_drain_delay);
3672 t4_set_reg_field(adap, SGE_INT_ENABLE3,
3673 DBFIFO_HP_INT | DBFIFO_LP_INT,
3674 DBFIFO_HP_INT | DBFIFO_LP_INT);
3675 notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
3678 static void process_db_drop(struct work_struct *work)
3680 struct adapter *adap;
3682 adap = container_of(work, struct adapter, db_drop_task);
3684 if (is_t4(adap->params.chip)) {
3686 notify_rdma_uld(adap, CXGB4_CONTROL_DB_DROP);
3687 drain_db_fifo(adap, 1);
3688 recover_all_queues(adap);
3691 u32 dropped_db = t4_read_reg(adap, 0x010ac);
3692 u16 qid = (dropped_db >> 15) & 0x1ffff;
3693 u16 pidx_inc = dropped_db & 0x1fff;
3695 unsigned short udb_density;
3696 unsigned long qpshift;
3700 dev_warn(adap->pdev_dev,
3701 "Dropped DB 0x%x qid %d bar2 %d coalesce %d pidx %d\n",
3703 (dropped_db >> 14) & 1,
3704 (dropped_db >> 13) & 1,
3707 drain_db_fifo(adap, 1);
3709 s_qpp = QUEUESPERPAGEPF1 * adap->fn;
3710 udb_density = 1 << QUEUESPERPAGEPF0_GET(t4_read_reg(adap,
3711 SGE_EGRESS_QUEUES_PER_PAGE_PF) >> s_qpp);
3712 qpshift = PAGE_SHIFT - ilog2(udb_density);
3713 udb = qid << qpshift;
3715 page = udb / PAGE_SIZE;
3716 udb += (qid - (page * udb_density)) * 128;
3718 writel(PIDX(pidx_inc), adap->bar2 + udb + 8);
3720 /* Re-enable BAR2 WC */
3721 t4_set_reg_field(adap, 0x10b0, 1<<15, 1<<15);
3724 t4_set_reg_field(adap, A_SGE_DOORBELL_CONTROL, F_DROPPED_DB, 0);
3727 void t4_db_full(struct adapter *adap)
3729 if (is_t4(adap->params.chip)) {
3730 t4_set_reg_field(adap, SGE_INT_ENABLE3,
3731 DBFIFO_HP_INT | DBFIFO_LP_INT, 0);
3732 queue_work(workq, &adap->db_full_task);
3736 void t4_db_dropped(struct adapter *adap)
3738 if (is_t4(adap->params.chip))
3739 queue_work(workq, &adap->db_drop_task);
3742 static void uld_attach(struct adapter *adap, unsigned int uld)
3745 struct cxgb4_lld_info lli;
3748 lli.pdev = adap->pdev;
3749 lli.l2t = adap->l2t;
3750 lli.tids = &adap->tids;
3751 lli.ports = adap->port;
3752 lli.vr = &adap->vres;
3753 lli.mtus = adap->params.mtus;
3754 if (uld == CXGB4_ULD_RDMA) {
3755 lli.rxq_ids = adap->sge.rdma_rxq;
3756 lli.nrxq = adap->sge.rdmaqs;
3757 } else if (uld == CXGB4_ULD_ISCSI) {
3758 lli.rxq_ids = adap->sge.ofld_rxq;
3759 lli.nrxq = adap->sge.ofldqsets;
3761 lli.ntxq = adap->sge.ofldqsets;
3762 lli.nchan = adap->params.nports;
3763 lli.nports = adap->params.nports;
3764 lli.wr_cred = adap->params.ofldq_wr_cred;
3765 lli.adapter_type = adap->params.chip;
3766 lli.iscsi_iolen = MAXRXDATA_GET(t4_read_reg(adap, TP_PARA_REG2));
3767 lli.udb_density = 1 << QUEUESPERPAGEPF0_GET(
3768 t4_read_reg(adap, SGE_EGRESS_QUEUES_PER_PAGE_PF) >>
3770 lli.ucq_density = 1 << QUEUESPERPAGEPF0_GET(
3771 t4_read_reg(adap, SGE_INGRESS_QUEUES_PER_PAGE_PF) >>
3773 lli.filt_mode = adap->params.tp.vlan_pri_map;
3774 /* MODQ_REQ_MAP sets queues 0-3 to chan 0-3 */
3775 for (i = 0; i < NCHAN; i++)
3777 lli.gts_reg = adap->regs + MYPF_REG(SGE_PF_GTS);
3778 lli.db_reg = adap->regs + MYPF_REG(SGE_PF_KDOORBELL);
3779 lli.fw_vers = adap->params.fw_vers;
3780 lli.dbfifo_int_thresh = dbfifo_int_thresh;
3781 lli.sge_pktshift = adap->sge.pktshift;
3782 lli.enable_fw_ofld_conn = adap->flags & FW_OFLD_CONN;
3783 lli.ulptx_memwrite_dsgl = adap->params.ulptx_memwrite_dsgl;
3785 handle = ulds[uld].add(&lli);
3786 if (IS_ERR(handle)) {
3787 dev_warn(adap->pdev_dev,
3788 "could not attach to the %s driver, error %ld\n",
3789 uld_str[uld], PTR_ERR(handle));
3793 adap->uld_handle[uld] = handle;
3795 if (!netevent_registered) {
3796 register_netevent_notifier(&cxgb4_netevent_nb);
3797 netevent_registered = true;
3800 if (adap->flags & FULL_INIT_DONE)
3801 ulds[uld].state_change(handle, CXGB4_STATE_UP);
3804 static void attach_ulds(struct adapter *adap)
3808 spin_lock(&adap_rcu_lock);
3809 list_add_tail_rcu(&adap->rcu_node, &adap_rcu_list);
3810 spin_unlock(&adap_rcu_lock);
3812 mutex_lock(&uld_mutex);
3813 list_add_tail(&adap->list_node, &adapter_list);
3814 for (i = 0; i < CXGB4_ULD_MAX; i++)
3816 uld_attach(adap, i);
3817 mutex_unlock(&uld_mutex);
3820 static void detach_ulds(struct adapter *adap)
3824 mutex_lock(&uld_mutex);
3825 list_del(&adap->list_node);
3826 for (i = 0; i < CXGB4_ULD_MAX; i++)
3827 if (adap->uld_handle[i]) {
3828 ulds[i].state_change(adap->uld_handle[i],
3829 CXGB4_STATE_DETACH);
3830 adap->uld_handle[i] = NULL;
3832 if (netevent_registered && list_empty(&adapter_list)) {
3833 unregister_netevent_notifier(&cxgb4_netevent_nb);
3834 netevent_registered = false;
3836 mutex_unlock(&uld_mutex);
3838 spin_lock(&adap_rcu_lock);
3839 list_del_rcu(&adap->rcu_node);
3840 spin_unlock(&adap_rcu_lock);
3843 static void notify_ulds(struct adapter *adap, enum cxgb4_state new_state)
3847 mutex_lock(&uld_mutex);
3848 for (i = 0; i < CXGB4_ULD_MAX; i++)
3849 if (adap->uld_handle[i])
3850 ulds[i].state_change(adap->uld_handle[i], new_state);
3851 mutex_unlock(&uld_mutex);
3855 * cxgb4_register_uld - register an upper-layer driver
3856 * @type: the ULD type
3857 * @p: the ULD methods
3859 * Registers an upper-layer driver with this driver and notifies the ULD
3860 * about any presently available devices that support its type. Returns
3861 * %-EBUSY if a ULD of the same type is already registered.
3863 int cxgb4_register_uld(enum cxgb4_uld type, const struct cxgb4_uld_info *p)
3866 struct adapter *adap;
3868 if (type >= CXGB4_ULD_MAX)
3870 mutex_lock(&uld_mutex);
3871 if (ulds[type].add) {
3876 list_for_each_entry(adap, &adapter_list, list_node)
3877 uld_attach(adap, type);
3878 out: mutex_unlock(&uld_mutex);
3881 EXPORT_SYMBOL(cxgb4_register_uld);
3884 * cxgb4_unregister_uld - unregister an upper-layer driver
3885 * @type: the ULD type
3887 * Unregisters an existing upper-layer driver.
3889 int cxgb4_unregister_uld(enum cxgb4_uld type)
3891 struct adapter *adap;
3893 if (type >= CXGB4_ULD_MAX)
3895 mutex_lock(&uld_mutex);
3896 list_for_each_entry(adap, &adapter_list, list_node)
3897 adap->uld_handle[type] = NULL;
3898 ulds[type].add = NULL;
3899 mutex_unlock(&uld_mutex);
3902 EXPORT_SYMBOL(cxgb4_unregister_uld);
3904 /* Check if netdev on which event is occured belongs to us or not. Return
3905 * suceess (1) if it belongs otherwise failure (0).
3907 static int cxgb4_netdev(struct net_device *netdev)
3909 struct adapter *adap;
3912 spin_lock(&adap_rcu_lock);
3913 list_for_each_entry_rcu(adap, &adap_rcu_list, rcu_node)
3914 for (i = 0; i < MAX_NPORTS; i++)
3915 if (adap->port[i] == netdev) {
3916 spin_unlock(&adap_rcu_lock);
3919 spin_unlock(&adap_rcu_lock);
3923 static int clip_add(struct net_device *event_dev, struct inet6_ifaddr *ifa,
3924 unsigned long event)
3926 int ret = NOTIFY_DONE;
3929 if (cxgb4_netdev(event_dev)) {
3932 ret = cxgb4_clip_get(event_dev,
3933 (const struct in6_addr *)ifa->addr.s6_addr);
3941 cxgb4_clip_release(event_dev,
3942 (const struct in6_addr *)ifa->addr.s6_addr);
3953 static int cxgb4_inet6addr_handler(struct notifier_block *this,
3954 unsigned long event, void *data)
3956 struct inet6_ifaddr *ifa = data;
3957 struct net_device *event_dev;
3958 int ret = NOTIFY_DONE;
3959 struct bonding *bond = netdev_priv(ifa->idev->dev);
3960 struct list_head *iter;
3961 struct slave *slave;
3962 struct pci_dev *first_pdev = NULL;
3964 if (ifa->idev->dev->priv_flags & IFF_802_1Q_VLAN) {
3965 event_dev = vlan_dev_real_dev(ifa->idev->dev);
3966 ret = clip_add(event_dev, ifa, event);
3967 } else if (ifa->idev->dev->flags & IFF_MASTER) {
3968 /* It is possible that two different adapters are bonded in one
3969 * bond. We need to find such different adapters and add clip
3970 * in all of them only once.
