]> git.ipfire.org Git - thirdparty/linux.git/blame - drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
projects / thirdparty / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
net: sched: make egress_dev flag part of flower offload struct
[thirdparty/linux.git] / drivers / net / ethernet / chelsio / cxgb4 / cxgb4_main.c
CommitLineData
b8ff05a9
DM		 1/*
2 * This file is part of the Chelsio T4 Ethernet driver for Linux.
3 *
b72a32da 4 * Copyright (c) 2003-2016 Chelsio Communications, Inc. All rights reserved.
b8ff05a9
DM		 5 *
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:
11 *
12 * Redistribution and use in source and binary forms, with or
13 * without modification, are permitted provided that the following
14 * conditions are met:
15 *
16 * - Redistributions of source code must retain the above
17 * copyright notice, this list of conditions and the following
18 * disclaimer.
19 *
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.
24 *
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
32 * SOFTWARE.
33 */
34
35#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
36
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>
01789349 44#include <linux/if.h>
b8ff05a9
DM		 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>
01bcca68 63#include <net/addrconf.h>
1ef8019b 64#include <net/bonding.h>
b5a02f50 65#include <net/addrconf.h>
7c0f6ba6 66#include <linux/uaccess.h>
c5a8c0f3 67#include <linux/crash_dump.h>
b8ff05a9
DM		 68
69#include "cxgb4.h"
d57fd6ca 70#include "cxgb4_filter.h"
b8ff05a9 71#include "t4_regs.h"
f612b815 72#include "t4_values.h"
b8ff05a9
DM		 73#include "t4_msg.h"
74#include "t4fw_api.h"
cd6c2f12 75#include "t4fw_version.h"
688848b1 76#include "cxgb4_dcb.h"
fd88b31a 77#include "cxgb4_debugfs.h"
b5a02f50 78#include "clip_tbl.h"
b8ff05a9 79#include "l2t.h"
b72a32da 80#include "sched.h"
d8931847 81#include "cxgb4_tc_u32.h"
a4569504 82#include "cxgb4_ptp.h"
b8ff05a9 83
812034f1
HS		 84char cxgb4_driver_name[] = KBUILD_MODNAME;
85
01bcca68
VP		 86#ifdef DRV_VERSION
87#undef DRV_VERSION
88#endif
3a7f8554 89#define DRV_VERSION "2.0.0-ko"
812034f1 90const char cxgb4_driver_version[] = DRV_VERSION;
52a5f846 91#define DRV_DESC "Chelsio T4/T5/T6 Network Driver"
b8ff05a9 92
b8ff05a9
DM		 93#define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
94 NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
95 NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
96
3fedeab1
HS		 97/* Macros needed to support the PCI Device ID Table ...
98 */
99#define CH_PCI_DEVICE_ID_TABLE_DEFINE_BEGIN \
768ffc66 100 static const struct pci_device_id cxgb4_pci_tbl[] = {
3fedeab1 101#define CH_PCI_DEVICE_ID_FUNCTION 0x4
b8ff05a9 102
3fedeab1
HS		 103/* Include PCI Device IDs for both PF4 and PF0-3 so our PCI probe() routine is
104 * called for both.
105 */
106#define CH_PCI_DEVICE_ID_FUNCTION2 0x0
107
108#define CH_PCI_ID_TABLE_ENTRY(devid) \
109 {PCI_VDEVICE(CHELSIO, (devid)), 4}
110
111#define CH_PCI_DEVICE_ID_TABLE_DEFINE_END \
112 { 0, } \
113 }
114
115#include "t4_pci_id_tbl.h"
b8ff05a9 116
16e47624 117#define FW4_FNAME "cxgb4/t4fw.bin"
0a57a536 118#define FW5_FNAME "cxgb4/t5fw.bin"
3ccc6cf7 119#define FW6_FNAME "cxgb4/t6fw.bin"
16e47624 120#define FW4_CFNAME "cxgb4/t4-config.txt"
0a57a536 121#define FW5_CFNAME "cxgb4/t5-config.txt"
3ccc6cf7 122#define FW6_CFNAME "cxgb4/t6-config.txt"
01b69614
HS		 123#define PHY_AQ1202_FIRMWARE "cxgb4/aq1202_fw.cld"
124#define PHY_BCM84834_FIRMWARE "cxgb4/bcm8483.bin"
125#define PHY_AQ1202_DEVICEID 0x4409
126#define PHY_BCM84834_DEVICEID 0x4486
b8ff05a9
DM		 127
128MODULE_DESCRIPTION(DRV_DESC);
129MODULE_AUTHOR("Chelsio Communications");
130MODULE_LICENSE("Dual BSD/GPL");
131MODULE_VERSION(DRV_VERSION);
132MODULE_DEVICE_TABLE(pci, cxgb4_pci_tbl);
16e47624 133MODULE_FIRMWARE(FW4_FNAME);
0a57a536 134MODULE_FIRMWARE(FW5_FNAME);
52a5f846 135MODULE_FIRMWARE(FW6_FNAME);
b8ff05a9 136
b8ff05a9
DM		 137/*
138 * The driver uses the best interrupt scheme available on a platform in the
139 * order MSI-X, MSI, legacy INTx interrupts. This parameter determines which
140 * of these schemes the driver may consider as follows:
141 *
142 * msi = 2: choose from among all three options
143 * msi = 1: only consider MSI and INTx interrupts
144 * msi = 0: force INTx interrupts
145 */
146static int msi = 2;
147
148module_param(msi, int, 0644);
149MODULE_PARM_DESC(msi, "whether to use INTx (0), MSI (1) or MSI-X (2)");
150
636f9d37
VP		 151/*
152 * Normally we tell the chip to deliver Ingress Packets into our DMA buffers
153 * offset by 2 bytes in order to have the IP headers line up on 4-byte
154 * boundaries. This is a requirement for many architectures which will throw
155 * a machine check fault if an attempt is made to access one of the 4-byte IP
156 * header fields on a non-4-byte boundary. And it's a major performance issue
157 * even on some architectures which allow it like some implementations of the
158 * x86 ISA. However, some architectures don't mind this and for some very
159 * edge-case performance sensitive applications (like forwarding large volumes
160 * of small packets), setting this DMA offset to 0 will decrease the number of
161 * PCI-E Bus transfers enough to measurably affect performance.
162 */
163static int rx_dma_offset = 2;
164
688848b1
AB		 165/* TX Queue select used to determine what algorithm to use for selecting TX
166 * queue. Select between the kernel provided function (select_queue=0) or user
167 * cxgb_select_queue function (select_queue=1)
168 *
169 * Default: select_queue=0
170 */
171static int select_queue;
172module_param(select_queue, int, 0644);
173MODULE_PARM_DESC(select_queue,
174 "Select between kernel provided method of selecting or driver method of selecting TX queue. Default is kernel method.");
175
b8ff05a9
DM		 176static struct dentry *cxgb4_debugfs_root;
177
94cdb8bb
HS		 178LIST_HEAD(adapter_list);
179DEFINE_MUTEX(uld_mutex);
b8ff05a9
DM		 180
181static void link_report(struct net_device *dev)
182{
183 if (!netif_carrier_ok(dev))
184 netdev_info(dev, "link down\n");
185 else {
186 static const char *fc[] = { "no", "Rx", "Tx", "Tx/Rx" };
187
85412255 188 const char *s;
b8ff05a9
DM		 189 const struct port_info *p = netdev_priv(dev);
190
191 switch (p->link_cfg.speed) {
5e78f7fd
GG		 192 case 100:
193 s = "100Mbps";
b8ff05a9 194 break;
e8b39015 195 case 1000:
5e78f7fd 196 s = "1Gbps";
b8ff05a9 197 break;
5e78f7fd
GG		 198 case 10000:
199 s = "10Gbps";
200 break;
201 case 25000:
202 s = "25Gbps";
b8ff05a9 203 break;
e8b39015 204 case 40000:
72aca4bf
KS		 205 s = "40Gbps";
206 break;
5e78f7fd
GG		 207 case 100000:
208 s = "100Gbps";
209 break;
85412255
HS		 210 default:
211 pr_info("%s: unsupported speed: %d\n",
212 dev->name, p->link_cfg.speed);
213 return;
b8ff05a9
DM		 214 }
215
216 netdev_info(dev, "link up, %s, full-duplex, %s PAUSE\n", s,
217 fc[p->link_cfg.fc]);
218 }
219}
220
688848b1
AB		 221#ifdef CONFIG_CHELSIO_T4_DCB
222/* Set up/tear down Data Center Bridging Priority mapping for a net device. */
223static void dcb_tx_queue_prio_enable(struct net_device *dev, int enable)
224{
225 struct port_info *pi = netdev_priv(dev);
226 struct adapter *adap = pi->adapter;
227 struct sge_eth_txq *txq = &adap->sge.ethtxq[pi->first_qset];
228 int i;
229
230 /* We use a simple mapping of Port TX Queue Index to DCB
231 * Priority when we're enabling DCB.
232 */
233 for (i = 0; i < pi->nqsets; i++, txq++) {
234 u32 name, value;
235 int err;
236
5167865a
HS		 237 name = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DMAQ) |
238 FW_PARAMS_PARAM_X_V(
239 FW_PARAMS_PARAM_DMAQ_EQ_DCBPRIO_ETH) |
240 FW_PARAMS_PARAM_YZ_V(txq->q.cntxt_id));
688848b1
AB		 241 value = enable ? i : 0xffffffff;
242
243 /* Since we can be called while atomic (from "interrupt
244 * level") we need to issue the Set Parameters Commannd
245 * without sleeping (timeout < 0).
246 */
b2612722 247 err = t4_set_params_timeout(adap, adap->mbox, adap->pf, 0, 1,
01b69614
HS		 248 &name, &value,
249 -FW_CMD_MAX_TIMEOUT);
688848b1
AB		 250
251 if (err)
252 dev_err(adap->pdev_dev,
253 "Can't %s DCB Priority on port %d, TX Queue %d: err=%d\n",
254 enable ? "set" : "unset", pi->port_id, i, -err);
10b00466
AB		 255 else
256 txq->dcb_prio = value;
688848b1
AB		 257 }
258}
688848b1 259
50935857 260static int cxgb4_dcb_enabled(const struct net_device *dev)
218d48e7 261{
218d48e7
HS		 262 struct port_info *pi = netdev_priv(dev);
263
264 if (!pi->dcb.enabled)
265 return 0;
266
267 return ((pi->dcb.state == CXGB4_DCB_STATE_FW_ALLSYNCED) ||
268 (pi->dcb.state == CXGB4_DCB_STATE_HOST));
218d48e7 269}
7c70c4f8 270#endif /* CONFIG_CHELSIO_T4_DCB */
218d48e7 271
b8ff05a9
DM		 272void t4_os_link_changed(struct adapter *adapter, int port_id, int link_stat)
273{
274 struct net_device *dev = adapter->port[port_id];
275
276 /* Skip changes from disabled ports. */
277 if (netif_running(dev) && link_stat != netif_carrier_ok(dev)) {
278 if (link_stat)
279 netif_carrier_on(dev);
688848b1
AB		 280 else {
281#ifdef CONFIG_CHELSIO_T4_DCB
218d48e7
HS		 282 if (cxgb4_dcb_enabled(dev)) {
283 cxgb4_dcb_state_init(dev);
284 dcb_tx_queue_prio_enable(dev, false);
285 }
688848b1 286#endif /* CONFIG_CHELSIO_T4_DCB */
b8ff05a9 287 netif_carrier_off(dev);
688848b1 288 }
b8ff05a9
DM		 289
290 link_report(dev);
291 }
292}
293
294void t4_os_portmod_changed(const struct adapter *adap, int port_id)
295{
296 static const char *mod_str[] = {
a0881cab 297 NULL, "LR", "SR", "ER", "passive DA", "active DA", "LRM"
b8ff05a9
DM		 298 };
299
300 const struct net_device *dev = adap->port[port_id];
301 const struct port_info *pi = netdev_priv(dev);
302
303 if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
304 netdev_info(dev, "port module unplugged\n");
a0881cab 305 else if (pi->mod_type < ARRAY_SIZE(mod_str))
b8ff05a9 306 netdev_info(dev, "%s module inserted\n", mod_str[pi->mod_type]);
be81a2de
HS		 307 else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
308 netdev_info(dev, "%s: unsupported port module inserted\n",
309 dev->name);
310 else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
311 netdev_info(dev, "%s: unknown port module inserted\n",
312 dev->name);
313 else if (pi->mod_type == FW_PORT_MOD_TYPE_ERROR)
314 netdev_info(dev, "%s: transceiver module error\n", dev->name);
315 else
316 netdev_info(dev, "%s: unknown module type %d inserted\n",
317 dev->name, pi->mod_type);
b8ff05a9
DM		 318}
319
fc08a01a
HS		 320int dbfifo_int_thresh = 10; /* 10 == 640 entry threshold */
321module_param(dbfifo_int_thresh, int, 0644);
322MODULE_PARM_DESC(dbfifo_int_thresh, "doorbell fifo interrupt threshold");
323
b8ff05a9 324/*
fc08a01a 325 * usecs to sleep while draining the dbfifo
b8ff05a9 326 */
fc08a01a
HS		 327static int dbfifo_drain_delay = 1000;
328module_param(dbfifo_drain_delay, int, 0644);
329MODULE_PARM_DESC(dbfifo_drain_delay,
330 "usecs to sleep while draining the dbfifo");
331
332static inline int cxgb4_set_addr_hash(struct port_info *pi)
b8ff05a9 333{
fc08a01a
HS		 334 struct adapter *adap = pi->adapter;
335 u64 vec = 0;
336 bool ucast = false;
337 struct hash_mac_addr *entry;
338
339 /* Calculate the hash vector for the updated list and program it */
340 list_for_each_entry(entry, &adap->mac_hlist, list) {
341 ucast |= is_unicast_ether_addr(entry->addr);
342 vec |= (1ULL << hash_mac_addr(entry->addr));
343 }
344 return t4_set_addr_hash(adap, adap->mbox, pi->viid, ucast,
345 vec, false);
346}
347
348static int cxgb4_mac_sync(struct net_device *netdev, const u8 *mac_addr)
349{
350 struct port_info *pi = netdev_priv(netdev);
351 struct adapter *adap = pi->adapter;
352 int ret;
b8ff05a9
DM		 353 u64 mhash = 0;
354 u64 uhash = 0;
fc08a01a
HS		 355 bool free = false;
356 bool ucast = is_unicast_ether_addr(mac_addr);
357 const u8 *maclist[1] = {mac_addr};
358 struct hash_mac_addr *new_entry;
359
360 ret = t4_alloc_mac_filt(adap, adap->mbox, pi->viid, free, 1, maclist,
361 NULL, ucast ? &uhash : &mhash, false);
362 if (ret < 0)
363 goto out;
364 /* if hash != 0, then add the addr to hash addr list
365 * so on the end we will calculate the hash for the
366 * list and program it
367 */
368 if (uhash || mhash) {
369 new_entry = kzalloc(sizeof(*new_entry), GFP_ATOMIC);
370 if (!new_entry)
371 return -ENOMEM;
372 ether_addr_copy(new_entry->addr, mac_addr);
373 list_add_tail(&new_entry->list, &adap->mac_hlist);
374 ret = cxgb4_set_addr_hash(pi);
b8ff05a9 375 }
fc08a01a
HS		 376out:
377 return ret < 0 ? ret : 0;
378}
b8ff05a9 379
fc08a01a
HS		 380static int cxgb4_mac_unsync(struct net_device *netdev, const u8 *mac_addr)
381{
382 struct port_info *pi = netdev_priv(netdev);
383 struct adapter *adap = pi->adapter;
384 int ret;
385 const u8 *maclist[1] = {mac_addr};
386 struct hash_mac_addr *entry, *tmp;
b8ff05a9 387
fc08a01a
HS		 388 /* If the MAC address to be removed is in the hash addr
389 * list, delete it from the list and update hash vector
390 */
391 list_for_each_entry_safe(entry, tmp, &adap->mac_hlist, list) {
392 if (ether_addr_equal(entry->addr, mac_addr)) {
393 list_del(&entry->list);
394 kfree(entry);
395 return cxgb4_set_addr_hash(pi);
b8ff05a9
DM		 396 }
397 }
398
fc08a01a
HS		 399 ret = t4_free_mac_filt(adap, adap->mbox, pi->viid, 1, maclist, false);
400 return ret < 0 ? -EINVAL : 0;
b8ff05a9
DM		 401}
402
403/*
404 * Set Rx properties of a port, such as promiscruity, address filters, and MTU.
405 * If @mtu is -1 it is left unchanged.
406 */
407static int set_rxmode(struct net_device *dev, int mtu, bool sleep_ok)
408{
b8ff05a9 409 struct port_info *pi = netdev_priv(dev);
fc08a01a 410 struct adapter *adapter = pi->adapter;
b8ff05a9 411
d01f7abc
HS		 412 __dev_uc_sync(dev, cxgb4_mac_sync, cxgb4_mac_unsync);
413 __dev_mc_sync(dev, cxgb4_mac_sync, cxgb4_mac_unsync);
fc08a01a
HS		 414
415 return t4_set_rxmode(adapter, adapter->mbox, pi->viid, mtu,
416 (dev->flags & IFF_PROMISC) ? 1 : 0,
417 (dev->flags & IFF_ALLMULTI) ? 1 : 0, 1, -1,
418 sleep_ok);
b8ff05a9
DM		 419}
420
421/**
422 * link_start - enable a port
423 * @dev: the port to enable
424 *
425 * Performs the MAC and PHY actions needed to enable a port.
426 */
427static int link_start(struct net_device *dev)
428{
429 int ret;
430 struct port_info *pi = netdev_priv(dev);
b2612722 431 unsigned int mb = pi->adapter->pf;
b8ff05a9
DM		 432
433 /*
434 * We do not set address filters and promiscuity here, the stack does
435 * that step explicitly.
436 */
060e0c75 437 ret = t4_set_rxmode(pi->adapter, mb, pi->viid, dev->mtu, -1, -1, -1,
f646968f 438 !!(dev->features & NETIF_F_HW_VLAN_CTAG_RX), true);
b8ff05a9 439 if (ret == 0) {
060e0c75 440 ret = t4_change_mac(pi->adapter, mb, pi->viid,
b8ff05a9 441 pi->xact_addr_filt, dev->dev_addr, true,
b6bd29e7 442 true);
b8ff05a9
DM		 443 if (ret >= 0) {
444 pi->xact_addr_filt = ret;
445 ret = 0;
446 }
447 }
448 if (ret == 0)
4036da90 449 ret = t4_link_l1cfg(pi->adapter, mb, pi->tx_chan,
060e0c75 450 &pi->link_cfg);
30f00847
AB		 451 if (ret == 0) {
452 local_bh_disable();
688848b1
AB		 453 ret = t4_enable_vi_params(pi->adapter, mb, pi->viid, true,
454 true, CXGB4_DCB_ENABLED);
30f00847
AB		 455 local_bh_enable();
456 }
688848b1 457
b8ff05a9
DM		 458 return ret;
459}
460
688848b1
AB		 461#ifdef CONFIG_CHELSIO_T4_DCB
462/* Handle a Data Center Bridging update message from the firmware. */
463static void dcb_rpl(struct adapter *adap, const struct fw_port_cmd *pcmd)
464{
2b5fb1f2 465 int port = FW_PORT_CMD_PORTID_G(ntohl(pcmd->op_to_portid));
134491fd 466 struct net_device *dev = adap->port[adap->chan_map[port]];
688848b1
AB		 467 int old_dcb_enabled = cxgb4_dcb_enabled(dev);
468 int new_dcb_enabled;
469
470 cxgb4_dcb_handle_fw_update(adap, pcmd);
471 new_dcb_enabled = cxgb4_dcb_enabled(dev);
472
473 /* If the DCB has become enabled or disabled on the port then we're
474 * going to need to set up/tear down DCB Priority parameters for the
475 * TX Queues associated with the port.