3972 read_lock(&bond->lock);
3973 bond_for_each_slave(bond, slave, iter) {
3975 ret = clip_add(slave->dev, ifa, event);
3976 /* If clip_add is success then only initialize
3977 * first_pdev since it means it is our device
3979 if (ret == NOTIFY_OK)
3980 first_pdev = to_pci_dev(
3981 slave->dev->dev.parent);
3982 } else if (first_pdev !=
3983 to_pci_dev(slave->dev->dev.parent))
3984 ret = clip_add(slave->dev, ifa, event);
3986 read_unlock(&bond->lock);
3988 ret = clip_add(ifa->idev->dev, ifa, event);
3993 static struct notifier_block cxgb4_inet6addr_notifier = {
3994 .notifier_call = cxgb4_inet6addr_handler
3997 /* Retrieves IPv6 addresses from a root device (bond, vlan) associated with
3998 * a physical device.
3999 * The physical device reference is needed to send the actul CLIP command.
4001 static int update_dev_clip(struct net_device *root_dev, struct net_device *dev)
4003 struct inet6_dev *idev = NULL;
4004 struct inet6_ifaddr *ifa;
4007 idev = __in6_dev_get(root_dev);
4011 read_lock_bh(&idev->lock);
4012 list_for_each_entry(ifa, &idev->addr_list, if_list) {
4013 ret = cxgb4_clip_get(dev,
4014 (const struct in6_addr *)ifa->addr.s6_addr);
4018 read_unlock_bh(&idev->lock);
4023 static int update_root_dev_clip(struct net_device *dev)
4025 struct net_device *root_dev = NULL;
4028 /* First populate the real net device's IPv6 addresses */
4029 ret = update_dev_clip(dev, dev);
4033 /* Parse all bond and vlan devices layered on top of the physical dev */
4034 for (i = 0; i < VLAN_N_VID; i++) {
4035 root_dev = __vlan_find_dev_deep(dev, htons(ETH_P_8021Q), i);
4039 ret = update_dev_clip(root_dev, dev);
4046 static void update_clip(const struct adapter *adap)
4049 struct net_device *dev;
4054 for (i = 0; i < MAX_NPORTS; i++) {
4055 dev = adap->port[i];
4059 ret = update_root_dev_clip(dev);
4068 * cxgb_up - enable the adapter
4069 * @adap: adapter being enabled
4071 * Called when the first port is enabled, this function performs the
4072 * actions necessary to make an adapter operational, such as completing
4073 * the initialization of HW modules, and enabling interrupts.
4075 * Must be called with the rtnl lock held.
4077 static int cxgb_up(struct adapter *adap)
4081 err = setup_sge_queues(adap);
4084 err = setup_rss(adap);
4088 if (adap->flags & USING_MSIX) {
4089 name_msix_vecs(adap);
4090 err = request_irq(adap->msix_info[0].vec, t4_nondata_intr, 0,
4091 adap->msix_info[0].desc, adap);
4095 err = request_msix_queue_irqs(adap);
4097 free_irq(adap->msix_info[0].vec, adap);
4101 err = request_irq(adap->pdev->irq, t4_intr_handler(adap),
4102 (adap->flags & USING_MSI) ? 0 : IRQF_SHARED,
4103 adap->port[0]->name, adap);
4109 t4_intr_enable(adap);
4110 adap->flags |= FULL_INIT_DONE;
4111 notify_ulds(adap, CXGB4_STATE_UP);
4116 dev_err(adap->pdev_dev, "request_irq failed, err %d\n", err);
4118 t4_free_sge_resources(adap);
4122 static void cxgb_down(struct adapter *adapter)
4124 t4_intr_disable(adapter);
4125 cancel_work_sync(&adapter->tid_release_task);
4126 cancel_work_sync(&adapter->db_full_task);
4127 cancel_work_sync(&adapter->db_drop_task);
4128 adapter->tid_release_task_busy = false;
4129 adapter->tid_release_head = NULL;
4131 if (adapter->flags & USING_MSIX) {
4132 free_msix_queue_irqs(adapter);
4133 free_irq(adapter->msix_info[0].vec, adapter);
4135 free_irq(adapter->pdev->irq, adapter);
4136 quiesce_rx(adapter);
4137 t4_sge_stop(adapter);
4138 t4_free_sge_resources(adapter);
4139 adapter->flags &= ~FULL_INIT_DONE;
4143 * net_device operations
4145 static int cxgb_open(struct net_device *dev)
4148 struct port_info *pi = netdev_priv(dev);
4149 struct adapter *adapter = pi->adapter;
4151 netif_carrier_off(dev);
4153 if (!(adapter->flags & FULL_INIT_DONE)) {
4154 err = cxgb_up(adapter);
4159 err = link_start(dev);
4161 netif_tx_start_all_queues(dev);
4165 static int cxgb_close(struct net_device *dev)
4167 struct port_info *pi = netdev_priv(dev);
4168 struct adapter *adapter = pi->adapter;
4170 netif_tx_stop_all_queues(dev);
4171 netif_carrier_off(dev);
4172 return t4_enable_vi(adapter, adapter->fn, pi->viid, false, false);
4175 /* Return an error number if the indicated filter isn't writable ...
4177 static int writable_filter(struct filter_entry *f)
4187 /* Delete the filter at the specified index (if valid). The checks for all
4188 * the common problems with doing this like the filter being locked, currently
4189 * pending in another operation, etc.
4191 static int delete_filter(struct adapter *adapter, unsigned int fidx)
4193 struct filter_entry *f;
4196 if (fidx >= adapter->tids.nftids + adapter->tids.nsftids)
4199 f = &adapter->tids.ftid_tab[fidx];
4200 ret = writable_filter(f);
4204 return del_filter_wr(adapter, fidx);
4209 int cxgb4_create_server_filter(const struct net_device *dev, unsigned int stid,
4210 __be32 sip, __be16 sport, __be16 vlan,
4211 unsigned int queue, unsigned char port, unsigned char mask)
4214 struct filter_entry *f;
4215 struct adapter *adap;
4219 adap = netdev2adap(dev);
4221 /* Adjust stid to correct filter index */
4222 stid -= adap->tids.sftid_base;
4223 stid += adap->tids.nftids;
4225 /* Check to make sure the filter requested is writable ...
4227 f = &adap->tids.ftid_tab[stid];
4228 ret = writable_filter(f);
4232 /* Clear out any old resources being used by the filter before
4233 * we start constructing the new filter.
4236 clear_filter(adap, f);
4238 /* Clear out filter specifications */
4239 memset(&f->fs, 0, sizeof(struct ch_filter_specification));
4240 f->fs.val.lport = cpu_to_be16(sport);
4241 f->fs.mask.lport = ~0;
4243 if ((val[0] | val[1] | val[2] | val[3]) != 0) {
4244 for (i = 0; i < 4; i++) {
4245 f->fs.val.lip[i] = val[i];
4246 f->fs.mask.lip[i] = ~0;
4248 if (adap->params.tp.vlan_pri_map & F_PORT) {
4249 f->fs.val.iport = port;
4250 f->fs.mask.iport = mask;
4254 if (adap->params.tp.vlan_pri_map & F_PROTOCOL) {
4255 f->fs.val.proto = IPPROTO_TCP;
4256 f->fs.mask.proto = ~0;
4261 /* Mark filter as locked */
4265 ret = set_filter_wr(adap, stid);
4267 clear_filter(adap, f);
4273 EXPORT_SYMBOL(cxgb4_create_server_filter);
4275 int cxgb4_remove_server_filter(const struct net_device *dev, unsigned int stid,
4276 unsigned int queue, bool ipv6)
4279 struct filter_entry *f;
4280 struct adapter *adap;
4282 adap = netdev2adap(dev);
4284 /* Adjust stid to correct filter index */
4285 stid -= adap->tids.sftid_base;
4286 stid += adap->tids.nftids;
4288 f = &adap->tids.ftid_tab[stid];
4289 /* Unlock the filter */
4292 ret = delete_filter(adap, stid);
4298 EXPORT_SYMBOL(cxgb4_remove_server_filter);
4300 static struct rtnl_link_stats64 *cxgb_get_stats(struct net_device *dev,
4301 struct rtnl_link_stats64 *ns)
4303 struct port_stats stats;
4304 struct port_info *p = netdev_priv(dev);
4305 struct adapter *adapter = p->adapter;
4307 /* Block retrieving statistics during EEH error
4308 * recovery. Otherwise, the recovery might fail
4309 * and the PCI device will be removed permanently
4311 spin_lock(&adapter->stats_lock);
4312 if (!netif_device_present(dev)) {
4313 spin_unlock(&adapter->stats_lock);
4316 t4_get_port_stats(adapter, p->tx_chan, &stats);
4317 spin_unlock(&adapter->stats_lock);
4319 ns->tx_bytes = stats.tx_octets;
4320 ns->tx_packets = stats.tx_frames;
4321 ns->rx_bytes = stats.rx_octets;
4322 ns->rx_packets = stats.rx_frames;
4323 ns->multicast = stats.rx_mcast_frames;
4325 /* detailed rx_errors */
4326 ns->rx_length_errors = stats.rx_jabber + stats.rx_too_long +
4328 ns->rx_over_errors = 0;
4329 ns->rx_crc_errors = stats.rx_fcs_err;
4330 ns->rx_frame_errors = stats.rx_symbol_err;
4331 ns->rx_fifo_errors = stats.rx_ovflow0 + stats.rx_ovflow1 +
4332 stats.rx_ovflow2 + stats.rx_ovflow3 +
4333 stats.rx_trunc0 + stats.rx_trunc1 +
4334 stats.rx_trunc2 + stats.rx_trunc3;
4335 ns->rx_missed_errors = 0;
4337 /* detailed tx_errors */
4338 ns->tx_aborted_errors = 0;
4339 ns->tx_carrier_errors = 0;
4340 ns->tx_fifo_errors = 0;
4341 ns->tx_heartbeat_errors = 0;
4342 ns->tx_window_errors = 0;
4344 ns->tx_errors = stats.tx_error_frames;
4345 ns->rx_errors = stats.rx_symbol_err + stats.rx_fcs_err +
4346 ns->rx_length_errors + stats.rx_len_err + ns->rx_fifo_errors;
4350 static int cxgb_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
4353 int ret = 0, prtad, devad;
4354 struct port_info *pi = netdev_priv(dev);
4355 struct mii_ioctl_data *data = (struct mii_ioctl_data *)&req->ifr_data;
4359 if (pi->mdio_addr < 0)
4361 data->phy_id = pi->mdio_addr;
4365 if (mdio_phy_id_is_c45(data->phy_id)) {
4366 prtad = mdio_phy_id_prtad(data->phy_id);
4367 devad = mdio_phy_id_devad(data->phy_id);
4368 } else if (data->phy_id < 32) {
4369 prtad = data->phy_id;
4371 data->reg_num &= 0x1f;
4375 mbox = pi->adapter->fn;
4376 if (cmd == SIOCGMIIREG)
4377 ret = t4_mdio_rd(pi->adapter, mbox, prtad, devad,
4378 data->reg_num, &data->val_out);
4380 ret = t4_mdio_wr(pi->adapter, mbox, prtad, devad,
4381 data->reg_num, data->val_in);
4389 static void cxgb_set_rxmode(struct net_device *dev)
4391 /* unfortunately we can't return errors to the stack */
4392 set_rxmode(dev, -1, false);
4395 static int cxgb_change_mtu(struct net_device *dev, int new_mtu)
4398 struct port_info *pi = netdev_priv(dev);
4400 if (new_mtu < 81 || new_mtu > MAX_MTU) /* accommodate SACK */
4402 ret = t4_set_rxmode(pi->adapter, pi->adapter->fn, pi->viid, new_mtu, -1,
4409 static int cxgb_set_mac_addr(struct net_device *dev, void *p)
4412 struct sockaddr *addr = p;
4413 struct port_info *pi = netdev_priv(dev);
4415 if (!is_valid_ether_addr(addr->sa_data))
4416 return -EADDRNOTAVAIL;
4418 ret = t4_change_mac(pi->adapter, pi->adapter->fn, pi->viid,
4419 pi->xact_addr_filt, addr->sa_data, true, true);
4423 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
4424 pi->xact_addr_filt = ret;
4428 #ifdef CONFIG_NET_POLL_CONTROLLER
4429 static void cxgb_netpoll(struct net_device *dev)
4431 struct port_info *pi = netdev_priv(dev);
4432 struct adapter *adap = pi->adapter;
4434 if (adap->flags & USING_MSIX) {
4436 struct sge_eth_rxq *rx = &adap->sge.ethrxq[pi->first_qset];
4438 for (i = pi->nqsets; i; i--, rx++)
4439 t4_sge_intr_msix(0, &rx->rspq);
4441 t4_intr_handler(adap)(0, adap);
4445 static const struct net_device_ops cxgb4_netdev_ops = {
4446 .ndo_open = cxgb_open,
4447 .ndo_stop = cxgb_close,
4448 .ndo_start_xmit = t4_eth_xmit,
4449 .ndo_get_stats64 = cxgb_get_stats,
4450 .ndo_set_rx_mode = cxgb_set_rxmode,
4451 .ndo_set_mac_address = cxgb_set_mac_addr,
4452 .ndo_set_features = cxgb_set_features,
4453 .ndo_validate_addr = eth_validate_addr,
4454 .ndo_do_ioctl = cxgb_ioctl,
4455 .ndo_change_mtu = cxgb_change_mtu,
4456 #ifdef CONFIG_NET_POLL_CONTROLLER
4457 .ndo_poll_controller = cxgb_netpoll,
4461 void t4_fatal_err(struct adapter *adap)
4463 t4_set_reg_field(adap, SGE_CONTROL, GLOBALENABLE, 0);
4464 t4_intr_disable(adap);
4465 dev_alert(adap->pdev_dev, "encountered fatal error, adapter stopped\n");
4468 static void setup_memwin(struct adapter *adap)
4470 u32 bar0, mem_win0_base, mem_win1_base, mem_win2_base;
4472 bar0 = pci_resource_start(adap->pdev, 0); /* truncation intentional */
4473 if (is_t4(adap->params.chip)) {
4474 mem_win0_base = bar0 + MEMWIN0_BASE;
4475 mem_win1_base = bar0 + MEMWIN1_BASE;
4476 mem_win2_base = bar0 + MEMWIN2_BASE;
4478 /* For T5, only relative offset inside the PCIe BAR is passed */
4479 mem_win0_base = MEMWIN0_BASE;
4480 mem_win1_base = MEMWIN1_BASE_T5;
4481 mem_win2_base = MEMWIN2_BASE_T5;
4483 t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 0),
4484 mem_win0_base | BIR(0) |
4485 WINDOW(ilog2(MEMWIN0_APERTURE) - 10));
4486 t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 1),
4487 mem_win1_base | BIR(0) |
4488 WINDOW(ilog2(MEMWIN1_APERTURE) - 10));
4489 t4_write_reg(adap, PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 2),
4490 mem_win2_base | BIR(0) |
4491 WINDOW(ilog2(MEMWIN2_APERTURE) - 10));
4494 static void setup_memwin_rdma(struct adapter *adap)
4496 if (adap->vres.ocq.size) {
4497 unsigned int start, sz_kb;
4499 start = pci_resource_start(adap->pdev, 2) +
4500 OCQ_WIN_OFFSET(adap->pdev, &adap->vres);
4501 sz_kb = roundup_pow_of_two(adap->vres.ocq.size) >> 10;
4503 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN, 3),
4504 start | BIR(1) | WINDOW(ilog2(sz_kb)));
4506 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET, 3),
4507 adap->vres.ocq.start);
4509 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET, 3));
4513 static int adap_init1(struct adapter *adap, struct fw_caps_config_cmd *c)
4518 /* get device capabilities */
4519 memset(c, 0, sizeof(*c));
4520 c->op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4521 FW_CMD_REQUEST | FW_CMD_READ);
4522 c->cfvalid_to_len16 = htonl(FW_LEN16(*c));
4523 ret = t4_wr_mbox(adap, adap->fn, c, sizeof(*c), c);
4527 /* select capabilities we'll be using */
4528 if (c->niccaps & htons(FW_CAPS_CONFIG_NIC_VM)) {
4530 c->niccaps ^= htons(FW_CAPS_CONFIG_NIC_VM);
4532 c->niccaps = htons(FW_CAPS_CONFIG_NIC_VM);
4533 } else if (vf_acls) {
4534 dev_err(adap->pdev_dev, "virtualization ACLs not supported");
4537 c->op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4538 FW_CMD_REQUEST | FW_CMD_WRITE);
4539 ret = t4_wr_mbox(adap, adap->fn, c, sizeof(*c), NULL);
4543 ret = t4_config_glbl_rss(adap, adap->fn,
4544 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL,
4545 FW_RSS_GLB_CONFIG_CMD_TNLMAPEN |
4546 FW_RSS_GLB_CONFIG_CMD_TNLALLLKP);
4550 ret = t4_cfg_pfvf(adap, adap->fn, adap->fn, 0, MAX_EGRQ, 64, MAX_INGQ,
4551 0, 0, 4, 0xf, 0xf, 16, FW_CMD_CAP_PF, FW_CMD_CAP_PF);
4557 /* tweak some settings */
4558 t4_write_reg(adap, TP_SHIFT_CNT, 0x64f8849);
4559 t4_write_reg(adap, ULP_RX_TDDP_PSZ, HPZ0(PAGE_SHIFT - 12));
4560 t4_write_reg(adap, TP_PIO_ADDR, TP_INGRESS_CONFIG);
4561 v = t4_read_reg(adap, TP_PIO_DATA);
4562 t4_write_reg(adap, TP_PIO_DATA, v & ~CSUM_HAS_PSEUDO_HDR);
4564 /* first 4 Tx modulation queues point to consecutive Tx channels */
4565 adap->params.tp.tx_modq_map = 0xE4;
4566 t4_write_reg(adap, A_TP_TX_MOD_QUEUE_REQ_MAP,
4567 V_TX_MOD_QUEUE_REQ_MAP(adap->params.tp.tx_modq_map));
4569 /* associate each Tx modulation queue with consecutive Tx channels */
4571 t4_write_indirect(adap, TP_PIO_ADDR, TP_PIO_DATA,
4572 &v, 1, A_TP_TX_SCHED_HDR);
4573 t4_write_indirect(adap, TP_PIO_ADDR, TP_PIO_DATA,
4574 &v, 1, A_TP_TX_SCHED_FIFO);
4575 t4_write_indirect(adap, TP_PIO_ADDR, TP_PIO_DATA,
4576 &v, 1, A_TP_TX_SCHED_PCMD);
4578 #define T4_TX_MODQ_10G_WEIGHT_DEFAULT 16 /* in KB units */
4579 if (is_offload(adap)) {
4580 t4_write_reg(adap, A_TP_TX_MOD_QUEUE_WEIGHT0,
4581 V_TX_MODQ_WEIGHT0(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4582 V_TX_MODQ_WEIGHT1(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4583 V_TX_MODQ_WEIGHT2(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4584 V_TX_MODQ_WEIGHT3(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
4585 t4_write_reg(adap, A_TP_TX_MOD_CHANNEL_WEIGHT,
4586 V_TX_MODQ_WEIGHT0(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4587 V_TX_MODQ_WEIGHT1(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4588 V_TX_MODQ_WEIGHT2(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
4589 V_TX_MODQ_WEIGHT3(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
4592 /* get basic stuff going */
4593 return t4_early_init(adap, adap->fn);
4597 * Max # of ATIDs. The absolute HW max is 16K but we keep it lower.
4599 #define MAX_ATIDS 8192U
4602 * Phase 0 of initialization: contact FW, obtain config, perform basic init.