476 */
477 if (new_dcb_enabled != old_dcb_enabled)
478 dcb_tx_queue_prio_enable(dev, new_dcb_enabled);
479}
480#endif /* CONFIG_CHELSIO_T4_DCB */
481
f2b7e78d 482/* Response queue handler for the FW event queue.
b8ff05a9
DM		 483 */
484static int fwevtq_handler(struct sge_rspq *q, const __be64 *rsp,
485 const struct pkt_gl *gl)
486{
487 u8 opcode = ((const struct rss_header *)rsp)->opcode;
488
489 rsp++; /* skip RSS header */
b407a4a9
VP		 490
491 /* FW can send EGR_UPDATEs encapsulated in a CPL_FW4_MSG.
492 */
493 if (unlikely(opcode == CPL_FW4_MSG &&
494 ((const struct cpl_fw4_msg *)rsp)->type == FW_TYPE_RSSCPL)) {
495 rsp++;
496 opcode = ((const struct rss_header *)rsp)->opcode;
497 rsp++;
498 if (opcode != CPL_SGE_EGR_UPDATE) {
499 dev_err(q->adap->pdev_dev, "unexpected FW4/CPL %#x on FW event queue\n"
500 , opcode);
501 goto out;
502 }
503 }
504
b8ff05a9
DM		 505 if (likely(opcode == CPL_SGE_EGR_UPDATE)) {
506 const struct cpl_sge_egr_update *p = (void *)rsp;
bdc590b9 507 unsigned int qid = EGR_QID_G(ntohl(p->opcode_qid));
e46dab4d 508 struct sge_txq *txq;
b8ff05a9 509
e46dab4d 510 txq = q->adap->sge.egr_map[qid - q->adap->sge.egr_start];
b8ff05a9 511 txq->restarts++;
ab677ff4 512 if (txq->q_type == CXGB4_TXQ_ETH) {
b8ff05a9
DM		 513 struct sge_eth_txq *eq;
514
515 eq = container_of(txq, struct sge_eth_txq, q);
516 netif_tx_wake_queue(eq->txq);
517 } else {
ab677ff4 518 struct sge_uld_txq *oq;
b8ff05a9 519
ab677ff4 520 oq = container_of(txq, struct sge_uld_txq, q);
b8ff05a9
DM		 521 tasklet_schedule(&oq->qresume_tsk);
522 }
523 } else if (opcode == CPL_FW6_MSG || opcode == CPL_FW4_MSG) {
524 const struct cpl_fw6_msg *p = (void *)rsp;
525
688848b1
AB		 526#ifdef CONFIG_CHELSIO_T4_DCB
527 const struct fw_port_cmd *pcmd = (const void *)p->data;
e2ac9628 528 unsigned int cmd = FW_CMD_OP_G(ntohl(pcmd->op_to_portid));
688848b1 529 unsigned int action =
2b5fb1f2 530 FW_PORT_CMD_ACTION_G(ntohl(pcmd->action_to_len16));
688848b1
AB		 531
532 if (cmd == FW_PORT_CMD &&
533 action == FW_PORT_ACTION_GET_PORT_INFO) {
2b5fb1f2 534 int port = FW_PORT_CMD_PORTID_G(
688848b1 535 be32_to_cpu(pcmd->op_to_portid));
134491fd
HS		 536 struct net_device *dev =
537 q->adap->port[q->adap->chan_map[port]];
688848b1 538 int state_input = ((pcmd->u.info.dcbxdis_pkd &
2b5fb1f2 539 FW_PORT_CMD_DCBXDIS_F)
688848b1
AB		 540 ? CXGB4_DCB_INPUT_FW_DISABLED
541 : CXGB4_DCB_INPUT_FW_ENABLED);
542
543 cxgb4_dcb_state_fsm(dev, state_input);
544 }
545
546 if (cmd == FW_PORT_CMD &&
547 action == FW_PORT_ACTION_L2_DCB_CFG)
548 dcb_rpl(q->adap, pcmd);
549 else
550#endif
551 if (p->type == 0)
552 t4_handle_fw_rpl(q->adap, p->data);
b8ff05a9
DM		 553 } else if (opcode == CPL_L2T_WRITE_RPL) {
554 const struct cpl_l2t_write_rpl *p = (void *)rsp;
555
556 do_l2t_write_rpl(q->adap, p);
f2b7e78d
VP		 557 } else if (opcode == CPL_SET_TCB_RPL) {
558 const struct cpl_set_tcb_rpl *p = (void *)rsp;
559
560 filter_rpl(q->adap, p);
b8ff05a9
DM		 561 } else
562 dev_err(q->adap->pdev_dev,
563 "unexpected CPL %#x on FW event queue\n", opcode);
b407a4a9 564out:
b8ff05a9
DM		 565 return 0;
566}
567
b8ff05a9
DM		 568static void disable_msi(struct adapter *adapter)
569{
570 if (adapter->flags & USING_MSIX) {
571 pci_disable_msix(adapter->pdev);
572 adapter->flags &= ~USING_MSIX;
573 } else if (adapter->flags & USING_MSI) {
574 pci_disable_msi(adapter->pdev);
575 adapter->flags &= ~USING_MSI;
576 }
577}
578
579/*
580 * Interrupt handler for non-data events used with MSI-X.
581 */
582static irqreturn_t t4_nondata_intr(int irq, void *cookie)
583{
584 struct adapter *adap = cookie;
0d804338 585 u32 v = t4_read_reg(adap, MYPF_REG(PL_PF_INT_CAUSE_A));
b8ff05a9 586
0d804338 587 if (v & PFSW_F) {
b8ff05a9 588 adap->swintr = 1;
0d804338 589 t4_write_reg(adap, MYPF_REG(PL_PF_INT_CAUSE_A), v);
b8ff05a9 590 }
c3c7b121
HS		 591 if (adap->flags & MASTER_PF)
592 t4_slow_intr_handler(adap);
b8ff05a9
DM		 593 return IRQ_HANDLED;
594}
595
596/*
597 * Name the MSI-X interrupts.
598 */
599static void name_msix_vecs(struct adapter *adap)
600{
ba27816c 601 int i, j, msi_idx = 2, n = sizeof(adap->msix_info[0].desc);
b8ff05a9
DM		 602
603 /* non-data interrupts */
b1a3c2b6 604 snprintf(adap->msix_info[0].desc, n, "%s", adap->port[0]->name);
b8ff05a9
DM		 605
606 /* FW events */
b1a3c2b6
DM		 607 snprintf(adap->msix_info[1].desc, n, "%s-FWeventq",
608 adap->port[0]->name);
b8ff05a9
DM		 609
610 /* Ethernet queues */
611 for_each_port(adap, j) {
612 struct net_device *d = adap->port[j];
613 const struct port_info *pi = netdev_priv(d);
614
ba27816c 615 for (i = 0; i < pi->nqsets; i++, msi_idx++)
b8ff05a9
DM		 616 snprintf(adap->msix_info[msi_idx].desc, n, "%s-Rx%d",
617 d->name, i);
b8ff05a9 618 }
b8ff05a9
DM		 619}
620
621static int request_msix_queue_irqs(struct adapter *adap)
622{
623 struct sge *s = &adap->sge;
0fbc81b3 624 int err, ethqidx;
cf38be6d 625 int msi_index = 2;
b8ff05a9
DM		 626
627 err = request_irq(adap->msix_info[1].vec, t4_sge_intr_msix, 0,
628 adap->msix_info[1].desc, &s->fw_evtq);
629 if (err)
630 return err;
631
632 for_each_ethrxq(s, ethqidx) {
404d9e3f
VP		 633 err = request_irq(adap->msix_info[msi_index].vec,
634 t4_sge_intr_msix, 0,
635 adap->msix_info[msi_index].desc,
b8ff05a9
DM		 636 &s->ethrxq[ethqidx].rspq);
637 if (err)
638 goto unwind;
404d9e3f 639 msi_index++;
b8ff05a9 640 }
b8ff05a9
DM		 641 return 0;
642
643unwind:
b8ff05a9 644 while (--ethqidx >= 0)
404d9e3f
VP		 645 free_irq(adap->msix_info[--msi_index].vec,
646 &s->ethrxq[ethqidx].rspq);
b8ff05a9
DM		 647 free_irq(adap->msix_info[1].vec, &s->fw_evtq);
648 return err;
649}
650
651static void free_msix_queue_irqs(struct adapter *adap)
652{
404d9e3f 653 int i, msi_index = 2;
b8ff05a9
DM		 654 struct sge *s = &adap->sge;
655
656 free_irq(adap->msix_info[1].vec, &s->fw_evtq);
657 for_each_ethrxq(s, i)
404d9e3f 658 free_irq(adap->msix_info[msi_index++].vec, &s->ethrxq[i].rspq);
b8ff05a9
DM		 659}
660
671b0060 661/**
812034f1 662 * cxgb4_write_rss - write the RSS table for a given port
671b0060
DM		 663 * @pi: the port
664 * @queues: array of queue indices for RSS
665 *
666 * Sets up the portion of the HW RSS table for the port's VI to distribute
667 * packets to the Rx queues in @queues.
c035e183 668 * Should never be called before setting up sge eth rx queues
671b0060 669 */
812034f1 670int cxgb4_write_rss(const struct port_info *pi, const u16 *queues)
671b0060
DM		 671{
672 u16 *rss;
673 int i, err;
c035e183
HS		 674 struct adapter *adapter = pi->adapter;
675 const struct sge_eth_rxq *rxq;
671b0060 676
c035e183 677 rxq = &adapter->sge.ethrxq[pi->first_qset];
671b0060
DM		 678 rss = kmalloc(pi->rss_size * sizeof(u16), GFP_KERNEL);
679 if (!rss)
680 return -ENOMEM;
681
682 /* map the queue indices to queue ids */
683 for (i = 0; i < pi->rss_size; i++, queues++)
c035e183 684 rss[i] = rxq[*queues].rspq.abs_id;
671b0060 685
b2612722 686 err = t4_config_rss_range(adapter, adapter->pf, pi->viid, 0,
060e0c75 687 pi->rss_size, rss, pi->rss_size);
c035e183
HS		 688 /* If Tunnel All Lookup isn't specified in the global RSS
689 * Configuration, then we need to specify a default Ingress
690 * Queue for any ingress packets which aren't hashed. We'll
691 * use our first ingress queue ...
692 */
693 if (!err)
694 err = t4_config_vi_rss(adapter, adapter->mbox, pi->viid,
695 FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN_F |
696 FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN_F |
697 FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN_F |
698 FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN_F |
699 FW_RSS_VI_CONFIG_CMD_UDPEN_F,
700 rss[0]);
671b0060
DM		 701 kfree(rss);
702 return err;
703}
704
b8ff05a9
DM		 705/**
706 * setup_rss - configure RSS
707 * @adap: the adapter
708 *
671b0060 709 * Sets up RSS for each port.
b8ff05a9
DM		 710 */
711static int setup_rss(struct adapter *adap)
712{
c035e183 713 int i, j, err;
b8ff05a9
DM		 714
715 for_each_port(adap, i) {
716 const struct port_info *pi = adap2pinfo(adap, i);
b8ff05a9 717
c035e183
HS		 718 /* Fill default values with equal distribution */
719 for (j = 0; j < pi->rss_size; j++)
720 pi->rss[j] = j % pi->nqsets;
721
812034f1 722 err = cxgb4_write_rss(pi, pi->rss);
b8ff05a9
DM		 723 if (err)
724 return err;
725 }
726 return 0;
727}
728
e46dab4d
DM		 729/*
730 * Return the channel of the ingress queue with the given qid.
731 */
732static unsigned int rxq_to_chan(const struct sge *p, unsigned int qid)
733{
734 qid -= p->ingr_start;
735 return netdev2pinfo(p->ingr_map[qid]->netdev)->tx_chan;
736}
737
b8ff05a9
DM		 738/*
739 * Wait until all NAPI handlers are descheduled.
740 */
741static void quiesce_rx(struct adapter *adap)
742{
743 int i;
744
4b8e27a8 745 for (i = 0; i < adap->sge.ingr_sz; i++) {
b8ff05a9
DM		 746 struct sge_rspq *q = adap->sge.ingr_map[i];
747
5226b791 748 if (q && q->handler)
b8ff05a9
DM		 749 napi_disable(&q->napi);
750 }
751}
752
b37987e8
HS		 753/* Disable interrupt and napi handler */
754static void disable_interrupts(struct adapter *adap)
755{
756 if (adap->flags & FULL_INIT_DONE) {
757 t4_intr_disable(adap);
758 if (adap->flags & USING_MSIX) {
759 free_msix_queue_irqs(adap);
760 free_irq(adap->msix_info[0].vec, adap);
761 } else {
762 free_irq(adap->pdev->irq, adap);
763 }
764 quiesce_rx(adap);
765 }
766}
767
b8ff05a9
DM		 768/*
769 * Enable NAPI scheduling and interrupt generation for all Rx queues.
770 */
771static void enable_rx(struct adapter *adap)
772{
773 int i;
774
4b8e27a8 775 for (i = 0; i < adap->sge.ingr_sz; i++) {
b8ff05a9
DM		 776 struct sge_rspq *q = adap->sge.ingr_map[i];
777
778 if (!q)
779 continue;
5226b791 780 if (q->handler)
b8ff05a9 781 napi_enable(&q->napi);
5226b791 782
b8ff05a9 783 /* 0-increment GTS to start the timer and enable interrupts */
f612b815
HS		 784 t4_write_reg(adap, MYPF_REG(SGE_PF_GTS_A),
785 SEINTARM_V(q->intr_params) |
786 INGRESSQID_V(q->cntxt_id));
b8ff05a9
DM		 787 }
788}
789
1c6a5b0e 790
0fbc81b3 791static int setup_fw_sge_queues(struct adapter *adap)
b8ff05a9 792{
b8ff05a9 793 struct sge *s = &adap->sge;
0fbc81b3 794 int err = 0;
b8ff05a9 795
4b8e27a8
HS		 796 bitmap_zero(s->starving_fl, s->egr_sz);
797 bitmap_zero(s->txq_maperr, s->egr_sz);
b8ff05a9
DM		 798
799 if (adap->flags & USING_MSIX)
94cdb8bb 800 adap->msi_idx = 1; /* vector 0 is for non-queue interrupts */
b8ff05a9
DM		 801 else {
802 err = t4_sge_alloc_rxq(adap, &s->intrq, false, adap->port[0], 0,
2337ba42 803 NULL, NULL, NULL, -1);
b8ff05a9
DM		 804 if (err)
805 return err;
94cdb8bb 806 adap->msi_idx = -((int)s->intrq.abs_id + 1);
b8ff05a9
DM		 807 }
808
809 err = t4_sge_alloc_rxq(adap, &s->fw_evtq, true, adap->port[0],
94cdb8bb 810 adap->msi_idx, NULL, fwevtq_handler, NULL, -1);
0fbc81b3
HS		 811 if (err)
812 t4_free_sge_resources(adap);
813 return err;
814}
815
816/**
817 * setup_sge_queues - configure SGE Tx/Rx/response queues
818 * @adap: the adapter
819 *
820 * Determines how many sets of SGE queues to use and initializes them.
821 * We support multiple queue sets per port if we have MSI-X, otherwise
822 * just one queue set per port.
823 */
824static int setup_sge_queues(struct adapter *adap)
825{
826 int err, i, j;
827 struct sge *s = &adap->sge;
d427caee 828 struct sge_uld_rxq_info *rxq_info = NULL;
0fbc81b3 829 unsigned int cmplqid = 0;
b8ff05a9 830
d427caee
GG		 831 if (is_uld(adap))
832 rxq_info = s->uld_rxq_info[CXGB4_ULD_RDMA];
833
b8ff05a9
DM		 834 for_each_port(adap, i) {
835 struct net_device *dev = adap->port[i];
836 struct port_info *pi = netdev_priv(dev);
837 struct sge_eth_rxq *q = &s->ethrxq[pi->first_qset];
838 struct sge_eth_txq *t = &s->ethtxq[pi->first_qset];
839
840 for (j = 0; j < pi->nqsets; j++, q++) {
94cdb8bb
HS		 841 if (adap->msi_idx > 0)
842 adap->msi_idx++;
b8ff05a9 843 err = t4_sge_alloc_rxq(adap, &q->rspq, false, dev,
94cdb8bb 844 adap->msi_idx, &q->fl,
145ef8a5 845 t4_ethrx_handler,
2337ba42 846 NULL,
193c4c28
AV		 847 t4_get_tp_ch_map(adap,
848 pi->tx_chan));
b8ff05a9
DM		 849 if (err)
850 goto freeout;
851 q->rspq.idx = j;
852 memset(&q->stats, 0, sizeof(q->stats));
853 }
854 for (j = 0; j < pi->nqsets; j++, t++) {
855 err = t4_sge_alloc_eth_txq(adap, t, dev,
856 netdev_get_tx_queue(dev, j),
857 s->fw_evtq.cntxt_id);
858 if (err)
859 goto freeout;
860 }
861 }
862
b8ff05a9 863 for_each_port(adap, i) {
0fbc81b3 864 /* Note that cmplqid below is 0 if we don't
b8ff05a9
DM		 865 * have RDMA queues, and that's the right value.
866 */
0fbc81b3
HS		 867 if (rxq_info)
868 cmplqid = rxq_info->uldrxq[i].rspq.cntxt_id;
869
b8ff05a9 870 err = t4_sge_alloc_ctrl_txq(adap, &s->ctrlq[i], adap->port[i],
0fbc81b3 871 s->fw_evtq.cntxt_id, cmplqid);
b8ff05a9
DM		 872 if (err)
873 goto freeout;
874 }
875
a4569504
AG		 876 if (!is_t4(adap->params.chip)) {
877 err = t4_sge_alloc_eth_txq(adap, &s->ptptxq, adap->port[0],
878 netdev_get_tx_queue(adap->port[0], 0)
879 , s->fw_evtq.cntxt_id);
880 if (err)
881 goto freeout;
882 }
883
9bb59b96 884 t4_write_reg(adap, is_t4(adap->params.chip) ?
837e4a42
HS		 885 MPS_TRC_RSS_CONTROL_A :
886 MPS_T5_TRC_RSS_CONTROL_A,
887 RSSCONTROL_V(netdev2pinfo(adap->port[0])->tx_chan) |
888 QUEUENUMBER_V(s->ethrxq[0].rspq.abs_id));
b8ff05a9 889 return 0;
0fbc81b3
HS		 890freeout:
891 t4_free_sge_resources(adap);
892 return err;
b8ff05a9
DM		 893}
894
688848b1
AB		 895static u16 cxgb_select_queue(struct net_device *dev, struct sk_buff *skb,
896 void *accel_priv, select_queue_fallback_t fallback)
897{
898 int txq;
899
900#ifdef CONFIG_CHELSIO_T4_DCB
901 /* If a Data Center Bridging has been successfully negotiated on this
902 * link then we'll use the skb's priority to map it to a TX Queue.
903 * The skb's priority is determined via the VLAN Tag Priority Code
904 * Point field.