4604 * If the firmware we're dealing with has Configuration File support, then
4605 * we use that to perform all configuration
4609 * Tweak configuration based on module parameters, etc. Most of these have
4610 * defaults assigned to them by Firmware Configuration Files (if we're using
4611 * them) but need to be explicitly set if we're using hard-coded
4612 * initialization. But even in the case of using Firmware Configuration
4613 * Files, we'd like to expose the ability to change these via module
4614 * parameters so these are essentially common tweaks/settings for
4615 * Configuration Files and hard-coded initialization ...
4617 static int adap_init0_tweaks(struct adapter *adapter)
4620 * Fix up various Host-Dependent Parameters like Page Size, Cache
4621 * Line Size, etc. The firmware default is for a 4KB Page Size and
4622 * 64B Cache Line Size ...
4624 t4_fixup_host_params(adapter, PAGE_SIZE, L1_CACHE_BYTES);
4627 * Process module parameters which affect early initialization.
4629 if (rx_dma_offset != 2 && rx_dma_offset != 0) {
4630 dev_err(&adapter->pdev->dev,
4631 "Ignoring illegal rx_dma_offset=%d, using 2\n",
4635 t4_set_reg_field(adapter, SGE_CONTROL,
4637 PKTSHIFT(rx_dma_offset));
4640 * Don't include the "IP Pseudo Header" in CPL_RX_PKT checksums: Linux
4641 * adds the pseudo header itself.
4643 t4_tp_wr_bits_indirect(adapter, TP_INGRESS_CONFIG,
4644 CSUM_HAS_PSEUDO_HDR, 0);
4650 * Attempt to initialize the adapter via a Firmware Configuration File.
4652 static int adap_init0_config(struct adapter *adapter, int reset)
4654 struct fw_caps_config_cmd caps_cmd;
4655 const struct firmware *cf;
4656 unsigned long mtype = 0, maddr = 0;
4657 u32 finiver, finicsum, cfcsum;
4659 int config_issued = 0;
4660 char *fw_config_file, fw_config_file_path[256];
4661 char *config_name = NULL;
4664 * Reset device if necessary.
4667 ret = t4_fw_reset(adapter, adapter->mbox,
4668 PIORSTMODE | PIORST);
4674 * If we have a T4 configuration file under /lib/firmware/cxgb4/,
4675 * then use that. Otherwise, use the configuration file stored
4676 * in the adapter flash ...
4678 switch (CHELSIO_CHIP_VERSION(adapter->params.chip)) {
4680 fw_config_file = FW4_CFNAME;
4683 fw_config_file = FW5_CFNAME;
4686 dev_err(adapter->pdev_dev, "Device %d is not supported\n",
4687 adapter->pdev->device);
4692 ret = request_firmware(&cf, fw_config_file, adapter->pdev_dev);
4694 config_name = "On FLASH";
4695 mtype = FW_MEMTYPE_CF_FLASH;
4696 maddr = t4_flash_cfg_addr(adapter);
4698 u32 params[7], val[7];
4700 sprintf(fw_config_file_path,
4701 "/lib/firmware/%s", fw_config_file);
4702 config_name = fw_config_file_path;
4704 if (cf->size >= FLASH_CFG_MAX_SIZE)
4707 params[0] = (FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
4708 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_CF));
4709 ret = t4_query_params(adapter, adapter->mbox,
4710 adapter->fn, 0, 1, params, val);
4713 * For t4_memory_write() below addresses and
4714 * sizes have to be in terms of multiples of 4
4715 * bytes. So, if the Configuration File isn't
4716 * a multiple of 4 bytes in length we'll have
4717 * to write that out separately since we can't
4718 * guarantee that the bytes following the
4719 * residual byte in the buffer returned by
4720 * request_firmware() are zeroed out ...
4722 size_t resid = cf->size & 0x3;
4723 size_t size = cf->size & ~0x3;
4724 __be32 *data = (__be32 *)cf->data;
4726 mtype = FW_PARAMS_PARAM_Y_GET(val[0]);
4727 maddr = FW_PARAMS_PARAM_Z_GET(val[0]) << 16;
4729 ret = t4_memory_write(adapter, mtype, maddr,
4731 if (ret == 0 && resid != 0) {
4738 last.word = data[size >> 2];
4739 for (i = resid; i < 4; i++)
4741 ret = t4_memory_write(adapter, mtype,
4748 release_firmware(cf);
4754 * Issue a Capability Configuration command to the firmware to get it
4755 * to parse the Configuration File. We don't use t4_fw_config_file()
4756 * because we want the ability to modify various features after we've
4757 * processed the configuration file ...
4759 memset(&caps_cmd, 0, sizeof(caps_cmd));
4760 caps_cmd.op_to_write =
4761 htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4764 caps_cmd.cfvalid_to_len16 =
4765 htonl(FW_CAPS_CONFIG_CMD_CFVALID |
4766 FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
4767 FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(maddr >> 16) |
4768 FW_LEN16(caps_cmd));
4769 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
4772 /* If the CAPS_CONFIG failed with an ENOENT (for a Firmware
4773 * Configuration File in FLASH), our last gasp effort is to use the
4774 * Firmware Configuration File which is embedded in the firmware. A
4775 * very few early versions of the firmware didn't have one embedded
4776 * but we can ignore those.
4778 if (ret == -ENOENT) {
4779 memset(&caps_cmd, 0, sizeof(caps_cmd));
4780 caps_cmd.op_to_write =
4781 htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4784 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
4785 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd,
4786 sizeof(caps_cmd), &caps_cmd);
4787 config_name = "Firmware Default";
4794 finiver = ntohl(caps_cmd.finiver);
4795 finicsum = ntohl(caps_cmd.finicsum);
4796 cfcsum = ntohl(caps_cmd.cfcsum);
4797 if (finicsum != cfcsum)
4798 dev_warn(adapter->pdev_dev, "Configuration File checksum "\
4799 "mismatch: [fini] csum=%#x, computed csum=%#x\n",
4803 * And now tell the firmware to use the configuration we just loaded.
4805 caps_cmd.op_to_write =
4806 htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4809 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
4810 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
4816 * Tweak configuration based on system architecture, module
4819 ret = adap_init0_tweaks(adapter);
4824 * And finally tell the firmware to initialize itself using the
4825 * parameters from the Configuration File.
4827 ret = t4_fw_initialize(adapter, adapter->mbox);
4832 * Return successfully and note that we're operating with parameters
4833 * not supplied by the driver, rather than from hard-wired
4834 * initialization constants burried in the driver.
4836 adapter->flags |= USING_SOFT_PARAMS;
4837 dev_info(adapter->pdev_dev, "Successfully configured using Firmware "\
4838 "Configuration File \"%s\", version %#x, computed checksum %#x\n",
4839 config_name, finiver, cfcsum);
4843 * Something bad happened. Return the error ... (If the "error"
4844 * is that there's no Configuration File on the adapter we don't
4845 * want to issue a warning since this is fairly common.)
4848 if (config_issued && ret != -ENOENT)
4849 dev_warn(adapter->pdev_dev, "\"%s\" configuration file error %d\n",
4855 * Attempt to initialize the adapter via hard-coded, driver supplied
4858 static int adap_init0_no_config(struct adapter *adapter, int reset)
4860 struct sge *s = &adapter->sge;
4861 struct fw_caps_config_cmd caps_cmd;
4866 * Reset device if necessary
4869 ret = t4_fw_reset(adapter, adapter->mbox,
4870 PIORSTMODE | PIORST);
4876 * Get device capabilities and select which we'll be using.
4878 memset(&caps_cmd, 0, sizeof(caps_cmd));
4879 caps_cmd.op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4880 FW_CMD_REQUEST | FW_CMD_READ);
4881 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
4882 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
4887 if (caps_cmd.niccaps & htons(FW_CAPS_CONFIG_NIC_VM)) {
4889 caps_cmd.niccaps ^= htons(FW_CAPS_CONFIG_NIC_VM);
4891 caps_cmd.niccaps = htons(FW_CAPS_CONFIG_NIC_VM);
4892 } else if (vf_acls) {
4893 dev_err(adapter->pdev_dev, "virtualization ACLs not supported");
4896 caps_cmd.op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4897 FW_CMD_REQUEST | FW_CMD_WRITE);
4898 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
4904 * Tweak configuration based on system architecture, module
4907 ret = adap_init0_tweaks(adapter);
4912 * Select RSS Global Mode we want to use. We use "Basic Virtual"
4913 * mode which maps each Virtual Interface to its own section of
4914 * the RSS Table and we turn on all map and hash enables ...
4916 adapter->flags |= RSS_TNLALLLOOKUP;
4917 ret = t4_config_glbl_rss(adapter, adapter->mbox,
4918 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL,
4919 FW_RSS_GLB_CONFIG_CMD_TNLMAPEN |
4920 FW_RSS_GLB_CONFIG_CMD_HASHTOEPLITZ |
4921 ((adapter->flags & RSS_TNLALLLOOKUP) ?
4922 FW_RSS_GLB_CONFIG_CMD_TNLALLLKP : 0));
4927 * Set up our own fundamental resource provisioning ...
4929 ret = t4_cfg_pfvf(adapter, adapter->mbox, adapter->fn, 0,
4930 PFRES_NEQ, PFRES_NETHCTRL,
4931 PFRES_NIQFLINT, PFRES_NIQ,
4932 PFRES_TC, PFRES_NVI,
4933 FW_PFVF_CMD_CMASK_MASK,
4934 pfvfres_pmask(adapter, adapter->fn, 0),
4936 PFRES_R_CAPS, PFRES_WX_CAPS);
4941 * Perform low level SGE initialization. We need to do this before we
4942 * send the firmware the INITIALIZE command because that will cause
4943 * any other PF Drivers which are waiting for the Master
4944 * Initialization to proceed forward.
4946 for (i = 0; i < SGE_NTIMERS - 1; i++)
4947 s->timer_val[i] = min(intr_holdoff[i], MAX_SGE_TIMERVAL);
4948 s->timer_val[SGE_NTIMERS - 1] = MAX_SGE_TIMERVAL;
4949 s->counter_val[0] = 1;
4950 for (i = 1; i < SGE_NCOUNTERS; i++)
4951 s->counter_val[i] = min(intr_cnt[i - 1],
4952 THRESHOLD_0_GET(THRESHOLD_0_MASK));
4953 t4_sge_init(adapter);
4955 #ifdef CONFIG_PCI_IOV
4957 * Provision resource limits for Virtual Functions. We currently
4958 * grant them all the same static resource limits except for the Port
4959 * Access Rights Mask which we're assigning based on the PF. All of
4960 * the static provisioning stuff for both the PF and VF really needs
4961 * to be managed in a persistent manner for each device which the
4962 * firmware controls.
4967 for (pf = 0; pf < ARRAY_SIZE(num_vf); pf++) {
4968 if (num_vf[pf] <= 0)
4971 /* VF numbering starts at 1! */
4972 for (vf = 1; vf <= num_vf[pf]; vf++) {
4973 ret = t4_cfg_pfvf(adapter, adapter->mbox,
4975 VFRES_NEQ, VFRES_NETHCTRL,
4976 VFRES_NIQFLINT, VFRES_NIQ,
4977 VFRES_TC, VFRES_NVI,
4978 FW_PFVF_CMD_CMASK_MASK,
4982 VFRES_R_CAPS, VFRES_WX_CAPS);
4984 dev_warn(adapter->pdev_dev,
4986 "provision pf/vf=%d/%d; "
4987 "err=%d\n", pf, vf, ret);
4994 * Set up the default filter mode. Later we'll want to implement this
4995 * via a firmware command, etc. ... This needs to be done before the
4996 * firmare initialization command ... If the selected set of fields
4997 * isn't equal to the default value, we'll need to make sure that the
4998 * field selections will fit in the 36-bit budget.
5000 if (tp_vlan_pri_map != TP_VLAN_PRI_MAP_DEFAULT) {
5003 for (j = TP_VLAN_PRI_MAP_FIRST; j <= TP_VLAN_PRI_MAP_LAST; j++)
5004 switch (tp_vlan_pri_map & (1 << j)) {
5006 /* compressed filter field not enabled */
5026 case ETHERTYPE_MASK:
5032 case MPSHITTYPE_MASK:
5035 case FRAGMENTATION_MASK:
5041 dev_err(adapter->pdev_dev,
5042 "tp_vlan_pri_map=%#x needs %d bits > 36;"\
5043 " using %#x\n", tp_vlan_pri_map, bits,
5044 TP_VLAN_PRI_MAP_DEFAULT);
5045 tp_vlan_pri_map = TP_VLAN_PRI_MAP_DEFAULT;
5048 v = tp_vlan_pri_map;
5049 t4_write_indirect(adapter, TP_PIO_ADDR, TP_PIO_DATA,
5050 &v, 1, TP_VLAN_PRI_MAP);
5053 * We need Five Tuple Lookup mode to be set in TP_GLOBAL_CONFIG order
5054 * to support any of the compressed filter fields above. Newer
5055 * versions of the firmware do this automatically but it doesn't hurt
5056 * to set it here. Meanwhile, we do _not_ need to set Lookup Every
5057 * Packet in TP_INGRESS_CONFIG to support matching non-TCP packets
5058 * since the firmware automatically turns this on and off when we have
5059 * a non-zero number of filters active (since it does have a
5060 * performance impact).
5062 if (tp_vlan_pri_map)
5063 t4_set_reg_field(adapter, TP_GLOBAL_CONFIG,
5064 FIVETUPLELOOKUP_MASK,
5065 FIVETUPLELOOKUP_MASK);
5068 * Tweak some settings.
5070 t4_write_reg(adapter, TP_SHIFT_CNT, SYNSHIFTMAX(6) |
5071 RXTSHIFTMAXR1(4) | RXTSHIFTMAXR2(15) |
5072 PERSHIFTBACKOFFMAX(8) | PERSHIFTMAX(8) |
5073 KEEPALIVEMAXR1(4) | KEEPALIVEMAXR2(9));
5076 * Get basic stuff going by issuing the Firmware Initialize command.
5077 * Note that this _must_ be after all PFVF commands ...
5079 ret = t4_fw_initialize(adapter, adapter->mbox);
5084 * Return successfully!
5086 dev_info(adapter->pdev_dev, "Successfully configured using built-in "\
5087 "driver parameters\n");
5091 * Something bad happened. Return the error ...
5097 static struct fw_info fw_info_array[] = {
5100 .fs_name = FW4_CFNAME,
5101 .fw_mod_name = FW4_FNAME,
5103 .chip = FW_HDR_CHIP_T4,
5104 .fw_ver = __cpu_to_be32(FW_VERSION(T4)),
5105 .intfver_nic = FW_INTFVER(T4, NIC),
5106 .intfver_vnic = FW_INTFVER(T4, VNIC),
5107 .intfver_ri = FW_INTFVER(T4, RI),
5108 .intfver_iscsi = FW_INTFVER(T4, ISCSI),
5109 .intfver_fcoe = FW_INTFVER(T4, FCOE),
5113 .fs_name = FW5_CFNAME,
5114 .fw_mod_name = FW5_FNAME,
5116 .chip = FW_HDR_CHIP_T5,
5117 .fw_ver = __cpu_to_be32(FW_VERSION(T5)),
5118 .intfver_nic = FW_INTFVER(T5, NIC),
5119 .intfver_vnic = FW_INTFVER(T5, VNIC),
5120 .intfver_ri = FW_INTFVER(T5, RI),
5121 .intfver_iscsi = FW_INTFVER(T5, ISCSI),
5122 .intfver_fcoe = FW_INTFVER(T5, FCOE),
5127 static struct fw_info *find_fw_info(int chip)
5131 for (i = 0; i < ARRAY_SIZE(fw_info_array); i++) {
5132 if (fw_info_array[i].chip == chip)
5133 return &fw_info_array[i];
5139 * Phase 0 of initialization: contact FW, obtain config, perform basic init.
5141 static int adap_init0(struct adapter *adap)
5145 enum dev_state state;
5146 u32 params[7], val[7];
5147 struct fw_caps_config_cmd caps_cmd;
5151 * Contact FW, advertising Master capability (and potentially forcing
5152 * ourselves as the Master PF if our module parameter force_init is
5155 ret = t4_fw_hello(adap, adap->mbox, adap->fn,
5156 force_init ? MASTER_MUST : MASTER_MAY,
5159 dev_err(adap->pdev_dev, "could not connect to FW, error %d\n",
5163 if (ret == adap->mbox)
5164 adap->flags |= MASTER_PF;
5165 if (force_init && state == DEV_STATE_INIT)
5166 state = DEV_STATE_UNINIT;
5169 * If we're the Master PF Driver and the device is uninitialized,
5170 * then let's consider upgrading the firmware ... (We always want
5171 * to check the firmware version number in order to A. get it for
5172 * later reporting and B. to warn if the currently loaded firmware
5173 * is excessively mismatched relative to the driver.)
5175 t4_get_fw_version(adap, &adap->params.fw_vers);
5176 t4_get_tp_version(adap, &adap->params.tp_vers);
5177 if ((adap->flags & MASTER_PF) && state != DEV_STATE_INIT) {
5178 struct fw_info *fw_info;
5179 struct fw_hdr *card_fw;
5180 const struct firmware *fw;
5181 const u8 *fw_data = NULL;
5182 unsigned int fw_size = 0;
5184 /* This is the firmware whose headers the driver was compiled
5187 fw_info = find_fw_info(CHELSIO_CHIP_VERSION(adap->params.chip));
5188 if (fw_info == NULL) {
5189 dev_err(adap->pdev_dev,
5190 "unable to get firmware info for chip %d.\n",
5191 CHELSIO_CHIP_VERSION(adap->params.chip));
5195 /* allocate memory to read the header of the firmware on the
5198 card_fw = t4_alloc_mem(sizeof(*card_fw));
5200 /* Get FW from from /lib/firmware/ */
5201 ret = request_firmware(&fw, fw_info->fw_mod_name,
5204 dev_err(adap->pdev_dev,
5205 "unable to load firmware image %s, error %d\n",
5206 fw_info->fw_mod_name, ret);
5212 /* upgrade FW logic */
5213 ret = t4_prep_fw(adap, fw_info, fw_data, fw_size, card_fw,
5218 release_firmware(fw);
5219 t4_free_mem(card_fw);
5226 * Grab VPD parameters. This should be done after we establish a
5227 * connection to the firmware since some of the VPD parameters
5228 * (notably the Core Clock frequency) are retrieved via requests to
5229 * the firmware. On the other hand, we need these fairly early on
5230 * so we do this right after getting ahold of the firmware.
5232 ret = get_vpd_params(adap, &adap->params.vpd);
5237 * Find out what ports are available to us. Note that we need to do
5238 * this before calling adap_init0_no_config() since it needs nports
5242 FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
5243 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_PORTVEC);
5244 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 1, &v, &port_vec);
5248 adap->params.nports = hweight32(port_vec);
5249 adap->params.portvec = port_vec;
5252 * If the firmware is initialized already (and we're not forcing a
5253 * master initialization), note that we're living with existing
5254 * adapter parameters. Otherwise, it's time to try initializing the
5257 if (state == DEV_STATE_INIT) {
5258 dev_info(adap->pdev_dev, "Coming up as %s: "\
5259 "Adapter already initialized\n",
5260 adap->flags & MASTER_PF ? "MASTER" : "SLAVE");
5261 adap->flags |= USING_SOFT_PARAMS;
5263 dev_info(adap->pdev_dev, "Coming up as MASTER: "\
5264 "Initializing adapter\n");
5267 * If the firmware doesn't support Configuration
5268 * Files warn user and exit,
5271 dev_warn(adap->pdev_dev, "Firmware doesn't support "
5272 "configuration file.\n");
5274 ret = adap_init0_no_config(adap, reset);
5277 * Find out whether we're dealing with a version of
5278 * the firmware which has configuration file support.
5280 params[0] = (FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
5281 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_CF));
5282 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 1,
5286 * If the firmware doesn't support Configuration
5287 * Files, use the old Driver-based, hard-wired
5288 * initialization. Otherwise, try using the
5289 * Configuration File support and fall back to the
5290 * Driver-based initialization if there's no
5291 * Configuration File found.