905 */
85eacf3f 906 if (cxgb4_dcb_enabled(dev) && !is_kdump_kernel()) {
688848b1
AB		 907 u16 vlan_tci;
908 int err;
909
910 err = vlan_get_tag(skb, &vlan_tci);
911 if (unlikely(err)) {
912 if (net_ratelimit())
913 netdev_warn(dev,
914 "TX Packet without VLAN Tag on DCB Link\n");
915 txq = 0;
916 } else {
917 txq = (vlan_tci & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
84a200b3
VP		 918#ifdef CONFIG_CHELSIO_T4_FCOE
919 if (skb->protocol == htons(ETH_P_FCOE))
920 txq = skb->priority & 0x7;
921#endif /* CONFIG_CHELSIO_T4_FCOE */
688848b1
AB		 922 }
923 return txq;
924 }
925#endif /* CONFIG_CHELSIO_T4_DCB */
926
927 if (select_queue) {
928 txq = (skb_rx_queue_recorded(skb)
929 ? skb_get_rx_queue(skb)
930 : smp_processor_id());
931
932 while (unlikely(txq >= dev->real_num_tx_queues))
933 txq -= dev->real_num_tx_queues;
934
935 return txq;
936 }
937
938 return fallback(dev, skb) % dev->real_num_tx_queues;
939}
940
b8ff05a9
DM		 941static int closest_timer(const struct sge *s, int time)
942{
943 int i, delta, match = 0, min_delta = INT_MAX;
944
945 for (i = 0; i < ARRAY_SIZE(s->timer_val); i++) {
946 delta = time - s->timer_val[i];
947 if (delta < 0)
948 delta = -delta;
949 if (delta < min_delta) {
950 min_delta = delta;
951 match = i;
952 }
953 }
954 return match;
955}
956
957static int closest_thres(const struct sge *s, int thres)
958{
959 int i, delta, match = 0, min_delta = INT_MAX;
960
961 for (i = 0; i < ARRAY_SIZE(s->counter_val); i++) {
962 delta = thres - s->counter_val[i];
963 if (delta < 0)
964 delta = -delta;
965 if (delta < min_delta) {
966 min_delta = delta;
967 match = i;
968 }
969 }
970 return match;
971}
972
b8ff05a9 973/**
812034f1 974 * cxgb4_set_rspq_intr_params - set a queue's interrupt holdoff parameters
b8ff05a9
DM		 975 * @q: the Rx queue
976 * @us: the hold-off time in us, or 0 to disable timer
977 * @cnt: the hold-off packet count, or 0 to disable counter
978 *
979 * Sets an Rx queue's interrupt hold-off time and packet count. At least
980 * one of the two needs to be enabled for the queue to generate interrupts.
981 */
812034f1
HS		 982int cxgb4_set_rspq_intr_params(struct sge_rspq *q,
983 unsigned int us, unsigned int cnt)
b8ff05a9 984{
c887ad0e
HS		 985 struct adapter *adap = q->adap;
986
b8ff05a9
DM		 987 if ((us | cnt) == 0)
988 cnt = 1;
989
990 if (cnt) {
991 int err;
992 u32 v, new_idx;
993
994 new_idx = closest_thres(&adap->sge, cnt);
995 if (q->desc && q->pktcnt_idx != new_idx) {
996 /* the queue has already been created, update it */
5167865a
HS		 997 v = FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DMAQ) |
998 FW_PARAMS_PARAM_X_V(
999 FW_PARAMS_PARAM_DMAQ_IQ_INTCNTTHRESH) |
1000 FW_PARAMS_PARAM_YZ_V(q->cntxt_id);
b2612722
HS		 1001 err = t4_set_params(adap, adap->mbox, adap->pf, 0, 1,
1002 &v, &new_idx);
b8ff05a9
DM		 1003 if (err)
1004 return err;
1005 }
1006 q->pktcnt_idx = new_idx;
1007 }
1008
1009 us = us == 0 ? 6 : closest_timer(&adap->sge, us);
1ecc7b7a 1010 q->intr_params = QINTR_TIMER_IDX_V(us) | QINTR_CNT_EN_V(cnt > 0);
b8ff05a9
DM		 1011 return 0;
1012}
1013
c8f44aff 1014static int cxgb_set_features(struct net_device *dev, netdev_features_t features)
87b6cf51 1015{
2ed28baa 1016 const struct port_info *pi = netdev_priv(dev);
c8f44aff 1017 netdev_features_t changed = dev->features ^ features;
19ecae2c 1018 int err;
19ecae2c 1019
f646968f 1020 if (!(changed & NETIF_F_HW_VLAN_CTAG_RX))
2ed28baa 1021 return 0;
19ecae2c 1022
b2612722 1023 err = t4_set_rxmode(pi->adapter, pi->adapter->pf, pi->viid, -1,
2ed28baa 1024 -1, -1, -1,
f646968f 1025 !!(features & NETIF_F_HW_VLAN_CTAG_RX), true);
2ed28baa 1026 if (unlikely(err))
f646968f 1027 dev->features = features ^ NETIF_F_HW_VLAN_CTAG_RX;
19ecae2c 1028 return err;
87b6cf51
DM		 1029}
1030
91744948 1031static int setup_debugfs(struct adapter *adap)
b8ff05a9 1032{
b8ff05a9
DM		 1033 if (IS_ERR_OR_NULL(adap->debugfs_root))
1034 return -1;
1035
fd88b31a
HS		 1036#ifdef CONFIG_DEBUG_FS
1037 t4_setup_debugfs(adap);
1038#endif
b8ff05a9
DM		 1039 return 0;
1040}
1041
1042/*
1043 * upper-layer driver support
1044 */
1045
1046/*
1047 * Allocate an active-open TID and set it to the supplied value.
1048 */
1049int cxgb4_alloc_atid(struct tid_info *t, void *data)
1050{
1051 int atid = -1;
1052
1053 spin_lock_bh(&t->atid_lock);
1054 if (t->afree) {
1055 union aopen_entry *p = t->afree;
1056
f2b7e78d 1057 atid = (p - t->atid_tab) + t->atid_base;
b8ff05a9
DM		 1058 t->afree = p->next;
1059 p->data = data;
1060 t->atids_in_use++;
1061 }
1062 spin_unlock_bh(&t->atid_lock);
1063 return atid;
1064}
1065EXPORT_SYMBOL(cxgb4_alloc_atid);
1066
1067/*
1068 * Release an active-open TID.
1069 */
1070void cxgb4_free_atid(struct tid_info *t, unsigned int atid)
1071{
f2b7e78d 1072 union aopen_entry *p = &t->atid_tab[atid - t->atid_base];
b8ff05a9
DM		 1073
1074 spin_lock_bh(&t->atid_lock);
1075 p->next = t->afree;
1076 t->afree = p;
1077 t->atids_in_use--;
1078 spin_unlock_bh(&t->atid_lock);
1079}
1080EXPORT_SYMBOL(cxgb4_free_atid);
1081
1082/*
1083 * Allocate a server TID and set it to the supplied value.
1084 */
1085int cxgb4_alloc_stid(struct tid_info *t, int family, void *data)
1086{
1087 int stid;
1088
1089 spin_lock_bh(&t->stid_lock);
1090 if (family == PF_INET) {
1091 stid = find_first_zero_bit(t->stid_bmap, t->nstids);
1092 if (stid < t->nstids)
1093 __set_bit(stid, t->stid_bmap);
1094 else
1095 stid = -1;
1096 } else {
a99c683e 1097 stid = bitmap_find_free_region(t->stid_bmap, t->nstids, 1);
b8ff05a9
DM		 1098 if (stid < 0)
1099 stid = -1;
1100 }
1101 if (stid >= 0) {
1102 t->stid_tab[stid].data = data;
1103 stid += t->stid_base;
15f63b74
KS		 1104 /* IPv6 requires max of 520 bits or 16 cells in TCAM
1105 * This is equivalent to 4 TIDs. With CLIP enabled it
1106 * needs 2 TIDs.
1107 */
1dec4cec 1108 if (family == PF_INET6) {
a99c683e 1109 t->stids_in_use += 2;
1dec4cec
GG		 1110 t->v6_stids_in_use += 2;
1111 } else {
1112 t->stids_in_use++;
1113 }
b8ff05a9
DM		 1114 }
1115 spin_unlock_bh(&t->stid_lock);
1116 return stid;
1117}
1118EXPORT_SYMBOL(cxgb4_alloc_stid);
1119
dca4faeb
VP		 1120/* Allocate a server filter TID and set it to the supplied value.
1121 */
1122int cxgb4_alloc_sftid(struct tid_info *t, int family, void *data)
1123{
1124 int stid;
1125
1126 spin_lock_bh(&t->stid_lock);
1127 if (family == PF_INET) {
1128 stid = find_next_zero_bit(t->stid_bmap,
1129 t->nstids + t->nsftids, t->nstids);
1130 if (stid < (t->nstids + t->nsftids))
1131 __set_bit(stid, t->stid_bmap);
1132 else
1133 stid = -1;
1134 } else {
1135 stid = -1;
1136 }
1137 if (stid >= 0) {
1138 t->stid_tab[stid].data = data;
470c60c4
KS		 1139 stid -= t->nstids;
1140 stid += t->sftid_base;
2248b293 1141 t->sftids_in_use++;
dca4faeb
VP		 1142 }
1143 spin_unlock_bh(&t->stid_lock);
1144 return stid;
1145}
1146EXPORT_SYMBOL(cxgb4_alloc_sftid);
1147
1148/* Release a server TID.
b8ff05a9
DM		 1149 */
1150void cxgb4_free_stid(struct tid_info *t, unsigned int stid, int family)
1151{
470c60c4
KS		 1152 /* Is it a server filter TID? */
1153 if (t->nsftids && (stid >= t->sftid_base)) {
1154 stid -= t->sftid_base;
1155 stid += t->nstids;
1156 } else {
1157 stid -= t->stid_base;
1158 }
1159
b8ff05a9
DM		 1160 spin_lock_bh(&t->stid_lock);
1161 if (family == PF_INET)
1162 __clear_bit(stid, t->stid_bmap);
1163 else
a99c683e 1164 bitmap_release_region(t->stid_bmap, stid, 1);
b8ff05a9 1165 t->stid_tab[stid].data = NULL;
2248b293 1166 if (stid < t->nstids) {
1dec4cec 1167 if (family == PF_INET6) {
a99c683e 1168 t->stids_in_use -= 2;
1dec4cec
GG		 1169 t->v6_stids_in_use -= 2;
1170 } else {
1171 t->stids_in_use--;
1172 }
2248b293
HS		 1173 } else {
1174 t->sftids_in_use--;
1175 }
1dec4cec 1176
b8ff05a9
DM		 1177 spin_unlock_bh(&t->stid_lock);
1178}
1179EXPORT_SYMBOL(cxgb4_free_stid);
1180
1181/*
1182 * Populate a TID_RELEASE WR. Caller must properly size the skb.
1183 */
1184static void mk_tid_release(struct sk_buff *skb, unsigned int chan,
1185 unsigned int tid)
1186{
1187 struct cpl_tid_release *req;
1188
1189 set_wr_txq(skb, CPL_PRIORITY_SETUP, chan);
4df864c1 1190 req = __skb_put(skb, sizeof(*req));
b8ff05a9
DM		 1191 INIT_TP_WR(req, tid);
1192 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, tid));
1193}
1194
1195/*
1196 * Queue a TID release request and if necessary schedule a work queue to
1197 * process it.
1198 */
31b9c19b 1199static void cxgb4_queue_tid_release(struct tid_info *t, unsigned int chan,
1200 unsigned int tid)
b8ff05a9
DM		 1201{
1202 void **p = &t->tid_tab[tid];
1203 struct adapter *adap = container_of(t, struct adapter, tids);
1204
1205 spin_lock_bh(&adap->tid_release_lock);
1206 *p = adap->tid_release_head;
1207 /* Low 2 bits encode the Tx channel number */
1208 adap->tid_release_head = (void **)((uintptr_t)p | chan);
1209 if (!adap->tid_release_task_busy) {
1210 adap->tid_release_task_busy = true;
29aaee65 1211 queue_work(adap->workq, &adap->tid_release_task);
b8ff05a9
DM		 1212 }
1213 spin_unlock_bh(&adap->tid_release_lock);
1214}
b8ff05a9
DM		 1215
1216/*
1217 * Process the list of pending TID release requests.
1218 */
1219static void process_tid_release_list(struct work_struct *work)
1220{
1221 struct sk_buff *skb;
1222 struct adapter *adap;
1223
1224 adap = container_of(work, struct adapter, tid_release_task);
1225
1226 spin_lock_bh(&adap->tid_release_lock);
1227 while (adap->tid_release_head) {
1228 void **p = adap->tid_release_head;
1229 unsigned int chan = (uintptr_t)p & 3;
1230 p = (void *)p - chan;
1231
1232 adap->tid_release_head = *p;
1233 *p = NULL;
1234 spin_unlock_bh(&adap->tid_release_lock);
1235
1236 while (!(skb = alloc_skb(sizeof(struct cpl_tid_release),
1237 GFP_KERNEL)))
1238 schedule_timeout_uninterruptible(1);
1239
1240 mk_tid_release(skb, chan, p - adap->tids.tid_tab);
1241 t4_ofld_send(adap, skb);
1242 spin_lock_bh(&adap->tid_release_lock);
1243 }
1244 adap->tid_release_task_busy = false;
1245 spin_unlock_bh(&adap->tid_release_lock);
1246}
1247
1248/*
1249 * Release a TID and inform HW. If we are unable to allocate the release
1250 * message we defer to a work queue.
1251 */
1dec4cec
GG		 1252void cxgb4_remove_tid(struct tid_info *t, unsigned int chan, unsigned int tid,
1253 unsigned short family)
b8ff05a9 1254{
b8ff05a9
DM		 1255 struct sk_buff *skb;
1256 struct adapter *adap = container_of(t, struct adapter, tids);
1257
9a1bb9f6
HS		 1258 WARN_ON(tid >= t->ntids);
1259
1260 if (t->tid_tab[tid]) {
1261 t->tid_tab[tid] = NULL;
1dec4cec
GG		 1262 atomic_dec(&t->conns_in_use);
1263 if (t->hash_base && (tid >= t->hash_base)) {
1264 if (family == AF_INET6)
1265 atomic_sub(2, &t->hash_tids_in_use);
1266 else
1267 atomic_dec(&t->hash_tids_in_use);
1268 } else {
1269 if (family == AF_INET6)
1270 atomic_sub(2, &t->tids_in_use);
1271 else
1272 atomic_dec(&t->tids_in_use);
1273 }
9a1bb9f6
HS		 1274 }
1275
b8ff05a9
DM		 1276 skb = alloc_skb(sizeof(struct cpl_tid_release), GFP_ATOMIC);
1277 if (likely(skb)) {
b8ff05a9
DM		 1278 mk_tid_release(skb, chan, tid);
1279 t4_ofld_send(adap, skb);
1280 } else
1281 cxgb4_queue_tid_release(t, chan, tid);
b8ff05a9
DM		 1282}
1283EXPORT_SYMBOL(cxgb4_remove_tid);
1284
1285/*
1286 * Allocate and initialize the TID tables. Returns 0 on success.
1287 */
1288static int tid_init(struct tid_info *t)
1289{
b6f8eaec 1290 struct adapter *adap = container_of(t, struct adapter, tids);
578b46b9
RL		 1291 unsigned int max_ftids = t->nftids + t->nsftids;
1292 unsigned int natids = t->natids;
1293 unsigned int stid_bmap_size;
1294 unsigned int ftid_bmap_size;
1295 size_t size;
b8ff05a9 1296
dca4faeb 1297 stid_bmap_size = BITS_TO_LONGS(t->nstids + t->nsftids);
578b46b9 1298 ftid_bmap_size = BITS_TO_LONGS(t->nftids);
f2b7e78d
VP		 1299 size = t->ntids * sizeof(*t->tid_tab) +
1300 natids * sizeof(*t->atid_tab) +
b8ff05a9 1301 t->nstids * sizeof(*t->stid_tab) +
dca4faeb 1302 t->nsftids * sizeof(*t->stid_tab) +
f2b7e78d 1303 stid_bmap_size * sizeof(long) +
578b46b9
RL		 1304 max_ftids * sizeof(*t->ftid_tab) +
1305 ftid_bmap_size * sizeof(long);
f2b7e78d 1306
752ade68 1307 t->tid_tab = kvzalloc(size, GFP_KERNEL);
b8ff05a9
DM		 1308 if (!t->tid_tab)
1309 return -ENOMEM;
1310
1311 t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids];
1312 t->stid_tab = (struct serv_entry *)&t->atid_tab[natids];
dca4faeb 1313 t->stid_bmap = (unsigned long *)&t->stid_tab[t->nstids + t->nsftids];
f2b7e78d 1314 t->ftid_tab = (struct filter_entry *)&t->stid_bmap[stid_bmap_size];
578b46b9 1315 t->ftid_bmap = (unsigned long *)&t->ftid_tab[max_ftids];
b8ff05a9
DM		 1316 spin_lock_init(&t->stid_lock);
1317 spin_lock_init(&t->atid_lock);
578b46b9 1318 spin_lock_init(&t->ftid_lock);
b8ff05a9
DM		 1319
1320 t->stids_in_use = 0;
1dec4cec 1321 t->v6_stids_in_use = 0;
2248b293 1322 t->sftids_in_use = 0;
b8ff05a9
DM		 1323 t->afree = NULL;
1324 t->atids_in_use = 0;
1325 atomic_set(&t->tids_in_use, 0);
1dec4cec 1326 atomic_set(&t->conns_in_use, 0);
9a1bb9f6 1327 atomic_set(&t->hash_tids_in_use, 0);
b8ff05a9
DM		 1328
1329 /* Setup the free list for atid_tab and clear the stid bitmap. */
1330 if (natids) {
1331 while (--natids)
1332 t->atid_tab[natids - 1].next = &t->atid_tab[natids];
1333 t->afree = t->atid_tab;
1334 }
b6f8eaec 1335
578b46b9
RL		 1336 if (is_offload(adap)) {
1337 bitmap_zero(t->stid_bmap, t->nstids + t->nsftids);
1338 /* Reserve stid 0 for T4/T5 adapters */
1339 if (!t->stid_base &&
1340 CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5)
1341 __set_bit(0, t->stid_bmap);
1342 }
1343
1344 bitmap_zero(t->ftid_bmap, t->nftids);
b8ff05a9
DM		 1345 return 0;
1346}
1347
1348/**
1349 * cxgb4_create_server - create an IP server
1350 * @dev: the device
1351 * @stid: the server TID
1352 * @sip: local IP address to bind server to
1353 * @sport: the server's TCP port
1354 * @queue: queue to direct messages from this server to
1355 *
1356 * Create an IP server for the given port and address.
1357 * Returns <0 on error and one of the %NET_XMIT_* values on success.
1358 */
1359int cxgb4_create_server(const struct net_device *dev, unsigned int stid,
793dad94
VP		 1360 __be32 sip, __be16 sport, __be16 vlan,
1361 unsigned int queue)
b8ff05a9
DM		 1362{
1363 unsigned int chan;
1364 struct sk_buff *skb;
1365 struct adapter *adap;
1366 struct cpl_pass_open_req *req;
80f40c1f 1367 int ret;
b8ff05a9
DM		 1368
1369 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
1370 if (!skb)
1371 return -ENOMEM;
1372
1373 adap = netdev2adap(dev);
4df864c1 1374 req = __skb_put(skb, sizeof(*req));
b8ff05a9
DM		 1375 INIT_TP_WR(req, 0);
1376 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, stid));
1377 req->local_port = sport;
1378 req->peer_port = htons(0);
1379 req->local_ip = sip;
1380 req->peer_ip = htonl(0);
e46dab4d 1381 chan = rxq_to_chan(&adap->sge, queue);
d7990b0c 1382 req->opt0 = cpu_to_be64(TX_CHAN_V(chan));
6c53e938
HS		 1383 req->opt1 = cpu_to_be64(CONN_POLICY_V(CPL_CONN_POLICY_ASK) |
1384 SYN_RSS_ENABLE_F | SYN_RSS_QUEUE_V(queue));
80f40c1f
VP		 1385 ret = t4_mgmt_tx(adap, skb);
1386 return net_xmit_eval(ret);
b8ff05a9
DM		 1387}
1388EXPORT_SYMBOL(cxgb4_create_server);
1389
80f40c1f
VP		 1390/* cxgb4_create_server6 - create an IPv6 server
1391 * @dev: the device
1392 * @stid: the server TID
1393 * @sip: local IPv6 address to bind server to
1394 * @sport: the server's TCP port
1395 * @queue: queue to direct messages from this server to
1396 *
1397 * Create an IPv6 server for the given port and address.
1398 * Returns <0 on error and one of the %NET_XMIT_* values on success.
1399 */
1400int cxgb4_create_server6(const struct net_device *dev, unsigned int stid,
1401 const struct in6_addr *sip, __be16 sport,
1402 unsigned int queue)
1403{
1404 unsigned int chan;
1405 struct sk_buff *skb;
1406 struct adapter *adap;
1407 struct cpl_pass_open_req6 *req;
1408 int ret;
1409
1410 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
1411 if (!skb)
1412 return -ENOMEM;
1413
1414 adap = netdev2adap(dev);
4df864c1 1415 req = __skb_put(skb, sizeof(*req));
80f40c1f
VP		 1416 INIT_TP_WR(req, 0);
1417 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ6, stid));
1418 req->local_port = sport;
1419 req->peer_port = htons(0);
1420 req->local_ip_hi = *(__be64 *)(sip->s6_addr);
1421 req->local_ip_lo = *(__be64 *)(sip->s6_addr + 8);
1422 req->peer_ip_hi = cpu_to_be64(0);
1423 req->peer_ip_lo = cpu_to_be64(0);
1424 chan = rxq_to_chan(&adap->sge, queue);
d7990b0c 1425 req->opt0 = cpu_to_be64(TX_CHAN_V(chan));
6c53e938
HS		 1426 req->opt1 = cpu_to_be64(CONN_POLICY_V(CPL_CONN_POLICY_ASK) |
1427 SYN_RSS_ENABLE_F | SYN_RSS_QUEUE_V(queue));
80f40c1f
VP		 1428 ret = t4_mgmt_tx(adap, skb);
1429 return net_xmit_eval(ret);
1430}
1431EXPORT_SYMBOL(cxgb4_create_server6);
1432
1433int cxgb4_remove_server(const struct net_device *dev, unsigned int stid,
1434 unsigned int queue, bool ipv6)
1435{
1436 struct sk_buff *skb;
1437 struct adapter *adap;
1438 struct cpl_close_listsvr_req *req;
1439 int ret;
1440
1441 adap = netdev2adap(dev);
1442
1443 skb = alloc_skb(sizeof(*req), GFP_KERNEL);
1444 if (!skb)
1445 return -ENOMEM;
1446
4df864c1 1447 req = __skb_put(skb, sizeof(*req));
80f40c1f
VP		 1448 INIT_TP_WR(req, 0);
1449 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, stid));
bdc590b9
HS		 1450 req->reply_ctrl = htons(NO_REPLY_V(0) | (ipv6 ? LISTSVR_IPV6_V(1) :
1451 LISTSVR_IPV6_V(0)) | QUEUENO_V(queue));
80f40c1f
VP		 1452 ret = t4_mgmt_tx(adap, skb);
1453 return net_xmit_eval(ret);
1454}
1455EXPORT_SYMBOL(cxgb4_remove_server);
1456
b8ff05a9
DM		 1457/**
1458 * cxgb4_best_mtu - find the entry in the MTU table closest to an MTU
1459 * @mtus: the HW MTU table
1460 * @mtu: the target MTU
1461 * @idx: index of selected entry in the MTU table
1462 *
1463 * Returns the index and the value in the HW MTU table that is closest to
1464 * but does not exceed @mtu, unless @mtu is smaller than any value in the
1465 * table, in which case that smallest available value is selected.
1466 */
1467unsigned int cxgb4_best_mtu(const unsigned short *mtus, unsigned short mtu,
1468 unsigned int *idx)
1469{
1470 unsigned int i = 0;
1471
1472 while (i < NMTUS - 1 && mtus[i + 1] <= mtu)
1473 ++i;
1474 if (idx)
1475 *idx = i;
1476 return mtus[i];
1477}
1478EXPORT_SYMBOL(cxgb4_best_mtu);
1479
92e7ae71
HS		 1480/**
1481 * cxgb4_best_aligned_mtu - find best MTU, [hopefully] data size aligned
1482 * @mtus: the HW MTU table
1483 * @header_size: Header Size
1484 * @data_size_max: maximum Data Segment Size
1485 * @data_size_align: desired Data Segment Size Alignment (2^N)
1486 * @mtu_idxp: HW MTU Table Index return value pointer (possibly NULL)
1487 *
1488 * Similar to cxgb4_best_mtu() but instead of searching the Hardware
1489 * MTU Table based solely on a Maximum MTU parameter, we break that
1490 * parameter up into a Header Size and Maximum Data Segment Size, and
1491 * provide a desired Data Segment Size Alignment. If we find an MTU in
1492 * the Hardware MTU Table which will result in a Data Segment Size with
1493 * the requested alignment _and_ that MTU isn't "too far" from the
1494 * closest MTU, then we'll return that rather than the closest MTU.
1495 */
1496unsigned int cxgb4_best_aligned_mtu(const unsigned short *mtus,
1497 unsigned short header_size,
1498 unsigned short data_size_max,
1499 unsigned short data_size_align,
1500 unsigned int *mtu_idxp)
1501{
1502 unsigned short max_mtu = header_size + data_size_max;
1503 unsigned short data_size_align_mask = data_size_align - 1;
1504 int mtu_idx, aligned_mtu_idx;
1505
1506 /* Scan the MTU Table till we find an MTU which is larger than our
1507 * Maximum MTU or we reach the end of the table. Along the way,
1508 * record the last MTU found, if any, which will result in a Data
1509 * Segment Length matching the requested alignment.
1510 */
1511 for (mtu_idx = 0, aligned_mtu_idx = -1; mtu_idx < NMTUS; mtu_idx++) {
1512 unsigned short data_size = mtus[mtu_idx] - header_size;
1513
1514 /* If this MTU minus the Header Size would result in a
1515 * Data Segment Size of the desired alignment, remember it.
1516 */
1517 if ((data_size & data_size_align_mask) == 0)
1518 aligned_mtu_idx = mtu_idx;
1519
1520 /* If we're not at the end of the Hardware MTU Table and the
1521 * next element is larger than our Maximum MTU, drop out of
1522 * the loop.
1523 */
1524 if (mtu_idx+1 < NMTUS && mtus[mtu_idx+1] > max_mtu)
1525 break;
1526 }
1527
1528 /* If we fell out of the loop because we ran to the end of the table,
1529 * then we just have to use the last [largest] entry.
1530 */
1531 if (mtu_idx == NMTUS)
1532 mtu_idx--;
1533
1534 /* If we found an MTU which resulted in the requested Data Segment
1535 * Length alignment and that's "not far" from the largest MTU which is
1536 * less than or equal to the maximum MTU, then use that.
1537 */
1538 if (aligned_mtu_idx >= 0 &&
1539 mtu_idx - aligned_mtu_idx <= 1)
1540 mtu_idx = aligned_mtu_idx;
1541
1542 /* If the caller has passed in an MTU Index pointer, pass the
1543 * MTU Index back. Return the MTU value.
1544 */
1545 if (mtu_idxp)
1546 *mtu_idxp = mtu_idx;
1547 return mtus[mtu_idx];
1548}
1549EXPORT_SYMBOL(cxgb4_best_aligned_mtu);
1550
27999805
H		 1551/**
1552 * cxgb4_tp_smt_idx - Get the Source Mac Table index for this VI
1553 * @chip: chip type
1554 * @viid: VI id of the given port
1555 *
1556 * Return the SMT index for this VI.
1557 */
1558unsigned int cxgb4_tp_smt_idx(enum chip_type chip, unsigned int viid)
1559{
1560 /* In T4/T5, SMT contains 256 SMAC entries organized in
1561 * 128 rows of 2 entries each.
1562 * In T6, SMT contains 256 SMAC entries in 256 rows.
1563 * TODO: The below code needs to be updated when we add support
1564 * for 256 VFs.
1565 */
1566 if (CHELSIO_CHIP_VERSION(chip) <= CHELSIO_T5)
1567 return ((viid & 0x7f) << 1);
1568 else
1569 return (viid & 0x7f);
1570}
1571EXPORT_SYMBOL(cxgb4_tp_smt_idx);
1572
b8ff05a9
DM		 1573/**
1574 * cxgb4_port_chan - get the HW channel of a port
1575 * @dev: the net device for the port
1576 *
1577 * Return the HW Tx channel of the given port.
1578 */
1579unsigned int cxgb4_port_chan(const struct net_device *dev)
1580{
1581 return netdev2pinfo(dev)->tx_chan;
1582}
1583EXPORT_SYMBOL(cxgb4_port_chan);
1584
881806bc
VP		 1585unsigned int cxgb4_dbfifo_count(const struct net_device *dev, int lpfifo)
1586{
1587 struct adapter *adap = netdev2adap(dev);
2cc301d2 1588 u32 v1, v2, lp_count, hp_count;
881806bc 1589
f061de42
HS		 1590 v1 = t4_read_reg(adap, SGE_DBFIFO_STATUS_A);
1591 v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2_A);
d14807dd 1592 if (is_t4(adap->params.chip)) {
f061de42
HS		 1593 lp_count = LP_COUNT_G(v1);
1594 hp_count = HP_COUNT_G(v1);
2cc301d2 1595 } else {
f061de42
HS		 1596 lp_count = LP_COUNT_T5_G(v1);
1597 hp_count = HP_COUNT_T5_G(v2);
2cc301d2
SR		 1598 }
1599 return lpfifo ? lp_count : hp_count;
881806bc
VP		 1600}
1601EXPORT_SYMBOL(cxgb4_dbfifo_count);
1602
b8ff05a9
DM		 1603/**
1604 * cxgb4_port_viid - get the VI id of a port
1605 * @dev: the net device for the port
1606 *
1607 * Return the VI id of the given port.
1608 */
1609unsigned int cxgb4_port_viid(const struct net_device *dev)
1610{
1611 return netdev2pinfo(dev)->viid;
1612}
1613EXPORT_SYMBOL(cxgb4_port_viid);
1614
1615/**
1616 * cxgb4_port_idx - get the index of a port
1617 * @dev: the net device for the port
1618 *
1619 * Return the index of the given port.
1620 */
1621unsigned int cxgb4_port_idx(const struct net_device *dev)
1622{
1623 return netdev2pinfo(dev)->port_id;
1624}
1625EXPORT_SYMBOL(cxgb4_port_idx);
1626
b8ff05a9
DM		 1627void cxgb4_get_tcp_stats(struct pci_dev *pdev, struct tp_tcp_stats *v4,
1628 struct tp_tcp_stats *v6)
1629{
1630 struct adapter *adap = pci_get_drvdata(pdev);
1631
1632 spin_lock(&adap->stats_lock);
1633 t4_tp_get_tcp_stats(adap, v4, v6);
1634 spin_unlock(&adap->stats_lock);
1635}
1636EXPORT_SYMBOL(cxgb4_get_tcp_stats);
1637
1638void cxgb4_iscsi_init(struct net_device *dev, unsigned int tag_mask,
1639 const unsigned int *pgsz_order)
1640{
1641 struct adapter *adap = netdev2adap(dev);
1642
0d804338
HS		 1643 t4_write_reg(adap, ULP_RX_ISCSI_TAGMASK_A, tag_mask);
1644 t4_write_reg(adap, ULP_RX_ISCSI_PSZ_A, HPZ0_V(pgsz_order[0]) |
1645 HPZ1_V(pgsz_order[1]) | HPZ2_V(pgsz_order[2]) |
1646 HPZ3_V(pgsz_order[3]));
b8ff05a9
DM		 1647}
1648EXPORT_SYMBOL(cxgb4_iscsi_init);
1649
3069ee9b
VP		 1650int cxgb4_flush_eq_cache(struct net_device *dev)
1651{
1652 struct adapter *adap = netdev2adap(dev);
3069ee9b 1653
5d700ecb 1654 return t4_sge_ctxt_flush(adap, adap->mbox);
3069ee9b
VP		 1655}
1656EXPORT_SYMBOL(cxgb4_flush_eq_cache);
1657
1658static int read_eq_indices(struct adapter *adap, u16 qid, u16 *pidx, u16 *cidx)
1659{
f061de42 1660 u32 addr = t4_read_reg(adap, SGE_DBQ_CTXT_BADDR_A) + 24 * qid + 8;
3069ee9b
VP		 1661 __be64 indices;
1662 int ret;
1663
fc5ab020
HS		 1664 spin_lock(&adap->win0_lock);
1665 ret = t4_memory_rw(adap, 0, MEM_EDC0, addr,
1666 sizeof(indices), (__be32 *)&indices,
1667 T4_MEMORY_READ);
1668 spin_unlock(&adap->win0_lock);
3069ee9b 1669 if (!ret) {
404d9e3f
VP		 1670 *cidx = (be64_to_cpu(indices) >> 25) & 0xffff;
1671 *pidx = (be64_to_cpu(indices) >> 9) & 0xffff;
3069ee9b
VP		 1672 }
1673 return ret;
1674}
1675
1676int cxgb4_sync_txq_pidx(struct net_device *dev, u16 qid, u16 pidx,
1677 u16 size)
1678{
1679 struct adapter *adap = netdev2adap(dev);
1680 u16 hw_pidx, hw_cidx;
1681 int ret;
1682
1683 ret = read_eq_indices(adap, qid, &hw_pidx, &hw_cidx);
1684 if (ret)
1685 goto out;
1686
1687 if (pidx != hw_pidx) {
1688 u16 delta;
f612b815 1689 u32 val;
3069ee9b
VP		 1690
1691 if (pidx >= hw_pidx)
1692 delta = pidx - hw_pidx;
1693 else
1694 delta = size - hw_pidx + pidx;
f612b815
HS		 1695
1696 if (is_t4(adap->params.chip))
1697 val = PIDX_V(delta);
1698 else
1699 val = PIDX_T5_V(delta);
3069ee9b 1700 wmb();
f612b815
HS		 1701 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
1702 QID_V(qid) | val);
3069ee9b
VP		 1703 }
1704out:
1705 return ret;
1706}
1707EXPORT_SYMBOL(cxgb4_sync_txq_pidx);
1708
031cf476
HS		 1709int cxgb4_read_tpte(struct net_device *dev, u32 stag, __be32 *tpte)
1710{
1711 struct adapter *adap;
1712 u32 offset, memtype, memaddr;
6559a7e8 1713 u32 edc0_size, edc1_size, mc0_size, mc1_size, size;
031cf476
HS		 1714 u32 edc0_end, edc1_end, mc0_end, mc1_end;
1715 int ret;
1716
1717 adap = netdev2adap(dev);
1718
1719 offset = ((stag >> 8) * 32) + adap->vres.stag.start;
1720
1721 /* Figure out where the offset lands in the Memory Type/Address scheme.
1722 * This code assumes that the memory is laid out starting at offset 0
1723 * with no breaks as: EDC0, EDC1, MC0, MC1. All cards have both EDC0
1724 * and EDC1. Some cards will have neither MC0 nor MC1, most cards have
1725 * MC0, and some have both MC0 and MC1.