5294 ret = adap_init0_no_config(adap, reset);
5297 * The firmware provides us with a memory
5298 * buffer where we can load a Configuration
5299 * File from the host if we want to override
5300 * the Configuration File in flash.
5303 ret = adap_init0_config(adap, reset);
5304 if (ret == -ENOENT) {
5305 dev_info(adap->pdev_dev,
5306 "No Configuration File present "
5307 "on adapter. Using hard-wired "
5308 "configuration parameters.\n");
5309 ret = adap_init0_no_config(adap, reset);
5314 dev_err(adap->pdev_dev,
5315 "could not initialize adapter, error %d\n",
5322 * If we're living with non-hard-coded parameters (either from a
5323 * Firmware Configuration File or values programmed by a different PF
5324 * Driver), give the SGE code a chance to pull in anything that it
5325 * needs ... Note that this must be called after we retrieve our VPD
5326 * parameters in order to know how to convert core ticks to seconds.
5328 if (adap->flags & USING_SOFT_PARAMS) {
5329 ret = t4_sge_init(adap);
5334 if (is_bypass_device(adap->pdev->device))
5335 adap->params.bypass = 1;
5338 * Grab some of our basic fundamental operating parameters.
5340 #define FW_PARAM_DEV(param) \
5341 (FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
5342 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
5344 #define FW_PARAM_PFVF(param) \
5345 FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
5346 FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param)| \
5347 FW_PARAMS_PARAM_Y(0) | \
5348 FW_PARAMS_PARAM_Z(0)
5350 params[0] = FW_PARAM_PFVF(EQ_START);
5351 params[1] = FW_PARAM_PFVF(L2T_START);
5352 params[2] = FW_PARAM_PFVF(L2T_END);
5353 params[3] = FW_PARAM_PFVF(FILTER_START);
5354 params[4] = FW_PARAM_PFVF(FILTER_END);
5355 params[5] = FW_PARAM_PFVF(IQFLINT_START);
5356 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 6, params, val);
5359 adap->sge.egr_start = val[0];
5360 adap->l2t_start = val[1];
5361 adap->l2t_end = val[2];
5362 adap->tids.ftid_base = val[3];
5363 adap->tids.nftids = val[4] - val[3] + 1;
5364 adap->sge.ingr_start = val[5];
5366 /* query params related to active filter region */
5367 params[0] = FW_PARAM_PFVF(ACTIVE_FILTER_START);
5368 params[1] = FW_PARAM_PFVF(ACTIVE_FILTER_END);
5369 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 2, params, val);
5370 /* If Active filter size is set we enable establishing
5371 * offload connection through firmware work request
5373 if ((val[0] != val[1]) && (ret >= 0)) {
5374 adap->flags |= FW_OFLD_CONN;
5375 adap->tids.aftid_base = val[0];
5376 adap->tids.aftid_end = val[1];
5379 /* If we're running on newer firmware, let it know that we're
5380 * prepared to deal with encapsulated CPL messages. Older
5381 * firmware won't understand this and we'll just get
5382 * unencapsulated messages ...
5384 params[0] = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
5386 (void) t4_set_params(adap, adap->mbox, adap->fn, 0, 1, params, val);
5389 * Find out whether we're allowed to use the T5+ ULPTX MEMWRITE DSGL
5390 * capability. Earlier versions of the firmware didn't have the
5391 * ULPTX_MEMWRITE_DSGL so we'll interpret a query failure as no
5392 * permission to use ULPTX MEMWRITE DSGL.
5394 if (is_t4(adap->params.chip)) {
5395 adap->params.ulptx_memwrite_dsgl = false;
5397 params[0] = FW_PARAM_DEV(ULPTX_MEMWRITE_DSGL);
5398 ret = t4_query_params(adap, adap->mbox, adap->fn, 0,
5400 adap->params.ulptx_memwrite_dsgl = (ret == 0 && val[0] != 0);
5404 * Get device capabilities so we can determine what resources we need
5407 memset(&caps_cmd, 0, sizeof(caps_cmd));
5408 caps_cmd.op_to_write = htonl(FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
5409 FW_CMD_REQUEST | FW_CMD_READ);
5410 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
5411 ret = t4_wr_mbox(adap, adap->mbox, &caps_cmd, sizeof(caps_cmd),
5416 if (caps_cmd.ofldcaps) {
5417 /* query offload-related parameters */
5418 params[0] = FW_PARAM_DEV(NTID);
5419 params[1] = FW_PARAM_PFVF(SERVER_START);
5420 params[2] = FW_PARAM_PFVF(SERVER_END);
5421 params[3] = FW_PARAM_PFVF(TDDP_START);
5422 params[4] = FW_PARAM_PFVF(TDDP_END);
5423 params[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
5424 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 6,
5428 adap->tids.ntids = val[0];
5429 adap->tids.natids = min(adap->tids.ntids / 2, MAX_ATIDS);
5430 adap->tids.stid_base = val[1];
5431 adap->tids.nstids = val[2] - val[1] + 1;
5433 * Setup server filter region. Divide the availble filter
5434 * region into two parts. Regular filters get 1/3rd and server
5435 * filters get 2/3rd part. This is only enabled if workarond
5437 * 1. For regular filters.
5438 * 2. Server filter: This are special filters which are used
5439 * to redirect SYN packets to offload queue.
5441 if (adap->flags & FW_OFLD_CONN && !is_bypass(adap)) {
5442 adap->tids.sftid_base = adap->tids.ftid_base +
5443 DIV_ROUND_UP(adap->tids.nftids, 3);
5444 adap->tids.nsftids = adap->tids.nftids -
5445 DIV_ROUND_UP(adap->tids.nftids, 3);
5446 adap->tids.nftids = adap->tids.sftid_base -
5447 adap->tids.ftid_base;
5449 adap->vres.ddp.start = val[3];
5450 adap->vres.ddp.size = val[4] - val[3] + 1;
5451 adap->params.ofldq_wr_cred = val[5];
5453 adap->params.offload = 1;
5455 if (caps_cmd.rdmacaps) {
5456 params[0] = FW_PARAM_PFVF(STAG_START);
5457 params[1] = FW_PARAM_PFVF(STAG_END);
5458 params[2] = FW_PARAM_PFVF(RQ_START);
5459 params[3] = FW_PARAM_PFVF(RQ_END);
5460 params[4] = FW_PARAM_PFVF(PBL_START);
5461 params[5] = FW_PARAM_PFVF(PBL_END);
5462 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 6,
5466 adap->vres.stag.start = val[0];
5467 adap->vres.stag.size = val[1] - val[0] + 1;
5468 adap->vres.rq.start = val[2];
5469 adap->vres.rq.size = val[3] - val[2] + 1;
5470 adap->vres.pbl.start = val[4];
5471 adap->vres.pbl.size = val[5] - val[4] + 1;
5473 params[0] = FW_PARAM_PFVF(SQRQ_START);
5474 params[1] = FW_PARAM_PFVF(SQRQ_END);
5475 params[2] = FW_PARAM_PFVF(CQ_START);
5476 params[3] = FW_PARAM_PFVF(CQ_END);
5477 params[4] = FW_PARAM_PFVF(OCQ_START);
5478 params[5] = FW_PARAM_PFVF(OCQ_END);
5479 ret = t4_query_params(adap, 0, 0, 0, 6, params, val);
5482 adap->vres.qp.start = val[0];
5483 adap->vres.qp.size = val[1] - val[0] + 1;
5484 adap->vres.cq.start = val[2];
5485 adap->vres.cq.size = val[3] - val[2] + 1;
5486 adap->vres.ocq.start = val[4];
5487 adap->vres.ocq.size = val[5] - val[4] + 1;
5489 if (caps_cmd.iscsicaps) {
5490 params[0] = FW_PARAM_PFVF(ISCSI_START);
5491 params[1] = FW_PARAM_PFVF(ISCSI_END);
5492 ret = t4_query_params(adap, adap->mbox, adap->fn, 0, 2,
5496 adap->vres.iscsi.start = val[0];
5497 adap->vres.iscsi.size = val[1] - val[0] + 1;
5499 #undef FW_PARAM_PFVF
5503 * These are finalized by FW initialization, load their values now.
5505 t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
5506 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
5507 adap->params.b_wnd);
5509 t4_init_tp_params(adap);
5510 adap->flags |= FW_OK;
5514 * Something bad happened. If a command timed out or failed with EIO
5515 * FW does not operate within its spec or something catastrophic
5516 * happened to HW/FW, stop issuing commands.
5519 if (ret != -ETIMEDOUT && ret != -EIO)
5520 t4_fw_bye(adap, adap->mbox);
5526 static pci_ers_result_t eeh_err_detected(struct pci_dev *pdev,
5527 pci_channel_state_t state)
5530 struct adapter *adap = pci_get_drvdata(pdev);
5536 adap->flags &= ~FW_OK;
5537 notify_ulds(adap, CXGB4_STATE_START_RECOVERY);
5538 spin_lock(&adap->stats_lock);
5539 for_each_port(adap, i) {
5540 struct net_device *dev = adap->port[i];
5542 netif_device_detach(dev);
5543 netif_carrier_off(dev);
5545 spin_unlock(&adap->stats_lock);
5546 if (adap->flags & FULL_INIT_DONE)
5549 if ((adap->flags & DEV_ENABLED)) {
5550 pci_disable_device(pdev);
5551 adap->flags &= ~DEV_ENABLED;
5553 out: return state == pci_channel_io_perm_failure ?