1726 */
6559a7e8
HS		 1727 size = t4_read_reg(adap, MA_EDRAM0_BAR_A);
1728 edc0_size = EDRAM0_SIZE_G(size) << 20;
1729 size = t4_read_reg(adap, MA_EDRAM1_BAR_A);
1730 edc1_size = EDRAM1_SIZE_G(size) << 20;
1731 size = t4_read_reg(adap, MA_EXT_MEMORY0_BAR_A);
1732 mc0_size = EXT_MEM0_SIZE_G(size) << 20;
031cf476
HS		 1733
1734 edc0_end = edc0_size;
1735 edc1_end = edc0_end + edc1_size;
1736 mc0_end = edc1_end + mc0_size;
1737
1738 if (offset < edc0_end) {
1739 memtype = MEM_EDC0;
1740 memaddr = offset;
1741 } else if (offset < edc1_end) {
1742 memtype = MEM_EDC1;
1743 memaddr = offset - edc0_end;
1744 } else {
1745 if (offset < mc0_end) {
1746 memtype = MEM_MC0;
1747 memaddr = offset - edc1_end;
3ccc6cf7 1748 } else if (is_t5(adap->params.chip)) {
6559a7e8
HS		 1749 size = t4_read_reg(adap, MA_EXT_MEMORY1_BAR_A);
1750 mc1_size = EXT_MEM1_SIZE_G(size) << 20;
031cf476
HS		 1751 mc1_end = mc0_end + mc1_size;
1752 if (offset < mc1_end) {
1753 memtype = MEM_MC1;
1754 memaddr = offset - mc0_end;
1755 } else {
1756 /* offset beyond the end of any memory */
1757 goto err;
1758 }
3ccc6cf7
HS		 1759 } else {
1760 /* T4/T6 only has a single memory channel */
1761 goto err;
031cf476
HS		 1762 }
1763 }
1764
1765 spin_lock(&adap->win0_lock);
1766 ret = t4_memory_rw(adap, 0, memtype, memaddr, 32, tpte, T4_MEMORY_READ);
1767 spin_unlock(&adap->win0_lock);
1768 return ret;
1769
1770err:
1771 dev_err(adap->pdev_dev, "stag %#x, offset %#x out of range\n",
1772 stag, offset);
1773 return -EINVAL;
1774}
1775EXPORT_SYMBOL(cxgb4_read_tpte);
1776
7730b4c7
HS		 1777u64 cxgb4_read_sge_timestamp(struct net_device *dev)
1778{
1779 u32 hi, lo;
1780 struct adapter *adap;
1781
1782 adap = netdev2adap(dev);
f612b815
HS		 1783 lo = t4_read_reg(adap, SGE_TIMESTAMP_LO_A);
1784 hi = TSVAL_G(t4_read_reg(adap, SGE_TIMESTAMP_HI_A));
7730b4c7
HS		 1785
1786 return ((u64)hi << 32) | (u64)lo;
1787}
1788EXPORT_SYMBOL(cxgb4_read_sge_timestamp);
1789
df64e4d3
HS		 1790int cxgb4_bar2_sge_qregs(struct net_device *dev,
1791 unsigned int qid,
1792 enum cxgb4_bar2_qtype qtype,
66cf188e 1793 int user,
df64e4d3
HS		 1794 u64 *pbar2_qoffset,
1795 unsigned int *pbar2_qid)
1796{
b2612722 1797 return t4_bar2_sge_qregs(netdev2adap(dev),
df64e4d3
HS		 1798 qid,
1799 (qtype == CXGB4_BAR2_QTYPE_EGRESS
1800 ? T4_BAR2_QTYPE_EGRESS
1801 : T4_BAR2_QTYPE_INGRESS),
66cf188e 1802 user,
df64e4d3
HS		 1803 pbar2_qoffset,
1804 pbar2_qid);
1805}
1806EXPORT_SYMBOL(cxgb4_bar2_sge_qregs);
1807
b8ff05a9
DM		 1808static struct pci_driver cxgb4_driver;
1809
1810static void check_neigh_update(struct neighbour *neigh)
1811{
1812 const struct device *parent;
1813 const struct net_device *netdev = neigh->dev;
1814
d0d7b10b 1815 if (is_vlan_dev(netdev))
b8ff05a9
DM		 1816 netdev = vlan_dev_real_dev(netdev);
1817 parent = netdev->dev.parent;
1818 if (parent && parent->driver == &cxgb4_driver.driver)
1819 t4_l2t_update(dev_get_drvdata(parent), neigh);
1820}
1821
1822static int netevent_cb(struct notifier_block *nb, unsigned long event,
1823 void *data)
1824{
1825 switch (event) {
1826 case NETEVENT_NEIGH_UPDATE:
1827 check_neigh_update(data);
1828 break;
b8ff05a9
DM		 1829 case NETEVENT_REDIRECT:
1830 default:
1831 break;
1832 }
1833 return 0;
1834}
1835
1836static bool netevent_registered;
1837static struct notifier_block cxgb4_netevent_nb = {
1838 .notifier_call = netevent_cb
1839};
1840
3069ee9b
VP		 1841static void drain_db_fifo(struct adapter *adap, int usecs)
1842{
2cc301d2 1843 u32 v1, v2, lp_count, hp_count;
3069ee9b
VP		 1844
1845 do {
f061de42
HS		 1846 v1 = t4_read_reg(adap, SGE_DBFIFO_STATUS_A);
1847 v2 = t4_read_reg(adap, SGE_DBFIFO_STATUS2_A);
d14807dd 1848 if (is_t4(adap->params.chip)) {
f061de42
HS		 1849 lp_count = LP_COUNT_G(v1);
1850 hp_count = HP_COUNT_G(v1);
2cc301d2 1851 } else {
f061de42
HS		 1852 lp_count = LP_COUNT_T5_G(v1);
1853 hp_count = HP_COUNT_T5_G(v2);
2cc301d2
SR		 1854 }
1855
1856 if (lp_count == 0 && hp_count == 0)
1857 break;
3069ee9b
VP		 1858 set_current_state(TASK_UNINTERRUPTIBLE);
1859 schedule_timeout(usecs_to_jiffies(usecs));
3069ee9b
VP		 1860 } while (1);
1861}
1862
1863static void disable_txq_db(struct sge_txq *q)
1864{
05eb2389
SW		 1865 unsigned long flags;
1866
1867 spin_lock_irqsave(&q->db_lock, flags);
3069ee9b 1868 q->db_disabled = 1;
05eb2389 1869 spin_unlock_irqrestore(&q->db_lock, flags);
3069ee9b
VP		 1870}
1871
05eb2389 1872static void enable_txq_db(struct adapter *adap, struct sge_txq *q)
3069ee9b
VP		 1873{
1874 spin_lock_irq(&q->db_lock);
05eb2389
SW		 1875 if (q->db_pidx_inc) {
1876 /* Make sure that all writes to the TX descriptors
1877 * are committed before we tell HW about them.
1878 */
1879 wmb();
f612b815
HS		 1880 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
1881 QID_V(q->cntxt_id) | PIDX_V(q->db_pidx_inc));
05eb2389
SW		 1882 q->db_pidx_inc = 0;
1883 }
3069ee9b
VP		 1884 q->db_disabled = 0;
1885 spin_unlock_irq(&q->db_lock);
1886}
1887
1888static void disable_dbs(struct adapter *adap)
1889{
1890 int i;
1891
1892 for_each_ethrxq(&adap->sge, i)
1893 disable_txq_db(&adap->sge.ethtxq[i].q);
ab677ff4
HS		 1894 if (is_offload(adap)) {
1895 struct sge_uld_txq_info *txq_info =
1896 adap->sge.uld_txq_info[CXGB4_TX_OFLD];
1897
1898 if (txq_info) {
1899 for_each_ofldtxq(&adap->sge, i) {
1900 struct sge_uld_txq *txq = &txq_info->uldtxq[i];
1901
1902 disable_txq_db(&txq->q);
1903 }
1904 }
1905 }
3069ee9b
VP		 1906 for_each_port(adap, i)
1907 disable_txq_db(&adap->sge.ctrlq[i].q);
1908}
1909
1910static void enable_dbs(struct adapter *adap)
1911{
1912 int i;
1913
1914 for_each_ethrxq(&adap->sge, i)
05eb2389 1915 enable_txq_db(adap, &adap->sge.ethtxq[i].q);
ab677ff4
HS		 1916 if (is_offload(adap)) {
1917 struct sge_uld_txq_info *txq_info =
1918 adap->sge.uld_txq_info[CXGB4_TX_OFLD];
1919
1920 if (txq_info) {
1921 for_each_ofldtxq(&adap->sge, i) {
1922 struct sge_uld_txq *txq = &txq_info->uldtxq[i];
1923
1924 enable_txq_db(adap, &txq->q);
1925 }
1926 }
1927 }
3069ee9b 1928 for_each_port(adap, i)
05eb2389
SW		 1929 enable_txq_db(adap, &adap->sge.ctrlq[i].q);
1930}
1931
1932static void notify_rdma_uld(struct adapter *adap, enum cxgb4_control cmd)
1933{
0fbc81b3
HS		 1934 enum cxgb4_uld type = CXGB4_ULD_RDMA;
1935
1936 if (adap->uld && adap->uld[type].handle)
1937 adap->uld[type].control(adap->uld[type].handle, cmd);
05eb2389
SW		 1938}
1939
1940static void process_db_full(struct work_struct *work)
1941{
1942 struct adapter *adap;
1943
1944 adap = container_of(work, struct adapter, db_full_task);
1945
1946 drain_db_fifo(adap, dbfifo_drain_delay);
1947 enable_dbs(adap);
1948 notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
3ccc6cf7
HS		 1949 if (CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5)
1950 t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
1951 DBFIFO_HP_INT_F | DBFIFO_LP_INT_F,
1952 DBFIFO_HP_INT_F | DBFIFO_LP_INT_F);
1953 else
1954 t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
1955 DBFIFO_LP_INT_F, DBFIFO_LP_INT_F);
3069ee9b
VP		 1956}
1957
1958static void sync_txq_pidx(struct adapter *adap, struct sge_txq *q)
1959{
1960 u16 hw_pidx, hw_cidx;
1961 int ret;
1962
05eb2389 1963 spin_lock_irq(&q->db_lock);
3069ee9b
VP		 1964 ret = read_eq_indices(adap, (u16)q->cntxt_id, &hw_pidx, &hw_cidx);
1965 if (ret)
1966 goto out;
1967 if (q->db_pidx != hw_pidx) {
1968 u16 delta;
f612b815 1969 u32 val;
3069ee9b
VP		 1970
1971 if (q->db_pidx >= hw_pidx)
1972 delta = q->db_pidx - hw_pidx;
1973 else
1974 delta = q->size - hw_pidx + q->db_pidx;
f612b815
HS		 1975
1976 if (is_t4(adap->params.chip))
1977 val = PIDX_V(delta);
1978 else
1979 val = PIDX_T5_V(delta);
3069ee9b 1980 wmb();
f612b815
HS		 1981 t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL_A),
1982 QID_V(q->cntxt_id) | val);
3069ee9b
VP		 1983 }
1984out:
1985 q->db_disabled = 0;
05eb2389
SW		 1986 q->db_pidx_inc = 0;
1987 spin_unlock_irq(&q->db_lock);
3069ee9b
VP		 1988 if (ret)
1989 CH_WARN(adap, "DB drop recovery failed.\n");
1990}
0fbc81b3 1991
3069ee9b
VP		 1992static void recover_all_queues(struct adapter *adap)
1993{
1994 int i;
1995
1996 for_each_ethrxq(&adap->sge, i)
1997 sync_txq_pidx(adap, &adap->sge.ethtxq[i].q);
ab677ff4
HS		 1998 if (is_offload(adap)) {
1999 struct sge_uld_txq_info *txq_info =
2000 adap->sge.uld_txq_info[CXGB4_TX_OFLD];
2001 if (txq_info) {
2002 for_each_ofldtxq(&adap->sge, i) {
2003 struct sge_uld_txq *txq = &txq_info->uldtxq[i];
2004
2005 sync_txq_pidx(adap, &txq->q);
2006 }
2007 }
2008 }
3069ee9b
VP		 2009 for_each_port(adap, i)
2010 sync_txq_pidx(adap, &adap->sge.ctrlq[i].q);
2011}
2012
881806bc
VP		 2013static void process_db_drop(struct work_struct *work)
2014{
2015 struct adapter *adap;
881806bc 2016
3069ee9b 2017 adap = container_of(work, struct adapter, db_drop_task);
881806bc 2018
d14807dd 2019 if (is_t4(adap->params.chip)) {
05eb2389 2020 drain_db_fifo(adap, dbfifo_drain_delay);
2cc301d2 2021 notify_rdma_uld(adap, CXGB4_CONTROL_DB_DROP);
05eb2389 2022 drain_db_fifo(adap, dbfifo_drain_delay);
2cc301d2 2023 recover_all_queues(adap);
05eb2389 2024 drain_db_fifo(adap, dbfifo_drain_delay);
2cc301d2 2025 enable_dbs(adap);
05eb2389 2026 notify_rdma_uld(adap, CXGB4_CONTROL_DB_EMPTY);
3ccc6cf7 2027 } else if (is_t5(adap->params.chip)) {
2cc301d2
SR		 2028 u32 dropped_db = t4_read_reg(adap, 0x010ac);
2029 u16 qid = (dropped_db >> 15) & 0x1ffff;
2030 u16 pidx_inc = dropped_db & 0x1fff;
df64e4d3
HS		 2031 u64 bar2_qoffset;
2032 unsigned int bar2_qid;
2033 int ret;
2cc301d2 2034
b2612722 2035 ret = t4_bar2_sge_qregs(adap, qid, T4_BAR2_QTYPE_EGRESS,
e0456717 2036 0, &bar2_qoffset, &bar2_qid);
df64e4d3
HS		 2037 if (ret)
2038 dev_err(adap->pdev_dev, "doorbell drop recovery: "
2039 "qid=%d, pidx_inc=%d\n", qid, pidx_inc);
2040 else
f612b815 2041 writel(PIDX_T5_V(pidx_inc) | QID_V(bar2_qid),
df64e4d3 2042 adap->bar2 + bar2_qoffset + SGE_UDB_KDOORBELL);
2cc301d2
SR		 2043
2044 /* Re-enable BAR2 WC */
2045 t4_set_reg_field(adap, 0x10b0, 1<<15, 1<<15);
2046 }
2047
3ccc6cf7
HS		 2048 if (CHELSIO_CHIP_VERSION(adap->params.chip) <= CHELSIO_T5)
2049 t4_set_reg_field(adap, SGE_DOORBELL_CONTROL_A, DROPPED_DB_F, 0);
881806bc
VP		 2050}
2051
2052void t4_db_full(struct adapter *adap)
2053{
d14807dd 2054 if (is_t4(adap->params.chip)) {
05eb2389
SW		 2055 disable_dbs(adap);
2056 notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
f612b815
HS		 2057 t4_set_reg_field(adap, SGE_INT_ENABLE3_A,
2058 DBFIFO_HP_INT_F | DBFIFO_LP_INT_F, 0);
29aaee65 2059 queue_work(adap->workq, &adap->db_full_task);
2cc301d2 2060 }
881806bc
VP		 2061}
2062
2063void t4_db_dropped(struct adapter *adap)
2064{
05eb2389
SW		 2065 if (is_t4(adap->params.chip)) {
2066 disable_dbs(adap);
2067 notify_rdma_uld(adap, CXGB4_CONTROL_DB_FULL);
2068 }
29aaee65 2069 queue_work(adap->workq, &adap->db_drop_task);
881806bc
VP		 2070}
2071
0fbc81b3
HS		 2072void t4_register_netevent_notifier(void)
2073{
b8ff05a9
DM		 2074 if (!netevent_registered) {
2075 register_netevent_notifier(&cxgb4_netevent_nb);
2076 netevent_registered = true;
2077 }
b8ff05a9
DM		 2078}
2079
2080static void detach_ulds(struct adapter *adap)
2081{
2082 unsigned int i;
2083
2084 mutex_lock(&uld_mutex);
2085 list_del(&adap->list_node);
6a146f3a 2086
b8ff05a9 2087 for (i = 0; i < CXGB4_ULD_MAX; i++)
6a146f3a 2088 if (adap->uld && adap->uld[i].handle)
94cdb8bb
HS		 2089 adap->uld[i].state_change(adap->uld[i].handle,
2090 CXGB4_STATE_DETACH);
6a146f3a 2091
b8ff05a9
DM		 2092 if (netevent_registered && list_empty(&adapter_list)) {
2093 unregister_netevent_notifier(&cxgb4_netevent_nb);
2094 netevent_registered = false;
2095 }
2096 mutex_unlock(&uld_mutex);
2097}
2098
2099static void notify_ulds(struct adapter *adap, enum cxgb4_state new_state)
2100{
2101 unsigned int i;
2102
2103 mutex_lock(&uld_mutex);
2104 for (i = 0; i < CXGB4_ULD_MAX; i++)
94cdb8bb
HS		 2105 if (adap->uld && adap->uld[i].handle)
2106 adap->uld[i].state_change(adap->uld[i].handle,
2107 new_state);
b8ff05a9
DM		 2108 mutex_unlock(&uld_mutex);
2109}
2110
1bb60376 2111#if IS_ENABLED(CONFIG_IPV6)
b5a02f50
AB		 2112static int cxgb4_inet6addr_handler(struct notifier_block *this,
2113 unsigned long event, void *data)
01bcca68 2114{
b5a02f50
AB		 2115 struct inet6_ifaddr *ifa = data;
2116 struct net_device *event_dev = ifa->idev->dev;
2117 const struct device *parent = NULL;
2118#if IS_ENABLED(CONFIG_BONDING)
01bcca68 2119 struct adapter *adap;
b5a02f50 2120#endif
d0d7b10b 2121 if (is_vlan_dev(event_dev))
b5a02f50
AB		 2122 event_dev = vlan_dev_real_dev(event_dev);
2123#if IS_ENABLED(CONFIG_BONDING)
2124 if (event_dev->flags & IFF_MASTER) {
2125 list_for_each_entry(adap, &adapter_list, list_node) {
2126 switch (event) {
2127 case NETDEV_UP:
2128 cxgb4_clip_get(adap->port[0],
2129 (const u32 *)ifa, 1);
2130 break;
2131 case NETDEV_DOWN:
2132 cxgb4_clip_release(adap->port[0],
2133 (const u32 *)ifa, 1);
2134 break;
2135 default:
2136 break;
2137 }
2138 }
2139 return NOTIFY_OK;
2140 }
2141#endif
01bcca68 2142
b5a02f50
AB		 2143 if (event_dev)
2144 parent = event_dev->dev.parent;
01bcca68 2145
b5a02f50 2146 if (parent && parent->driver == &cxgb4_driver.driver) {
01bcca68
VP		 2147 switch (event) {
2148 case NETDEV_UP:
b5a02f50 2149 cxgb4_clip_get(event_dev, (const u32 *)ifa, 1);
01bcca68
VP		 2150 break;
2151 case NETDEV_DOWN:
b5a02f50 2152 cxgb4_clip_release(event_dev, (const u32 *)ifa, 1);
01bcca68
VP		 2153 break;
2154 default:
2155 break;
2156 }
2157 }
b5a02f50 2158 return NOTIFY_OK;
01bcca68
VP		 2159}
2160
b5a02f50 2161static bool inet6addr_registered;
01bcca68
VP		 2162static struct notifier_block cxgb4_inet6addr_notifier = {
2163 .notifier_call = cxgb4_inet6addr_handler
2164};
2165
01bcca68
VP		 2166static void update_clip(const struct adapter *adap)
2167{
2168 int i;
2169 struct net_device *dev;
2170 int ret;
2171
2172 rcu_read_lock();
2173
2174 for (i = 0; i < MAX_NPORTS; i++) {
2175 dev = adap->port[i];
2176 ret = 0;
2177
2178 if (dev)
b5a02f50 2179 ret = cxgb4_update_root_dev_clip(dev);
01bcca68
VP		 2180
2181 if (ret < 0)
2182 break;
2183 }
2184 rcu_read_unlock();
2185}
1bb60376 2186#endif /* IS_ENABLED(CONFIG_IPV6) */
01bcca68 2187
b8ff05a9
DM		 2188/**
2189 * cxgb_up - enable the adapter
2190 * @adap: adapter being enabled
2191 *
2192 * Called when the first port is enabled, this function performs the
2193 * actions necessary to make an adapter operational, such as completing
2194 * the initialization of HW modules, and enabling interrupts.
2195 *
2196 * Must be called with the rtnl lock held.
2197 */
2198static int cxgb_up(struct adapter *adap)
2199{
aaefae9b 2200 int err;
b8ff05a9 2201
91060381 2202 mutex_lock(&uld_mutex);
aaefae9b
DM		 2203 err = setup_sge_queues(adap);
2204 if (err)
91060381 2205 goto rel_lock;
aaefae9b
DM		 2206 err = setup_rss(adap);
2207 if (err)
2208 goto freeq;
b8ff05a9
DM		 2209
2210 if (adap->flags & USING_MSIX) {
aaefae9b 2211 name_msix_vecs(adap);
b8ff05a9
DM		 2212 err = request_irq(adap->msix_info[0].vec, t4_nondata_intr, 0,
2213 adap->msix_info[0].desc, adap);
2214 if (err)
2215 goto irq_err;
b8ff05a9
DM		 2216 err = request_msix_queue_irqs(adap);
2217 if (err) {
2218 free_irq(adap->msix_info[0].vec, adap);
2219 goto irq_err;
2220 }
2221 } else {
2222 err = request_irq(adap->pdev->irq, t4_intr_handler(adap),
2223 (adap->flags & USING_MSI) ? 0 : IRQF_SHARED,
b1a3c2b6 2224 adap->port[0]->name, adap);
b8ff05a9
DM		 2225 if (err)
2226 goto irq_err;
2227 }
e7519f99 2228
b8ff05a9
DM		 2229 enable_rx(adap);
2230 t4_sge_start(adap);
2231 t4_intr_enable(adap);
aaefae9b 2232 adap->flags |= FULL_INIT_DONE;
e7519f99
GG		 2233 mutex_unlock(&uld_mutex);
2234
b8ff05a9 2235 notify_ulds(adap, CXGB4_STATE_UP);
1bb60376 2236#if IS_ENABLED(CONFIG_IPV6)
01bcca68 2237 update_clip(adap);
1bb60376 2238#endif
fc08a01a
HS		 2239 /* Initialize hash mac addr list*/
2240 INIT_LIST_HEAD(&adap->mac_hlist);
b8ff05a9 2241 return err;
91060381 2242
b8ff05a9
DM		 2243 irq_err:
2244 dev_err(adap->pdev_dev, "request_irq failed, err %d\n", err);
aaefae9b
DM		 2245 freeq:
2246 t4_free_sge_resources(adap);
91060381
RR		 2247 rel_lock:
2248 mutex_unlock(&uld_mutex);
2249 return err;
b8ff05a9
DM		 2250}
2251
2252static void cxgb_down(struct adapter *adapter)
2253{
b8ff05a9 2254 cancel_work_sync(&adapter->tid_release_task);
881806bc
VP		 2255 cancel_work_sync(&adapter->db_full_task);
2256 cancel_work_sync(&adapter->db_drop_task);
b8ff05a9 2257 adapter->tid_release_task_busy = false;
204dc3c0 2258 adapter->tid_release_head = NULL;
b8ff05a9 2259
aaefae9b
DM		 2260 t4_sge_stop(adapter);
2261 t4_free_sge_resources(adapter);
2262 adapter->flags &= ~FULL_INIT_DONE;
b8ff05a9
DM		 2263}
2264
2265/*
2266 * net_device operations
2267 */
2268static int cxgb_open(struct net_device *dev)
2269{
2270 int err;
2271 struct port_info *pi = netdev_priv(dev);
2272 struct adapter *adapter = pi->adapter;
2273
6a3c869a
DM		 2274 netif_carrier_off(dev);
2275
aaefae9b
DM		 2276 if (!(adapter->flags & FULL_INIT_DONE)) {
2277 err = cxgb_up(adapter);
2278 if (err < 0)
2279 return err;
2280 }
b8ff05a9 2281
2061ec3f
GG		 2282 /* It's possible that the basic port information could have
2283 * changed since we first read it.
2284 */
2285 err = t4_update_port_info(pi);
2286 if (err < 0)
2287 return err;
2288
f68707b8
DM		 2289 err = link_start(dev);
2290 if (!err)
2291 netif_tx_start_all_queues(dev);
2292 return err;
b8ff05a9
DM		 2293}
2294
2295static int cxgb_close(struct net_device *dev)
2296{
b8ff05a9
DM		 2297 struct port_info *pi = netdev_priv(dev);
2298 struct adapter *adapter = pi->adapter;
2299
2300 netif_tx_stop_all_queues(dev);
2301 netif_carrier_off(dev);
b2612722 2302 return t4_enable_vi(adapter, adapter->pf, pi->viid, false, false);
b8ff05a9
DM		 2303}
2304
dca4faeb 2305int cxgb4_create_server_filter(const struct net_device *dev, unsigned int stid,
793dad94
VP		 2306 __be32 sip, __be16 sport, __be16 vlan,
2307 unsigned int queue, unsigned char port, unsigned char mask)
dca4faeb
VP		 2308{
2309 int ret;
2310 struct filter_entry *f;
2311 struct adapter *adap;
2312 int i;
2313 u8 *val;
2314
2315 adap = netdev2adap(dev);
2316
1cab775c 2317 /* Adjust stid to correct filter index */
470c60c4 2318 stid -= adap->tids.sftid_base;
1cab775c
VP		 2319 stid += adap->tids.nftids;
2320
dca4faeb
VP		 2321 /* Check to make sure the filter requested is writable ...
2322 */
2323 f = &adap->tids.ftid_tab[stid];
2324 ret = writable_filter(f);
2325 if (ret)
2326 return ret;
2327
2328 /* Clear out any old resources being used by the filter before
2329 * we start constructing the new filter.
2330 */
2331 if (f->valid)
2332 clear_filter(adap, f);
2333
2334 /* Clear out filter specifications */
2335 memset(&f->fs, 0, sizeof(struct ch_filter_specification));
2336 f->fs.val.lport = cpu_to_be16(sport);
2337 f->fs.mask.lport = ~0;
2338 val = (u8 *)&sip;
793dad94 2339 if ((val[0] | val[1] | val[2] | val[3]) != 0) {
dca4faeb
VP		 2340 for (i = 0; i < 4; i++) {
2341 f->fs.val.lip[i] = val[i];
2342 f->fs.mask.lip[i] = ~0;
2343 }
0d804338 2344 if (adap->params.tp.vlan_pri_map & PORT_F) {
793dad94
VP		 2345 f->fs.val.iport = port;
2346 f->fs.mask.iport = mask;
2347 }
2348 }
dca4faeb 2349
0d804338 2350 if (adap->params.tp.vlan_pri_map & PROTOCOL_F) {
7c89e555
KS		 2351 f->fs.val.proto = IPPROTO_TCP;
2352 f->fs.mask.proto = ~0;
2353 }
2354
dca4faeb
VP		 2355 f->fs.dirsteer = 1;
2356 f->fs.iq = queue;
2357 /* Mark filter as locked */
2358 f->locked = 1;
2359 f->fs.rpttid = 1;
2360
6b254afd
GG		 2361 /* Save the actual tid. We need this to get the corresponding
2362 * filter entry structure in filter_rpl.
2363 */
2364 f->tid = stid + adap->tids.ftid_base;
dca4faeb
VP		 2365 ret = set_filter_wr(adap, stid);
2366 if (ret) {
2367 clear_filter(adap, f);
2368 return ret;
2369 }
2370
2371 return 0;
2372}
2373EXPORT_SYMBOL(cxgb4_create_server_filter);
2374
2375int cxgb4_remove_server_filter(const struct net_device *dev, unsigned int stid,
2376 unsigned int queue, bool ipv6)
2377{
dca4faeb
VP		 2378 struct filter_entry *f;
2379 struct adapter *adap;
2380
2381 adap = netdev2adap(dev);
1cab775c
VP		 2382
2383 /* Adjust stid to correct filter index */
470c60c4 2384 stid -= adap->tids.sftid_base;
1cab775c
VP		 2385 stid += adap->tids.nftids;
2386
dca4faeb
VP		 2387 f = &adap->tids.ftid_tab[stid];
2388 /* Unlock the filter */
2389 f->locked = 0;
2390
8c14846d 2391 return delete_filter(adap, stid);
dca4faeb
VP		 2392}
2393EXPORT_SYMBOL(cxgb4_remove_server_filter);
2394
bc1f4470 2395static void cxgb_get_stats(struct net_device *dev,
2396 struct rtnl_link_stats64 *ns)
b8ff05a9
DM		 2397{
2398 struct port_stats stats;
2399 struct port_info *p = netdev_priv(dev);
2400 struct adapter *adapter = p->adapter;
b8ff05a9 2401
9fe6cb58
GS		 2402 /* Block retrieving statistics during EEH error
2403 * recovery. Otherwise, the recovery might fail
2404 * and the PCI device will be removed permanently
2405 */
b8ff05a9 2406 spin_lock(&adapter->stats_lock);
9fe6cb58
GS		 2407 if (!netif_device_present(dev)) {
2408 spin_unlock(&adapter->stats_lock);
bc1f4470 2409 return;
9fe6cb58 2410 }
a4cfd929
HS		 2411 t4_get_port_stats_offset(adapter, p->tx_chan, &stats,
2412 &p->stats_base);
b8ff05a9
DM		 2413 spin_unlock(&adapter->stats_lock);
2414
2415 ns->tx_bytes = stats.tx_octets;
2416 ns->tx_packets = stats.tx_frames;
2417 ns->rx_bytes = stats.rx_octets;
2418 ns->rx_packets = stats.rx_frames;
2419 ns->multicast = stats.rx_mcast_frames;
2420
2421 /* detailed rx_errors */
2422 ns->rx_length_errors = stats.rx_jabber + stats.rx_too_long +
2423 stats.rx_runt;
2424 ns->rx_over_errors = 0;
2425 ns->rx_crc_errors = stats.rx_fcs_err;
2426 ns->rx_frame_errors = stats.rx_symbol_err;
b93f79be 2427 ns->rx_dropped = stats.rx_ovflow0 + stats.rx_ovflow1 +
b8ff05a9
DM		 2428 stats.rx_ovflow2 + stats.rx_ovflow3 +
2429 stats.rx_trunc0 + stats.rx_trunc1 +
2430 stats.rx_trunc2 + stats.rx_trunc3;
2431 ns->rx_missed_errors = 0;
2432
2433 /* detailed tx_errors */
2434 ns->tx_aborted_errors = 0;
2435 ns->tx_carrier_errors = 0;
2436 ns->tx_fifo_errors = 0;
2437 ns->tx_heartbeat_errors = 0;
2438 ns->tx_window_errors = 0;
2439
2440 ns->tx_errors = stats.tx_error_frames;
2441 ns->rx_errors = stats.rx_symbol_err + stats.rx_fcs_err +
2442 ns->rx_length_errors + stats.rx_len_err + ns->rx_fifo_errors;
b8ff05a9
DM		 2443}
2444
2445static int cxgb_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
2446{
060e0c75 2447 unsigned int mbox;
b8ff05a9
DM		 2448 int ret = 0, prtad, devad;
2449 struct port_info *pi = netdev_priv(dev);
a4569504 2450 struct adapter *adapter = pi->adapter;
b8ff05a9
DM		 2451 struct mii_ioctl_data *data = (struct mii_ioctl_data *)&req->ifr_data;
2452
2453 switch (cmd) {
2454 case SIOCGMIIPHY:
2455 if (pi->mdio_addr < 0)
2456 return -EOPNOTSUPP;
2457 data->phy_id = pi->mdio_addr;
2458 break;
2459 case SIOCGMIIREG:
2460 case SIOCSMIIREG:
2461 if (mdio_phy_id_is_c45(data->phy_id)) {
2462 prtad = mdio_phy_id_prtad(data->phy_id);
2463 devad = mdio_phy_id_devad(data->phy_id);
2464 } else if (data->phy_id < 32) {
2465 prtad = data->phy_id;
2466 devad = 0;
2467 data->reg_num &= 0x1f;
2468 } else
2469 return -EINVAL;
2470
b2612722 2471 mbox = pi->adapter->pf;
b8ff05a9 2472 if (cmd == SIOCGMIIREG)
060e0c75 2473 ret = t4_mdio_rd(pi->adapter, mbox, prtad, devad,
b8ff05a9
DM		 2474 data->reg_num, &data->val_out);
2475 else
060e0c75 2476 ret = t4_mdio_wr(pi->adapter, mbox, prtad, devad,
b8ff05a9
DM		 2477 data->reg_num, data->val_in);
2478 break;
5e2a5ebc
HS		 2479 case SIOCGHWTSTAMP:
2480 return copy_to_user(req->ifr_data, &pi->tstamp_config,
2481 sizeof(pi->tstamp_config)) ?
2482 -EFAULT : 0;
2483 case SIOCSHWTSTAMP:
2484 if (copy_from_user(&pi->tstamp_config, req->ifr_data,
2485 sizeof(pi->tstamp_config)))
2486 return -EFAULT;
2487
a4569504
AG		 2488 if (!is_t4(adapter->params.chip)) {
2489 switch (pi->tstamp_config.tx_type) {
2490 case HWTSTAMP_TX_OFF:
2491 case HWTSTAMP_TX_ON:
2492 break;
2493 default:
2494 return -ERANGE;
2495 }
2496
2497 switch (pi->tstamp_config.rx_filter) {
2498 case HWTSTAMP_FILTER_NONE:
2499 pi->rxtstamp = false;
2500 break;
2501 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
2502 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
2503 cxgb4_ptprx_timestamping(pi, pi->port_id,
2504 PTP_TS_L4);
2505 break;
2506 case HWTSTAMP_FILTER_PTP_V2_EVENT:
2507 cxgb4_ptprx_timestamping(pi, pi->port_id,
2508 PTP_TS_L2_L4);
2509 break;
2510 case HWTSTAMP_FILTER_ALL:
2511 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
2512 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
2513 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
2514 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
2515 pi->rxtstamp = true;
2516 break;
2517 default:
2518 pi->tstamp_config.rx_filter =
2519 HWTSTAMP_FILTER_NONE;
2520 return -ERANGE;
2521 }
2522
2523 if ((pi->tstamp_config.tx_type == HWTSTAMP_TX_OFF) &&
2524 (pi->tstamp_config.rx_filter ==
2525 HWTSTAMP_FILTER_NONE)) {
2526 if (cxgb4_ptp_txtype(adapter, pi->port_id) >= 0)
2527 pi->ptp_enable = false;
2528 }
2529
2530 if (pi->tstamp_config.rx_filter !=
2531 HWTSTAMP_FILTER_NONE) {
2532 if (cxgb4_ptp_redirect_rx_packet(adapter,
2533 pi) >= 0)
2534 pi->ptp_enable = true;
2535 }
2536 } else {
2537 /* For T4 Adapters */
2538 switch (pi->tstamp_config.rx_filter) {
2539 case HWTSTAMP_FILTER_NONE:
5e2a5ebc
HS		 2540 pi->rxtstamp = false;
2541 break;
a4569504 2542 case HWTSTAMP_FILTER_ALL:
5e2a5ebc
HS		 2543 pi->rxtstamp = true;
2544 break;
a4569504
AG		 2545 default:
2546 pi->tstamp_config.rx_filter =
2547 HWTSTAMP_FILTER_NONE;
5e2a5ebc 2548 return -ERANGE;
a4569504 2549 }
5e2a5ebc 2550 }
5e2a5ebc
HS		 2551 return copy_to_user(req->ifr_data, &pi->tstamp_config,
2552 sizeof(pi->tstamp_config)) ?
2553 -EFAULT : 0;
b8ff05a9
DM		 2554 default:
2555 return -EOPNOTSUPP;
2556 }
2557 return ret;
2558}
2559
2560static void cxgb_set_rxmode(struct net_device *dev)
2561{
2562 /* unfortunately we can't return errors to the stack */
2563 set_rxmode(dev, -1, false);
2564}
2565
2566static int cxgb_change_mtu(struct net_device *dev, int new_mtu)
2567{
2568 int ret;
2569 struct port_info *pi = netdev_priv(dev);
2570
b2612722 2571 ret = t4_set_rxmode(pi->adapter, pi->adapter->pf, pi->viid, new_mtu, -1,
060e0c75 2572 -1, -1, -1, true);
b8ff05a9
DM		 2573 if (!ret)
2574 dev->mtu = new_mtu;
2575 return ret;
2576}
2577
858aa65c 2578#ifdef CONFIG_PCI_IOV
e7b48a32
HS		 2579static int dummy_open(struct net_device *dev)
2580{
2581 /* Turn carrier off since we don't have to transmit anything on this
2582 * interface.