5554 PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
5557 static pci_ers_result_t eeh_slot_reset(struct pci_dev *pdev)
5560 struct fw_caps_config_cmd c;
5561 struct adapter *adap = pci_get_drvdata(pdev);
5564 pci_restore_state(pdev);
5565 pci_save_state(pdev);
5566 return PCI_ERS_RESULT_RECOVERED;
5569 if (!(adap->flags & DEV_ENABLED)) {
5570 if (pci_enable_device(pdev)) {
5571 dev_err(&pdev->dev, "Cannot reenable PCI "
5572 "device after reset\n");
5573 return PCI_ERS_RESULT_DISCONNECT;
5575 adap->flags |= DEV_ENABLED;
5578 pci_set_master(pdev);
5579 pci_restore_state(pdev);
5580 pci_save_state(pdev);
5581 pci_cleanup_aer_uncorrect_error_status(pdev);
5583 if (t4_wait_dev_ready(adap) < 0)
5584 return PCI_ERS_RESULT_DISCONNECT;
5585 if (t4_fw_hello(adap, adap->fn, adap->fn, MASTER_MUST, NULL) < 0)
5586 return PCI_ERS_RESULT_DISCONNECT;
5587 adap->flags |= FW_OK;
5588 if (adap_init1(adap, &c))
5589 return PCI_ERS_RESULT_DISCONNECT;
5591 for_each_port(adap, i) {
5592 struct port_info *p = adap2pinfo(adap, i);
5594 ret = t4_alloc_vi(adap, adap->fn, p->tx_chan, adap->fn, 0, 1,
5597 return PCI_ERS_RESULT_DISCONNECT;
5599 p->xact_addr_filt = -1;
5602 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
5603 adap->params.b_wnd);
5606 return PCI_ERS_RESULT_DISCONNECT;
5607 return PCI_ERS_RESULT_RECOVERED;
5610 static void eeh_resume(struct pci_dev *pdev)
5613 struct adapter *adap = pci_get_drvdata(pdev);
5619 for_each_port(adap, i) {
5620 struct net_device *dev = adap->port[i];
5622 if (netif_running(dev)) {
5624 cxgb_set_rxmode(dev);
5626 netif_device_attach(dev);
5631 static const struct pci_error_handlers cxgb4_eeh = {
5632 .error_detected = eeh_err_detected,
5633 .slot_reset = eeh_slot_reset,
5634 .resume = eeh_resume,
5637 static inline bool is_x_10g_port(const struct link_config *lc)
5639 return (lc->supported & FW_PORT_CAP_SPEED_10G) != 0 ||
5640 (lc->supported & FW_PORT_CAP_SPEED_40G) != 0;
5643 static inline void init_rspq(struct sge_rspq *q, u8 timer_idx, u8 pkt_cnt_idx,
5644 unsigned int size, unsigned int iqe_size)
5646 q->intr_params = QINTR_TIMER_IDX(timer_idx) |
5647 (pkt_cnt_idx < SGE_NCOUNTERS ? QINTR_CNT_EN : 0);
5648 q->pktcnt_idx = pkt_cnt_idx < SGE_NCOUNTERS ? pkt_cnt_idx : 0;
5649 q->iqe_len = iqe_size;
5654 * Perform default configuration of DMA queues depending on the number and type
5655 * of ports we found and the number of available CPUs. Most settings can be
5656 * modified by the admin prior to actual use.
5658 static void cfg_queues(struct adapter *adap)
5660 struct sge *s = &adap->sge;
5661 int i, q10g = 0, n10g = 0, qidx = 0;
5663 for_each_port(adap, i)
5664 n10g += is_x_10g_port(&adap2pinfo(adap, i)->link_cfg);
5667 * We default to 1 queue per non-10G port and up to # of cores queues
5671 q10g = (MAX_ETH_QSETS - (adap->params.nports - n10g)) / n10g;
5672 if (q10g > netif_get_num_default_rss_queues())
5673 q10g = netif_get_num_default_rss_queues();
5675 for_each_port(adap, i) {
5676 struct port_info *pi = adap2pinfo(adap, i);
5678 pi->first_qset = qidx;
5679 pi->nqsets = is_x_10g_port(&pi->link_cfg) ? q10g : 1;
5684 s->max_ethqsets = qidx; /* MSI-X may lower it later */
5686 if (is_offload(adap)) {
5688 * For offload we use 1 queue/channel if all ports are up to 1G,
5689 * otherwise we divide all available queues amongst the channels
5690 * capped by the number of available cores.
5693 i = min_t(int, ARRAY_SIZE(s->ofldrxq),
5695 s->ofldqsets = roundup(i, adap->params.nports);
5697 s->ofldqsets = adap->params.nports;
5698 /* For RDMA one Rx queue per channel suffices */
5699 s->rdmaqs = adap->params.nports;
5702 for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
5703 struct sge_eth_rxq *r = &s->ethrxq[i];
5705 init_rspq(&r->rspq, 0, 0, 1024, 64);
5709 for (i = 0; i < ARRAY_SIZE(s->ethtxq); i++)
5710 s->ethtxq[i].q.size = 1024;
5712 for (i = 0; i < ARRAY_SIZE(s->ctrlq); i++)
5713 s->ctrlq[i].q.size = 512;
5715 for (i = 0; i < ARRAY_SIZE(s->ofldtxq); i++)
5716 s->ofldtxq[i].q.size = 1024;
5718 for (i = 0; i < ARRAY_SIZE(s->ofldrxq); i++) {
5719 struct sge_ofld_rxq *r = &s->ofldrxq[i];
5721 init_rspq(&r->rspq, 0, 0, 1024, 64);
5722 r->rspq.uld = CXGB4_ULD_ISCSI;
5726 for (i = 0; i < ARRAY_SIZE(s->rdmarxq); i++) {
5727 struct sge_ofld_rxq *r = &s->rdmarxq[i];
5729 init_rspq(&r->rspq, 0, 0, 511, 64);
5730 r->rspq.uld = CXGB4_ULD_RDMA;
5734 init_rspq(&s->fw_evtq, 6, 0, 512, 64);
5735 init_rspq(&s->intrq, 6, 0, 2 * MAX_INGQ, 64);
5739 * Reduce the number of Ethernet queues across all ports to at most n.
5740 * n provides at least one queue per port.
5742 static void reduce_ethqs(struct adapter *adap, int n)
5745 struct port_info *pi;
5747 while (n < adap->sge.ethqsets)
5748 for_each_port(adap, i) {
5749 pi = adap2pinfo(adap, i);
5750 if (pi->nqsets > 1) {
5752 adap->sge.ethqsets--;
5753 if (adap->sge.ethqsets <= n)
5759 for_each_port(adap, i) {
5760 pi = adap2pinfo(adap, i);
5766 /* 2 MSI-X vectors needed for the FW queue and non-data interrupts */
5767 #define EXTRA_VECS 2
5769 static int enable_msix(struct adapter *adap)
5773 struct sge *s = &adap->sge;
5774 unsigned int nchan = adap->params.nports;
5775 struct msix_entry entries[MAX_INGQ + 1];
5777 for (i = 0; i < ARRAY_SIZE(entries); ++i)
5778 entries[i].entry = i;
5780 want = s->max_ethqsets + EXTRA_VECS;
5781 if (is_offload(adap)) {
5782 want += s->rdmaqs + s->ofldqsets;
5783 /* need nchan for each possible ULD */
5784 ofld_need = 2 * nchan;
5786 need = adap->params.nports + EXTRA_VECS + ofld_need;
5788 want = pci_enable_msix_range(adap->pdev, entries, need, want);
5793 * Distribute available vectors to the various queue groups.
5794 * Every group gets its minimum requirement and NIC gets top
5795 * priority for leftovers.
5797 i = want - EXTRA_VECS - ofld_need;
5798 if (i < s->max_ethqsets) {
5799 s->max_ethqsets = i;
5800 if (i < s->ethqsets)
5801 reduce_ethqs(adap, i);
5803 if (is_offload(adap)) {
5804 i = want - EXTRA_VECS - s->max_ethqsets;
5805 i -= ofld_need - nchan;
5806 s->ofldqsets = (i / nchan) * nchan; /* round down */
5808 for (i = 0; i < want; ++i)
5809 adap->msix_info[i].vec = entries[i].vector;
5816 static int init_rss(struct adapter *adap)
5820 for_each_port(adap, i) {
5821 struct port_info *pi = adap2pinfo(adap, i);
5823 pi->rss = kcalloc(pi->rss_size, sizeof(u16), GFP_KERNEL);
5826 for (j = 0; j < pi->rss_size; j++)
5827 pi->rss[j] = ethtool_rxfh_indir_default(j, pi->nqsets);
5832 static void print_port_info(const struct net_device *dev)
5836 const char *spd = "";
5837 const struct port_info *pi = netdev_priv(dev);
5838 const struct adapter *adap = pi->adapter;
5840 if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_2_5GB)
5842 else if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_5_0GB)
5845 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_100M)
5846 bufp += sprintf(bufp, "100/");
5847 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_1G)
5848 bufp += sprintf(bufp, "1000/");
5849 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_10G)
5850 bufp += sprintf(bufp, "10G/");
5851 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_40G)
5852 bufp += sprintf(bufp, "40G/");
5855 sprintf(bufp, "BASE-%s", t4_get_port_type_description(pi->port_type));
5857 netdev_info(dev, "Chelsio %s rev %d %s %sNIC PCIe x%d%s%s\n",
5858 adap->params.vpd.id,
5859 CHELSIO_CHIP_RELEASE(adap->params.chip), buf,
5860 is_offload(adap) ? "R" : "", adap->params.pci.width, spd,
5861 (adap->flags & USING_MSIX) ? " MSI-X" :
5862 (adap->flags & USING_MSI) ? " MSI" : "");
5863 netdev_info(dev, "S/N: %s, P/N: %s\n",
5864 adap->params.vpd.sn, adap->params.vpd.pn);
5867 static void enable_pcie_relaxed_ordering(struct pci_dev *dev)
5869 pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_RELAX_EN);
5873 * Free the following resources:
5874 * - memory used for tables
5877 * - resources FW is holding for us
5879 static void free_some_resources(struct adapter *adapter)
5883 t4_free_mem(adapter->l2t);
5884 t4_free_mem(adapter->tids.tid_tab);
5885 disable_msi(adapter);
5887 for_each_port(adapter, i)
5888 if (adapter->port[i]) {
5889 kfree(adap2pinfo(adapter, i)->rss);
5890 free_netdev(adapter->port[i]);
5892 if (adapter->flags & FW_OK)
5893 t4_fw_bye(adapter, adapter->fn);
5896 #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
5897 #define VLAN_FEAT (NETIF_F_SG | NETIF_F_IP_CSUM | TSO_FLAGS | \