2583 */
2584 netif_carrier_off(dev);
2585 return 0;
2586}
2587
661dbeb9
HS		 2588/* Fill MAC address that will be assigned by the FW */
2589static void fill_vf_station_mac_addr(struct adapter *adap)
2590{
2591 unsigned int i;
2592 u8 hw_addr[ETH_ALEN], macaddr[ETH_ALEN];
2593 int err;
2594 u8 *na;
2595 u16 a, b;
2596
2597 err = t4_get_raw_vpd_params(adap, &adap->params.vpd);
2598 if (!err) {
2599 na = adap->params.vpd.na;
2600 for (i = 0; i < ETH_ALEN; i++)
2601 hw_addr[i] = (hex2val(na[2 * i + 0]) * 16 +
2602 hex2val(na[2 * i + 1]));
2603 a = (hw_addr[0] << 8) | hw_addr[1];
2604 b = (hw_addr[1] << 8) | hw_addr[2];
2605 a ^= b;
2606 a |= 0x0200; /* locally assigned Ethernet MAC address */
2607 a &= ~0x0100; /* not a multicast Ethernet MAC address */
2608 macaddr[0] = a >> 8;
2609 macaddr[1] = a & 0xff;
2610
2611 for (i = 2; i < 5; i++)
2612 macaddr[i] = hw_addr[i + 1];
2613
2614 for (i = 0; i < adap->num_vfs; i++) {
2615 macaddr[5] = adap->pf * 16 + i;
2616 ether_addr_copy(adap->vfinfo[i].vf_mac_addr, macaddr);
2617 }
2618 }
2619}
2620
858aa65c
HS		 2621static int cxgb_set_vf_mac(struct net_device *dev, int vf, u8 *mac)
2622{
2623 struct port_info *pi = netdev_priv(dev);
2624 struct adapter *adap = pi->adapter;
661dbeb9 2625 int ret;
858aa65c
HS		 2626
2627 /* verify MAC addr is valid */
2628 if (!is_valid_ether_addr(mac)) {
2629 dev_err(pi->adapter->pdev_dev,
2630 "Invalid Ethernet address %pM for VF %d\n",
2631 mac, vf);
2632 return -EINVAL;
2633 }
2634
2635 dev_info(pi->adapter->pdev_dev,
2636 "Setting MAC %pM on VF %d\n", mac, vf);
661dbeb9
HS		 2637 ret = t4_set_vf_mac_acl(adap, vf + 1, 1, mac);
2638 if (!ret)
2639 ether_addr_copy(adap->vfinfo[vf].vf_mac_addr, mac);
2640 return ret;
2641}
2642
2643static int cxgb_get_vf_config(struct net_device *dev,
2644 int vf, struct ifla_vf_info *ivi)
2645{
2646 struct port_info *pi = netdev_priv(dev);
2647 struct adapter *adap = pi->adapter;
2648
2649 if (vf >= adap->num_vfs)
2650 return -EINVAL;
2651 ivi->vf = vf;
8ea4fae9
GG		 2652 ivi->max_tx_rate = adap->vfinfo[vf].tx_rate;
2653 ivi->min_tx_rate = 0;
661dbeb9
HS		 2654 ether_addr_copy(ivi->mac, adap->vfinfo[vf].vf_mac_addr);
2655 return 0;
858aa65c 2656}
96fe11f2
GG		 2657
2658static int cxgb_get_phys_port_id(struct net_device *dev,
2659 struct netdev_phys_item_id *ppid)
2660{
2661 struct port_info *pi = netdev_priv(dev);
2662 unsigned int phy_port_id;
2663
2664 phy_port_id = pi->adapter->adap_idx * 10 + pi->port_id;
2665 ppid->id_len = sizeof(phy_port_id);
2666 memcpy(ppid->id, &phy_port_id, ppid->id_len);
2667 return 0;
2668}
2669
8ea4fae9
GG		 2670static int cxgb_set_vf_rate(struct net_device *dev, int vf, int min_tx_rate,
2671 int max_tx_rate)
2672{
2673 struct port_info *pi = netdev_priv(dev);
2674 struct adapter *adap = pi->adapter;
2675 struct fw_port_cmd port_cmd, port_rpl;
2676 u32 link_status, speed = 0;
2677 u32 fw_pfvf, fw_class;
2678 int class_id = vf;
2679 int link_ok, ret;
2680 u16 pktsize;
2681
2682 if (vf >= adap->num_vfs)
2683 return -EINVAL;
2684
2685 if (min_tx_rate) {
2686 dev_err(adap->pdev_dev,
2687 "Min tx rate (%d) (> 0) for VF %d is Invalid.\n",
2688 min_tx_rate, vf);
2689 return -EINVAL;
2690 }
2691 /* Retrieve link details for VF port */
2692 memset(&port_cmd, 0, sizeof(port_cmd));
2693 port_cmd.op_to_portid = cpu_to_be32(FW_CMD_OP_V(FW_PORT_CMD) |
2694 FW_CMD_REQUEST_F |
2695 FW_CMD_READ_F |
2696 FW_PORT_CMD_PORTID_V(pi->port_id));
2697 port_cmd.action_to_len16 =
2698 cpu_to_be32(FW_PORT_CMD_ACTION_V(FW_PORT_ACTION_GET_PORT_INFO) |
2699 FW_LEN16(port_cmd));
2700 ret = t4_wr_mbox(adap, adap->mbox, &port_cmd, sizeof(port_cmd),
2701 &port_rpl);
2702 if (ret != FW_SUCCESS) {
2703 dev_err(adap->pdev_dev,
2704 "Failed to get link status for VF %d\n", vf);
2705 return -EINVAL;
2706 }
2707 link_status = be32_to_cpu(port_rpl.u.info.lstatus_to_modtype);
2708 link_ok = (link_status & FW_PORT_CMD_LSTATUS_F) != 0;
2709 if (!link_ok) {
2710 dev_err(adap->pdev_dev, "Link down for VF %d\n", vf);
2711 return -EINVAL;
2712 }
2713 /* Determine link speed */
2714 if (link_status & FW_PORT_CMD_LSPEED_V(FW_PORT_CAP_SPEED_100M))
2715 speed = 100;
2716 else if (link_status & FW_PORT_CMD_LSPEED_V(FW_PORT_CAP_SPEED_1G))
2717 speed = 1000;
2718 else if (link_status & FW_PORT_CMD_LSPEED_V(FW_PORT_CAP_SPEED_10G))
2719 speed = 10000;
2720 else if (link_status & FW_PORT_CMD_LSPEED_V(FW_PORT_CAP_SPEED_25G))
2721 speed = 25000;
2722 else if (link_status & FW_PORT_CMD_LSPEED_V(FW_PORT_CAP_SPEED_40G))
2723 speed = 40000;
2724 else if (link_status & FW_PORT_CMD_LSPEED_V(FW_PORT_CAP_SPEED_100G))
2725 speed = 100000;
2726
2727 if (max_tx_rate > speed) {
2728 dev_err(adap->pdev_dev,
2729 "Max tx rate %d for VF %d can't be > link-speed %u",
2730 max_tx_rate, vf, speed);
2731 return -EINVAL;
2732 }
2733 pktsize = be16_to_cpu(port_rpl.u.info.mtu);
2734 /* subtract ethhdr size and 4 bytes crc since, f/w appends it */
2735 pktsize = pktsize - sizeof(struct ethhdr) - 4;
2736 /* subtract ipv4 hdr size, tcp hdr size to get typical IPv4 MSS size */
2737 pktsize = pktsize - sizeof(struct iphdr) - sizeof(struct tcphdr);
2738 /* configure Traffic Class for rate-limiting */
2739 ret = t4_sched_params(adap, SCHED_CLASS_TYPE_PACKET,
2740 SCHED_CLASS_LEVEL_CL_RL,
2741 SCHED_CLASS_MODE_CLASS,
2742 SCHED_CLASS_RATEUNIT_BITS,
2743 SCHED_CLASS_RATEMODE_ABS,
2744 pi->port_id, class_id, 0,
2745 max_tx_rate * 1000, 0, pktsize);
2746 if (ret) {
2747 dev_err(adap->pdev_dev, "Err %d for Traffic Class config\n",
2748 ret);
2749 return -EINVAL;
2750 }
2751 dev_info(adap->pdev_dev,
2752 "Class %d with MSS %u configured with rate %u\n",
2753 class_id, pktsize, max_tx_rate);
2754
2755 /* bind VF to configured Traffic Class */
2756 fw_pfvf = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) |
2757 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_SCHEDCLASS_ETH));
2758 fw_class = class_id;
2759 ret = t4_set_params(adap, adap->mbox, adap->pf, vf + 1, 1, &fw_pfvf,
2760 &fw_class);
2761 if (ret) {
2762 dev_err(adap->pdev_dev,
2763 "Err %d in binding VF %d to Traffic Class %d\n",
2764 ret, vf, class_id);
2765 return -EINVAL;
2766 }
2767 dev_info(adap->pdev_dev, "PF %d VF %d is bound to Class %d\n",
2768 adap->pf, vf, class_id);
2769 adap->vfinfo[vf].tx_rate = max_tx_rate;
2770 return 0;
2771}
2772
858aa65c
HS		 2773#endif
2774
b8ff05a9
DM		 2775static int cxgb_set_mac_addr(struct net_device *dev, void *p)
2776{
2777 int ret;
2778 struct sockaddr *addr = p;
2779 struct port_info *pi = netdev_priv(dev);
2780
2781 if (!is_valid_ether_addr(addr->sa_data))
504f9b5a 2782 return -EADDRNOTAVAIL;
b8ff05a9 2783
b2612722 2784 ret = t4_change_mac(pi->adapter, pi->adapter->pf, pi->viid,
060e0c75 2785 pi->xact_addr_filt, addr->sa_data, true, true);
b8ff05a9
DM		 2786 if (ret < 0)
2787 return ret;
2788
2789 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
2790 pi->xact_addr_filt = ret;
2791 return 0;
2792}
2793
b8ff05a9
DM		 2794#ifdef CONFIG_NET_POLL_CONTROLLER
2795static void cxgb_netpoll(struct net_device *dev)
2796{
2797 struct port_info *pi = netdev_priv(dev);
2798 struct adapter *adap = pi->adapter;
2799
2800 if (adap->flags & USING_MSIX) {
2801 int i;
2802 struct sge_eth_rxq *rx = &adap->sge.ethrxq[pi->first_qset];
2803
2804 for (i = pi->nqsets; i; i--, rx++)
2805 t4_sge_intr_msix(0, &rx->rspq);
2806 } else
2807 t4_intr_handler(adap)(0, adap);
2808}
2809#endif
2810
10a2604e
RL		 2811static int cxgb_set_tx_maxrate(struct net_device *dev, int index, u32 rate)
2812{
2813 struct port_info *pi = netdev_priv(dev);
2814 struct adapter *adap = pi->adapter;
2815 struct sched_class *e;
2816 struct ch_sched_params p;
2817 struct ch_sched_queue qe;
2818 u32 req_rate;
2819 int err = 0;
2820
2821 if (!can_sched(dev))
2822 return -ENOTSUPP;
2823
2824 if (index < 0 || index > pi->nqsets - 1)
2825 return -EINVAL;
2826
2827 if (!(adap->flags & FULL_INIT_DONE)) {
2828 dev_err(adap->pdev_dev,
2829 "Failed to rate limit on queue %d. Link Down?\n",
2830 index);
2831 return -EINVAL;
2832 }
2833
2834 /* Convert from Mbps to Kbps */
2835 req_rate = rate << 10;
2836
2837 /* Max rate is 10 Gbps */
2838 if (req_rate >= SCHED_MAX_RATE_KBPS) {
2839 dev_err(adap->pdev_dev,
2840 "Invalid rate %u Mbps, Max rate is %u Gbps\n",
2841 rate, SCHED_MAX_RATE_KBPS);
2842 return -ERANGE;
2843 }
2844
2845 /* First unbind the queue from any existing class */
2846 memset(&qe, 0, sizeof(qe));
2847 qe.queue = index;
2848 qe.class = SCHED_CLS_NONE;
2849
2850 err = cxgb4_sched_class_unbind(dev, (void *)(&qe), SCHED_QUEUE);
2851 if (err) {
2852 dev_err(adap->pdev_dev,
2853 "Unbinding Queue %d on port %d fail. Err: %d\n",
2854 index, pi->port_id, err);
2855 return err;
2856 }
2857
2858 /* Queue already unbound */
2859 if (!req_rate)
2860 return 0;
2861
2862 /* Fetch any available unused or matching scheduling class */
2863 memset(&p, 0, sizeof(p));
2864 p.type = SCHED_CLASS_TYPE_PACKET;
2865 p.u.params.level = SCHED_CLASS_LEVEL_CL_RL;
2866 p.u.params.mode = SCHED_CLASS_MODE_CLASS;
2867 p.u.params.rateunit = SCHED_CLASS_RATEUNIT_BITS;
2868 p.u.params.ratemode = SCHED_CLASS_RATEMODE_ABS;
2869 p.u.params.channel = pi->tx_chan;
2870 p.u.params.class = SCHED_CLS_NONE;
2871 p.u.params.minrate = 0;
2872 p.u.params.maxrate = req_rate;
2873 p.u.params.weight = 0;
2874 p.u.params.pktsize = dev->mtu;
2875
2876 e = cxgb4_sched_class_alloc(dev, &p);
2877 if (!e)
2878 return -ENOMEM;
2879
2880 /* Bind the queue to a scheduling class */
2881 memset(&qe, 0, sizeof(qe));
2882 qe.queue = index;
2883 qe.class = e->idx;
2884
2885 err = cxgb4_sched_class_bind(dev, (void *)(&qe), SCHED_QUEUE);
2886 if (err)
2887 dev_err(adap->pdev_dev,
2888 "Queue rate limiting failed. Err: %d\n", err);
2889 return err;
2890}
2891
2572ac53
JP		 2892static int cxgb_setup_tc(struct net_device *dev, enum tc_setup_type type,
2893 u32 handle, u32 chain_index, __be16 proto,
2894 struct tc_to_netdev *tc)
d8931847
RL		 2895{
2896 struct port_info *pi = netdev2pinfo(dev);
2897 struct adapter *adap = netdev2adap(dev);
2898
a5fcf8a6
JP		 2899 if (chain_index)
2900 return -EOPNOTSUPP;
2901
d8931847
RL		 2902 if (!(adap->flags & FULL_INIT_DONE)) {
2903 dev_err(adap->pdev_dev,
2904 "Failed to setup tc on port %d. Link Down?\n",
2905 pi->port_id);
2906 return -EINVAL;
2907 }
2908
2909 if (TC_H_MAJ(handle) == TC_H_MAJ(TC_H_INGRESS) &&
2572ac53 2910 type == TC_SETUP_CLSU32) {
d8931847
RL		 2911 switch (tc->cls_u32->command) {
2912 case TC_CLSU32_NEW_KNODE:
2913 case TC_CLSU32_REPLACE_KNODE:
2914 return cxgb4_config_knode(dev, proto, tc->cls_u32);
2915 case TC_CLSU32_DELETE_KNODE:
2916 return cxgb4_delete_knode(dev, proto, tc->cls_u32);
2917 default:
2918 return -EOPNOTSUPP;
2919 }
2920 }
2921
2922 return -EOPNOTSUPP;
2923}
2924
90592b9a
AV		 2925static netdev_features_t cxgb_fix_features(struct net_device *dev,
2926 netdev_features_t features)
2927{
2928 /* Disable GRO, if RX_CSUM is disabled */
2929 if (!(features & NETIF_F_RXCSUM))
2930 features &= ~NETIF_F_GRO;
2931
2932 return features;
2933}
2934
b8ff05a9
DM		 2935static const struct net_device_ops cxgb4_netdev_ops = {
2936 .ndo_open = cxgb_open,
2937 .ndo_stop = cxgb_close,
2938 .ndo_start_xmit = t4_eth_xmit,
688848b1 2939 .ndo_select_queue = cxgb_select_queue,
9be793bf 2940 .ndo_get_stats64 = cxgb_get_stats,
b8ff05a9
DM		 2941 .ndo_set_rx_mode = cxgb_set_rxmode,
2942 .ndo_set_mac_address = cxgb_set_mac_addr,
2ed28baa 2943 .ndo_set_features = cxgb_set_features,
b8ff05a9
DM		 2944 .ndo_validate_addr = eth_validate_addr,
2945 .ndo_do_ioctl = cxgb_ioctl,
2946 .ndo_change_mtu = cxgb_change_mtu,
b8ff05a9
DM		 2947#ifdef CONFIG_NET_POLL_CONTROLLER
2948 .ndo_poll_controller = cxgb_netpoll,
2949#endif
84a200b3
VP		 2950#ifdef CONFIG_CHELSIO_T4_FCOE
2951 .ndo_fcoe_enable = cxgb_fcoe_enable,
2952 .ndo_fcoe_disable = cxgb_fcoe_disable,
2953#endif /* CONFIG_CHELSIO_T4_FCOE */
10a2604e 2954 .ndo_set_tx_maxrate = cxgb_set_tx_maxrate,
d8931847 2955 .ndo_setup_tc = cxgb_setup_tc,
90592b9a 2956 .ndo_fix_features = cxgb_fix_features,
b8ff05a9
DM		 2957};
2958
858aa65c 2959#ifdef CONFIG_PCI_IOV
e7b48a32
HS		 2960static const struct net_device_ops cxgb4_mgmt_netdev_ops = {
2961 .ndo_open = dummy_open,
858aa65c 2962 .ndo_set_vf_mac = cxgb_set_vf_mac,
661dbeb9 2963 .ndo_get_vf_config = cxgb_get_vf_config,
8ea4fae9 2964 .ndo_set_vf_rate = cxgb_set_vf_rate,
96fe11f2 2965 .ndo_get_phys_port_id = cxgb_get_phys_port_id,
7829451c 2966};
e7b48a32 2967#endif
7829451c
HS		 2968
2969static void get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
2970{
2971 struct adapter *adapter = netdev2adap(dev);
2972
2973 strlcpy(info->driver, cxgb4_driver_name, sizeof(info->driver));
2974 strlcpy(info->version, cxgb4_driver_version,
2975 sizeof(info->version));
2976 strlcpy(info->bus_info, pci_name(adapter->pdev),
2977 sizeof(info->bus_info));
2978}
2979
2980static const struct ethtool_ops cxgb4_mgmt_ethtool_ops = {
2981 .get_drvinfo = get_drvinfo,
2982};
2983
b8ff05a9
DM		 2984void t4_fatal_err(struct adapter *adap)
2985{
3be0679b
HS		 2986 int port;
2987
025d0973
GP		 2988 if (pci_channel_offline(adap->pdev))
2989 return;
2990
3be0679b
HS		 2991 /* Disable the SGE since ULDs are going to free resources that
2992 * could be exposed to the adapter. RDMA MWs for example...
2993 */
2994 t4_shutdown_adapter(adap);
2995 for_each_port(adap, port) {
2996 struct net_device *dev = adap->port[port];
2997
2998 /* If we get here in very early initialization the network
2999 * devices may not have been set up yet.
3000 */
3001 if (!dev)
3002 continue;
3003
3004 netif_tx_stop_all_queues(dev);
3005 netif_carrier_off(dev);
3006 }
b8ff05a9
DM		 3007 dev_alert(adap->pdev_dev, "encountered fatal error, adapter stopped\n");
3008}
3009
3010static void setup_memwin(struct adapter *adap)
3011{
b562fc37 3012 u32 nic_win_base = t4_get_util_window(adap);
b8ff05a9 3013
b562fc37 3014 t4_setup_memwin(adap, nic_win_base, MEMWIN_NIC);
636f9d37
VP		 3015}
3016
3017static void setup_memwin_rdma(struct adapter *adap)
3018{
1ae970e0 3019 if (adap->vres.ocq.size) {
0abfd152
HS		 3020 u32 start;
3021 unsigned int sz_kb;
1ae970e0 3022
0abfd152
HS		 3023 start = t4_read_pcie_cfg4(adap, PCI_BASE_ADDRESS_2);
3024 start &= PCI_BASE_ADDRESS_MEM_MASK;
3025 start += OCQ_WIN_OFFSET(adap->pdev, &adap->vres);
1ae970e0
DM		 3026 sz_kb = roundup_pow_of_two(adap->vres.ocq.size) >> 10;
3027 t4_write_reg(adap,
f061de42
HS		 3028 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_BASE_WIN_A, 3),
3029 start | BIR_V(1) | WINDOW_V(ilog2(sz_kb)));
1ae970e0 3030 t4_write_reg(adap,
f061de42 3031 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET_A, 3),
1ae970e0
DM		 3032 adap->vres.ocq.start);
3033 t4_read_reg(adap,
f061de42 3034 PCIE_MEM_ACCESS_REG(PCIE_MEM_ACCESS_OFFSET_A, 3));
1ae970e0 3035 }
b8ff05a9
DM		 3036}
3037
02b5fb8e
DM		 3038static int adap_init1(struct adapter *adap, struct fw_caps_config_cmd *c)
3039{
3040 u32 v;
3041 int ret;
3042
3043 /* get device capabilities */
3044 memset(c, 0, sizeof(*c));
e2ac9628
HS		 3045 c->op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3046 FW_CMD_REQUEST_F | FW_CMD_READ_F);
ce91a923 3047 c->cfvalid_to_len16 = htonl(FW_LEN16(*c));
b2612722 3048 ret = t4_wr_mbox(adap, adap->mbox, c, sizeof(*c), c);
02b5fb8e
DM		 3049 if (ret < 0)
3050 return ret;
3051
e2ac9628
HS		 3052 c->op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3053 FW_CMD_REQUEST_F | FW_CMD_WRITE_F);
b2612722 3054 ret = t4_wr_mbox(adap, adap->mbox, c, sizeof(*c), NULL);
02b5fb8e
DM		 3055 if (ret < 0)
3056 return ret;
3057
b2612722 3058 ret = t4_config_glbl_rss(adap, adap->pf,
02b5fb8e 3059 FW_RSS_GLB_CONFIG_CMD_MODE_BASICVIRTUAL,
b2e1a3f0
HS		 3060 FW_RSS_GLB_CONFIG_CMD_TNLMAPEN_F |
3061 FW_RSS_GLB_CONFIG_CMD_TNLALLLKP_F);
02b5fb8e
DM		 3062 if (ret < 0)
3063 return ret;
3064
b2612722 3065 ret = t4_cfg_pfvf(adap, adap->mbox, adap->pf, 0, adap->sge.egr_sz, 64,
4b8e27a8
HS		 3066 MAX_INGQ, 0, 0, 4, 0xf, 0xf, 16, FW_CMD_CAP_PF,
3067 FW_CMD_CAP_PF);
02b5fb8e
DM		 3068 if (ret < 0)
3069 return ret;
3070
3071 t4_sge_init(adap);
3072
02b5fb8e 3073 /* tweak some settings */
837e4a42 3074 t4_write_reg(adap, TP_SHIFT_CNT_A, 0x64f8849);
0d804338 3075 t4_write_reg(adap, ULP_RX_TDDP_PSZ_A, HPZ0_V(PAGE_SHIFT - 12));
837e4a42
HS		 3076 t4_write_reg(adap, TP_PIO_ADDR_A, TP_INGRESS_CONFIG_A);
3077 v = t4_read_reg(adap, TP_PIO_DATA_A);
3078 t4_write_reg(adap, TP_PIO_DATA_A, v & ~CSUM_HAS_PSEUDO_HDR_F);
060e0c75 3079
dca4faeb
VP		 3080 /* first 4 Tx modulation queues point to consecutive Tx channels */
3081 adap->params.tp.tx_modq_map = 0xE4;
0d804338
HS		 3082 t4_write_reg(adap, TP_TX_MOD_QUEUE_REQ_MAP_A,
3083 TX_MOD_QUEUE_REQ_MAP_V(adap->params.tp.tx_modq_map));
dca4faeb
VP		 3084
3085 /* associate each Tx modulation queue with consecutive Tx channels */
3086 v = 0x84218421;
837e4a42 3087 t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
0d804338 3088 &v, 1, TP_TX_SCHED_HDR_A);
837e4a42 3089 t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
0d804338 3090 &v, 1, TP_TX_SCHED_FIFO_A);
837e4a42 3091 t4_write_indirect(adap, TP_PIO_ADDR_A, TP_PIO_DATA_A,
0d804338 3092 &v, 1, TP_TX_SCHED_PCMD_A);
dca4faeb
VP		 3093
3094#define T4_TX_MODQ_10G_WEIGHT_DEFAULT 16 /* in KB units */
3095 if (is_offload(adap)) {
0d804338
HS		 3096 t4_write_reg(adap, TP_TX_MOD_QUEUE_WEIGHT0_A,
3097 TX_MODQ_WEIGHT0_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3098 TX_MODQ_WEIGHT1_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3099 TX_MODQ_WEIGHT2_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3100 TX_MODQ_WEIGHT3_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
3101 t4_write_reg(adap, TP_TX_MOD_CHANNEL_WEIGHT_A,
3102 TX_MODQ_WEIGHT0_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3103 TX_MODQ_WEIGHT1_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3104 TX_MODQ_WEIGHT2_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT) |
3105 TX_MODQ_WEIGHT3_V(T4_TX_MODQ_10G_WEIGHT_DEFAULT));
dca4faeb
VP		 3106 }
3107
060e0c75 3108 /* get basic stuff going */
b2612722 3109 return t4_early_init(adap, adap->pf);
02b5fb8e
DM		 3110}
3111
b8ff05a9
DM		 3112/*
3113 * Max # of ATIDs. The absolute HW max is 16K but we keep it lower.