5898 NETIF_F_IPV6_CSUM | NETIF_F_HIGHDMA)
5899 #define SEGMENT_SIZE 128
5901 static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
5903 int func, i, err, s_qpp, qpp, num_seg;
5904 struct port_info *pi;
5905 bool highdma = false;
5906 struct adapter *adapter = NULL;
5908 printk_once(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
5910 err = pci_request_regions(pdev, KBUILD_MODNAME);
5912 /* Just info, some other driver may have claimed the device. */
5913 dev_info(&pdev->dev, "cannot obtain PCI resources\n");
5917 /* We control everything through one PF */
5918 func = PCI_FUNC(pdev->devfn);
5919 if (func != ent->driver_data) {
5920 pci_save_state(pdev); /* to restore SR-IOV later */
5924 err = pci_enable_device(pdev);
5926 dev_err(&pdev->dev, "cannot enable PCI device\n");
5927 goto out_release_regions;
5930 if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
5932 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
5934 dev_err(&pdev->dev, "unable to obtain 64-bit DMA for "
5935 "coherent allocations\n");
5936 goto out_disable_device;
5939 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
5941 dev_err(&pdev->dev, "no usable DMA configuration\n");
5942 goto out_disable_device;
5946 pci_enable_pcie_error_reporting(pdev);
5947 enable_pcie_relaxed_ordering(pdev);
5948 pci_set_master(pdev);
5949 pci_save_state(pdev);
5951 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
5954 goto out_disable_device;
5957 /* PCI device has been enabled */
5958 adapter->flags |= DEV_ENABLED;
5960 adapter->regs = pci_ioremap_bar(pdev, 0);
5961 if (!adapter->regs) {
5962 dev_err(&pdev->dev, "cannot map device registers\n");
5964 goto out_free_adapter;
5967 adapter->pdev = pdev;
5968 adapter->pdev_dev = &pdev->dev;
5969 adapter->mbox = func;
5971 adapter->msg_enable = dflt_msg_enable;
5972 memset(adapter->chan_map, 0xff, sizeof(adapter->chan_map));
5974 spin_lock_init(&adapter->stats_lock);
5975 spin_lock_init(&adapter->tid_release_lock);
5977 INIT_WORK(&adapter->tid_release_task, process_tid_release_list);
5978 INIT_WORK(&adapter->db_full_task, process_db_full);
5979 INIT_WORK(&adapter->db_drop_task, process_db_drop);
5981 err = t4_prep_adapter(adapter);
5983 goto out_unmap_bar0;
5985 if (!is_t4(adapter->params.chip)) {
5986 s_qpp = QUEUESPERPAGEPF1 * adapter->fn;
5987 qpp = 1 << QUEUESPERPAGEPF0_GET(t4_read_reg(adapter,
5988 SGE_EGRESS_QUEUES_PER_PAGE_PF) >> s_qpp);
5989 num_seg = PAGE_SIZE / SEGMENT_SIZE;
5991 /* Each segment size is 128B. Write coalescing is enabled only
5992 * when SGE_EGRESS_QUEUES_PER_PAGE_PF reg value for the
5993 * queue is less no of segments that can be accommodated in
5996 if (qpp > num_seg) {
5998 "Incorrect number of egress queues per page\n");
6000 goto out_unmap_bar0;
6002 adapter->bar2 = ioremap_wc(pci_resource_start(pdev, 2),
6003 pci_resource_len(pdev, 2));
6004 if (!adapter->bar2) {
6005 dev_err(&pdev->dev, "cannot map device bar2 region\n");
6007 goto out_unmap_bar0;
6011 setup_memwin(adapter);
6012 err = adap_init0(adapter);
6013 setup_memwin_rdma(adapter);
6017 for_each_port(adapter, i) {
6018 struct net_device *netdev;
6020 netdev = alloc_etherdev_mq(sizeof(struct port_info),
6027 SET_NETDEV_DEV(netdev, &pdev->dev);
6029 adapter->port[i] = netdev;
6030 pi = netdev_priv(netdev);
6031 pi->adapter = adapter;
6032 pi->xact_addr_filt = -1;
6034 netdev->irq = pdev->irq;
6036 netdev->hw_features = NETIF_F_SG | TSO_FLAGS |
6037 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
6038 NETIF_F_RXCSUM | NETIF_F_RXHASH |
6039 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
6041 netdev->hw_features |= NETIF_F_HIGHDMA;
6042 netdev->features |= netdev->hw_features;
6043 netdev->vlan_features = netdev->features & VLAN_FEAT;
6045 netdev->priv_flags |= IFF_UNICAST_FLT;
6047 netdev->netdev_ops = &cxgb4_netdev_ops;
6048 SET_ETHTOOL_OPS(netdev, &cxgb_ethtool_ops);
6051 pci_set_drvdata(pdev, adapter);
6053 if (adapter->flags & FW_OK) {
6054 err = t4_port_init(adapter, func, func, 0);
6060 * Configure queues and allocate tables now, they can be needed as
6061 * soon as the first register_netdev completes.
6063 cfg_queues(adapter);
6065 adapter->l2t = t4_init_l2t();
6066 if (!adapter->l2t) {
6067 /* We tolerate a lack of L2T, giving up some functionality */
6068 dev_warn(&pdev->dev, "could not allocate L2T, continuing\n");
6069 adapter->params.offload = 0;
6072 if (is_offload(adapter) && tid_init(&adapter->tids) < 0) {
6073 dev_warn(&pdev->dev, "could not allocate TID table, "
6075 adapter->params.offload = 0;
6078 /* See what interrupts we'll be using */
6079 if (msi > 1 && enable_msix(adapter) == 0)
6080 adapter->flags |= USING_MSIX;
6081 else if (msi > 0 && pci_enable_msi(pdev) == 0)
6082 adapter->flags |= USING_MSI;
6084 err = init_rss(adapter);
6089 * The card is now ready to go. If any errors occur during device
6090 * registration we do not fail the whole card but rather proceed only
6091 * with the ports we manage to register successfully. However we must
6092 * register at least one net device.
6094 for_each_port(adapter, i) {
6095 pi = adap2pinfo(adapter, i);
6096 netif_set_real_num_tx_queues(adapter->port[i], pi->nqsets);
6097 netif_set_real_num_rx_queues(adapter->port[i], pi->nqsets);
6099 err = register_netdev(adapter->port[i]);
6102 adapter->chan_map[pi->tx_chan] = i;
6103 print_port_info(adapter->port[i]);
6106 dev_err(&pdev->dev, "could not register any net devices\n");
6110 dev_warn(&pdev->dev, "only %d net devices registered\n", i);
6114 if (cxgb4_debugfs_root) {
6115 adapter->debugfs_root = debugfs_create_dir(pci_name(pdev),
6116 cxgb4_debugfs_root);
6117 setup_debugfs(adapter);
6120 /* PCIe EEH recovery on powerpc platforms needs fundamental reset */
6121 pdev->needs_freset = 1;
6123 if (is_offload(adapter))
6124 attach_ulds(adapter);
6127 #ifdef CONFIG_PCI_IOV
6128 if (func < ARRAY_SIZE(num_vf) && num_vf[func] > 0)
6129 if (pci_enable_sriov(pdev, num_vf[func]) == 0)
6130 dev_info(&pdev->dev,
6131 "instantiated %u virtual functions\n",
6137 free_some_resources(adapter);
6139 if (!is_t4(adapter->params.chip))
6140 iounmap(adapter->bar2);
6142 iounmap(adapter->regs);
6146 pci_disable_pcie_error_reporting(pdev);
6147 pci_disable_device(pdev);
6148 out_release_regions:
6149 pci_release_regions(pdev);
6153 static void remove_one(struct pci_dev *pdev)
6155 struct adapter *adapter = pci_get_drvdata(pdev);
6157 #ifdef CONFIG_PCI_IOV
6158 pci_disable_sriov(pdev);
6165 if (is_offload(adapter))
6166 detach_ulds(adapter);
6168 for_each_port(adapter, i)
6169 if (adapter->port[i]->reg_state == NETREG_REGISTERED)
6170 unregister_netdev(adapter->port[i]);
6172 if (adapter->debugfs_root)
6173 debugfs_remove_recursive(adapter->debugfs_root);
6175 /* If we allocated filters, free up state associated with any
6178 if (adapter->tids.ftid_tab) {
6179 struct filter_entry *f = &adapter->tids.ftid_tab[0];
6180 for (i = 0; i < (adapter->tids.nftids +
6181 adapter->tids.nsftids); i++, f++)
6183 clear_filter(adapter, f);
6186 if (adapter->flags & FULL_INIT_DONE)
6189 free_some_resources(adapter);
6190 iounmap(adapter->regs);
6191 if (!is_t4(adapter->params.chip))
6192 iounmap(adapter->bar2);
6193 pci_disable_pcie_error_reporting(pdev);
6194 if ((adapter->flags & DEV_ENABLED)) {
6195 pci_disable_device(pdev);
6196 adapter->flags &= ~DEV_ENABLED;
6198 pci_release_regions(pdev);
6201 pci_release_regions(pdev);
6204 static struct pci_driver cxgb4_driver = {
6205 .name = KBUILD_MODNAME,
6206 .id_table = cxgb4_pci_tbl,
6208 .remove = remove_one,
6209 .err_handler = &cxgb4_eeh,
6212 static int __init cxgb4_init_module(void)
6216 workq = create_singlethread_workqueue("cxgb4");
6220 /* Debugfs support is optional, just warn if this fails */
6221 cxgb4_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
6222 if (!cxgb4_debugfs_root)
6223 pr_warn("could not create debugfs entry, continuing\n");
6225 ret = pci_register_driver(&cxgb4_driver);
6227 debugfs_remove(cxgb4_debugfs_root);
6228 destroy_workqueue(workq);
6231 register_inet6addr_notifier(&cxgb4_inet6addr_notifier);
6236 static void __exit cxgb4_cleanup_module(void)
6238 unregister_inet6addr_notifier(&cxgb4_inet6addr_notifier);
6239 pci_unregister_driver(&cxgb4_driver);
6240 debugfs_remove(cxgb4_debugfs_root); /* NULL ok */
6241 flush_workqueue(workq);
6242 destroy_workqueue(workq);
6245 module_init(cxgb4_init_module);
6246 module_exit(cxgb4_cleanup_module);