3114 */
3115#define MAX_ATIDS 8192U
3116
636f9d37
VP		 3117/*
3118 * Phase 0 of initialization: contact FW, obtain config, perform basic init.
3119 *
3120 * If the firmware we're dealing with has Configuration File support, then
3121 * we use that to perform all configuration
3122 */
3123
3124/*
3125 * Tweak configuration based on module parameters, etc. Most of these have
3126 * defaults assigned to them by Firmware Configuration Files (if we're using
3127 * them) but need to be explicitly set if we're using hard-coded
3128 * initialization. But even in the case of using Firmware Configuration
3129 * Files, we'd like to expose the ability to change these via module
3130 * parameters so these are essentially common tweaks/settings for
3131 * Configuration Files and hard-coded initialization ...
3132 */
3133static int adap_init0_tweaks(struct adapter *adapter)
3134{
3135 /*
3136 * Fix up various Host-Dependent Parameters like Page Size, Cache
3137 * Line Size, etc. The firmware default is for a 4KB Page Size and
3138 * 64B Cache Line Size ...
3139 */
3140 t4_fixup_host_params(adapter, PAGE_SIZE, L1_CACHE_BYTES);
3141
3142 /*
3143 * Process module parameters which affect early initialization.
3144 */
3145 if (rx_dma_offset != 2 && rx_dma_offset != 0) {
3146 dev_err(&adapter->pdev->dev,
3147 "Ignoring illegal rx_dma_offset=%d, using 2\n",
3148 rx_dma_offset);
3149 rx_dma_offset = 2;
3150 }
f612b815
HS		 3151 t4_set_reg_field(adapter, SGE_CONTROL_A,
3152 PKTSHIFT_V(PKTSHIFT_M),
3153 PKTSHIFT_V(rx_dma_offset));
636f9d37
VP		 3154
3155 /*
3156 * Don't include the "IP Pseudo Header" in CPL_RX_PKT checksums: Linux
3157 * adds the pseudo header itself.
3158 */
837e4a42
HS		 3159 t4_tp_wr_bits_indirect(adapter, TP_INGRESS_CONFIG_A,
3160 CSUM_HAS_PSEUDO_HDR_F, 0);
636f9d37
VP		 3161
3162 return 0;
3163}
3164
01b69614
HS		 3165/* 10Gb/s-BT PHY Support. chip-external 10Gb/s-BT PHYs are complex chips
3166 * unto themselves and they contain their own firmware to perform their
3167 * tasks ...
3168 */
3169static int phy_aq1202_version(const u8 *phy_fw_data,
3170 size_t phy_fw_size)
3171{
3172 int offset;
3173
3174 /* At offset 0x8 you're looking for the primary image's
3175 * starting offset which is 3 Bytes wide
3176 *
3177 * At offset 0xa of the primary image, you look for the offset
3178 * of the DRAM segment which is 3 Bytes wide.
3179 *
3180 * The FW version is at offset 0x27e of the DRAM and is 2 Bytes
3181 * wide
3182 */
3183 #define be16(__p) (((__p)[0] << 8) | (__p)[1])
3184 #define le16(__p) ((__p)[0] | ((__p)[1] << 8))
3185 #define le24(__p) (le16(__p) | ((__p)[2] << 16))
3186
3187 offset = le24(phy_fw_data + 0x8) << 12;
3188 offset = le24(phy_fw_data + offset + 0xa);
3189 return be16(phy_fw_data + offset + 0x27e);
3190
3191 #undef be16
3192 #undef le16
3193 #undef le24
3194}
3195
3196static struct info_10gbt_phy_fw {
3197 unsigned int phy_fw_id; /* PCI Device ID */
3198 char *phy_fw_file; /* /lib/firmware/ PHY Firmware file */
3199 int (*phy_fw_version)(const u8 *phy_fw_data, size_t phy_fw_size);
3200 int phy_flash; /* Has FLASH for PHY Firmware */
3201} phy_info_array[] = {
3202 {
3203 PHY_AQ1202_DEVICEID,
3204 PHY_AQ1202_FIRMWARE,
3205 phy_aq1202_version,
3206 1,
3207 },
3208 {
3209 PHY_BCM84834_DEVICEID,
3210 PHY_BCM84834_FIRMWARE,
3211 NULL,
3212 0,
3213 },
3214 { 0, NULL, NULL },
3215};
3216
3217static struct info_10gbt_phy_fw *find_phy_info(int devid)
3218{
3219 int i;
3220
3221 for (i = 0; i < ARRAY_SIZE(phy_info_array); i++) {
3222 if (phy_info_array[i].phy_fw_id == devid)
3223 return &phy_info_array[i];
3224 }
3225 return NULL;
3226}
3227
3228/* Handle updating of chip-external 10Gb/s-BT PHY firmware. This needs to
3229 * happen after the FW_RESET_CMD but before the FW_INITIALIZE_CMD. On error
3230 * we return a negative error number. If we transfer new firmware we return 1
3231 * (from t4_load_phy_fw()). If we don't do anything we return 0.
3232 */
3233static int adap_init0_phy(struct adapter *adap)
3234{
3235 const struct firmware *phyf;
3236 int ret;
3237 struct info_10gbt_phy_fw *phy_info;
3238
3239 /* Use the device ID to determine which PHY file to flash.
3240 */
3241 phy_info = find_phy_info(adap->pdev->device);
3242 if (!phy_info) {
3243 dev_warn(adap->pdev_dev,
3244 "No PHY Firmware file found for this PHY\n");
3245 return -EOPNOTSUPP;
3246 }
3247
3248 /* If we have a T4 PHY firmware file under /lib/firmware/cxgb4/, then
3249 * use that. The adapter firmware provides us with a memory buffer
3250 * where we can load a PHY firmware file from the host if we want to
3251 * override the PHY firmware File in flash.
3252 */
3253 ret = request_firmware_direct(&phyf, phy_info->phy_fw_file,
3254 adap->pdev_dev);
3255 if (ret < 0) {
3256 /* For adapters without FLASH attached to PHY for their
3257 * firmware, it's obviously a fatal error if we can't get the
3258 * firmware to the adapter. For adapters with PHY firmware
3259 * FLASH storage, it's worth a warning if we can't find the
3260 * PHY Firmware but we'll neuter the error ...
3261 */
3262 dev_err(adap->pdev_dev, "unable to find PHY Firmware image "
3263 "/lib/firmware/%s, error %d\n",
3264 phy_info->phy_fw_file, -ret);
3265 if (phy_info->phy_flash) {
3266 int cur_phy_fw_ver = 0;
3267
3268 t4_phy_fw_ver(adap, &cur_phy_fw_ver);
3269 dev_warn(adap->pdev_dev, "continuing with, on-adapter "
3270 "FLASH copy, version %#x\n", cur_phy_fw_ver);
3271 ret = 0;
3272 }
3273
3274 return ret;
3275 }
3276
3277 /* Load PHY Firmware onto adapter.
3278 */
3279 ret = t4_load_phy_fw(adap, MEMWIN_NIC, &adap->win0_lock,
3280 phy_info->phy_fw_version,
3281 (u8 *)phyf->data, phyf->size);
3282 if (ret < 0)
3283 dev_err(adap->pdev_dev, "PHY Firmware transfer error %d\n",
3284 -ret);
3285 else if (ret > 0) {
3286 int new_phy_fw_ver = 0;
3287
3288 if (phy_info->phy_fw_version)
3289 new_phy_fw_ver = phy_info->phy_fw_version(phyf->data,
3290 phyf->size);
3291 dev_info(adap->pdev_dev, "Successfully transferred PHY "
3292 "Firmware /lib/firmware/%s, version %#x\n",
3293 phy_info->phy_fw_file, new_phy_fw_ver);
3294 }
3295
3296 release_firmware(phyf);
3297
3298 return ret;
3299}
3300
636f9d37
VP		 3301/*
3302 * Attempt to initialize the adapter via a Firmware Configuration File.
3303 */
3304static int adap_init0_config(struct adapter *adapter, int reset)
3305{
3306 struct fw_caps_config_cmd caps_cmd;
3307 const struct firmware *cf;
3308 unsigned long mtype = 0, maddr = 0;
3309 u32 finiver, finicsum, cfcsum;
16e47624
HS		 3310 int ret;
3311 int config_issued = 0;
0a57a536 3312 char *fw_config_file, fw_config_file_path[256];
16e47624 3313 char *config_name = NULL;
636f9d37
VP		 3314
3315 /*
3316 * Reset device if necessary.
3317 */
3318 if (reset) {
3319 ret = t4_fw_reset(adapter, adapter->mbox,
0d804338 3320 PIORSTMODE_F | PIORST_F);
636f9d37
VP		 3321 if (ret < 0)
3322 goto bye;
3323 }
3324
01b69614
HS		 3325 /* If this is a 10Gb/s-BT adapter make sure the chip-external
3326 * 10Gb/s-BT PHYs have up-to-date firmware. Note that this step needs
3327 * to be performed after any global adapter RESET above since some
3328 * PHYs only have local RAM copies of the PHY firmware.
3329 */
3330 if (is_10gbt_device(adapter->pdev->device)) {
3331 ret = adap_init0_phy(adapter);
3332 if (ret < 0)
3333 goto bye;
3334 }
636f9d37
VP		 3335 /*
3336 * If we have a T4 configuration file under /lib/firmware/cxgb4/,
3337 * then use that. Otherwise, use the configuration file stored
3338 * in the adapter flash ...
3339 */
d14807dd 3340 switch (CHELSIO_CHIP_VERSION(adapter->params.chip)) {
0a57a536 3341 case CHELSIO_T4:
16e47624 3342 fw_config_file = FW4_CFNAME;
0a57a536
SR		 3343 break;
3344 case CHELSIO_T5:
3345 fw_config_file = FW5_CFNAME;
3346 break;
3ccc6cf7
HS		 3347 case CHELSIO_T6:
3348 fw_config_file = FW6_CFNAME;
3349 break;
0a57a536
SR		 3350 default:
3351 dev_err(adapter->pdev_dev, "Device %d is not supported\n",
3352 adapter->pdev->device);
3353 ret = -EINVAL;
3354 goto bye;
3355 }
3356
3357 ret = request_firmware(&cf, fw_config_file, adapter->pdev_dev);
636f9d37 3358 if (ret < 0) {
16e47624 3359 config_name = "On FLASH";
636f9d37
VP		 3360 mtype = FW_MEMTYPE_CF_FLASH;
3361 maddr = t4_flash_cfg_addr(adapter);
3362 } else {
3363 u32 params[7], val[7];
3364
16e47624
HS		 3365 sprintf(fw_config_file_path,
3366 "/lib/firmware/%s", fw_config_file);
3367 config_name = fw_config_file_path;
3368
636f9d37
VP		 3369 if (cf->size >= FLASH_CFG_MAX_SIZE)
3370 ret = -ENOMEM;
3371 else {
5167865a
HS		 3372 params[0] = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
3373 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_CF));
636f9d37 3374 ret = t4_query_params(adapter, adapter->mbox,
b2612722 3375 adapter->pf, 0, 1, params, val);
636f9d37
VP		 3376 if (ret == 0) {
3377 /*
fc5ab020 3378 * For t4_memory_rw() below addresses and
636f9d37
VP		 3379 * sizes have to be in terms of multiples of 4
3380 * bytes. So, if the Configuration File isn't
3381 * a multiple of 4 bytes in length we'll have
3382 * to write that out separately since we can't
3383 * guarantee that the bytes following the
3384 * residual byte in the buffer returned by
3385 * request_firmware() are zeroed out ...
3386 */
3387 size_t resid = cf->size & 0x3;
3388 size_t size = cf->size & ~0x3;
3389 __be32 *data = (__be32 *)cf->data;
3390
5167865a
HS		 3391 mtype = FW_PARAMS_PARAM_Y_G(val[0]);
3392 maddr = FW_PARAMS_PARAM_Z_G(val[0]) << 16;
636f9d37 3393
fc5ab020
HS		 3394 spin_lock(&adapter->win0_lock);
3395 ret = t4_memory_rw(adapter, 0, mtype, maddr,
3396 size, data, T4_MEMORY_WRITE);
636f9d37
VP		 3397 if (ret == 0 && resid != 0) {
3398 union {
3399 __be32 word;
3400 char buf[4];
3401 } last;
3402 int i;
3403
3404 last.word = data[size >> 2];
3405 for (i = resid; i < 4; i++)
3406 last.buf[i] = 0;
fc5ab020
HS		 3407 ret = t4_memory_rw(adapter, 0, mtype,
3408 maddr + size,
3409 4, &last.word,
3410 T4_MEMORY_WRITE);
636f9d37 3411 }
fc5ab020 3412 spin_unlock(&adapter->win0_lock);
636f9d37
VP		 3413 }
3414 }
3415
3416 release_firmware(cf);
3417 if (ret)
3418 goto bye;
3419 }
3420
3421 /*
3422 * Issue a Capability Configuration command to the firmware to get it
3423 * to parse the Configuration File. We don't use t4_fw_config_file()
3424 * because we want the ability to modify various features after we've
3425 * processed the configuration file ...
3426 */
3427 memset(&caps_cmd, 0, sizeof(caps_cmd));
3428 caps_cmd.op_to_write =
e2ac9628
HS		 3429 htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3430 FW_CMD_REQUEST_F |
3431 FW_CMD_READ_F);
ce91a923 3432 caps_cmd.cfvalid_to_len16 =
5167865a
HS		 3433 htonl(FW_CAPS_CONFIG_CMD_CFVALID_F |
3434 FW_CAPS_CONFIG_CMD_MEMTYPE_CF_V(mtype) |
3435 FW_CAPS_CONFIG_CMD_MEMADDR64K_CF_V(maddr >> 16) |
636f9d37
VP		 3436 FW_LEN16(caps_cmd));
3437 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
3438 &caps_cmd);
16e47624
HS		 3439
3440 /* If the CAPS_CONFIG failed with an ENOENT (for a Firmware
3441 * Configuration File in FLASH), our last gasp effort is to use the
3442 * Firmware Configuration File which is embedded in the firmware. A
3443 * very few early versions of the firmware didn't have one embedded
3444 * but we can ignore those.
3445 */
3446 if (ret == -ENOENT) {
3447 memset(&caps_cmd, 0, sizeof(caps_cmd));
3448 caps_cmd.op_to_write =
e2ac9628
HS		 3449 htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3450 FW_CMD_REQUEST_F |
3451 FW_CMD_READ_F);
16e47624
HS		 3452 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
3453 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd,
3454 sizeof(caps_cmd), &caps_cmd);
3455 config_name = "Firmware Default";
3456 }
3457
3458 config_issued = 1;
636f9d37
VP		 3459 if (ret < 0)
3460 goto bye;
3461
3462 finiver = ntohl(caps_cmd.finiver);
3463 finicsum = ntohl(caps_cmd.finicsum);
3464 cfcsum = ntohl(caps_cmd.cfcsum);
3465 if (finicsum != cfcsum)
3466 dev_warn(adapter->pdev_dev, "Configuration File checksum "\
3467 "mismatch: [fini] csum=%#x, computed csum=%#x\n",
3468 finicsum, cfcsum);
3469
636f9d37
VP		 3470 /*
3471 * And now tell the firmware to use the configuration we just loaded.
3472 */
3473 caps_cmd.op_to_write =
e2ac9628
HS		 3474 htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3475 FW_CMD_REQUEST_F |
3476 FW_CMD_WRITE_F);
ce91a923 3477 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
636f9d37
VP		 3478 ret = t4_wr_mbox(adapter, adapter->mbox, &caps_cmd, sizeof(caps_cmd),
3479 NULL);
3480 if (ret < 0)
3481 goto bye;
3482
3483 /*
3484 * Tweak configuration based on system architecture, module
3485 * parameters, etc.
3486 */
3487 ret = adap_init0_tweaks(adapter);
3488 if (ret < 0)
3489 goto bye;
3490
3491 /*
3492 * And finally tell the firmware to initialize itself using the
3493 * parameters from the Configuration File.
3494 */
3495 ret = t4_fw_initialize(adapter, adapter->mbox);
3496 if (ret < 0)
3497 goto bye;
3498
06640310
HS		 3499 /* Emit Firmware Configuration File information and return
3500 * successfully.
636f9d37 3501 */
636f9d37 3502 dev_info(adapter->pdev_dev, "Successfully configured using Firmware "\
16e47624
HS		 3503 "Configuration File \"%s\", version %#x, computed checksum %#x\n",
3504 config_name, finiver, cfcsum);
636f9d37
VP		 3505 return 0;
3506
3507 /*
3508 * Something bad happened. Return the error ... (If the "error"
3509 * is that there's no Configuration File on the adapter we don't
3510 * want to issue a warning since this is fairly common.)
3511 */
3512bye:
16e47624
HS		 3513 if (config_issued && ret != -ENOENT)
3514 dev_warn(adapter->pdev_dev, "\"%s\" configuration file error %d\n",
3515 config_name, -ret);
636f9d37
VP		 3516 return ret;
3517}
3518
16e47624
HS		 3519static struct fw_info fw_info_array[] = {
3520 {
3521 .chip = CHELSIO_T4,
3522 .fs_name = FW4_CFNAME,
3523 .fw_mod_name = FW4_FNAME,
3524 .fw_hdr = {
3525 .chip = FW_HDR_CHIP_T4,
3526 .fw_ver = __cpu_to_be32(FW_VERSION(T4)),
3527 .intfver_nic = FW_INTFVER(T4, NIC),
3528 .intfver_vnic = FW_INTFVER(T4, VNIC),
3529 .intfver_ri = FW_INTFVER(T4, RI),
3530 .intfver_iscsi = FW_INTFVER(T4, ISCSI),
3531 .intfver_fcoe = FW_INTFVER(T4, FCOE),
3532 },
3533 }, {
3534 .chip = CHELSIO_T5,
3535 .fs_name = FW5_CFNAME,
3536 .fw_mod_name = FW5_FNAME,
3537 .fw_hdr = {
3538 .chip = FW_HDR_CHIP_T5,
3539 .fw_ver = __cpu_to_be32(FW_VERSION(T5)),
3540 .intfver_nic = FW_INTFVER(T5, NIC),
3541 .intfver_vnic = FW_INTFVER(T5, VNIC),
3542 .intfver_ri = FW_INTFVER(T5, RI),
3543 .intfver_iscsi = FW_INTFVER(T5, ISCSI),
3544 .intfver_fcoe = FW_INTFVER(T5, FCOE),
3545 },
3ccc6cf7
HS		 3546 }, {
3547 .chip = CHELSIO_T6,
3548 .fs_name = FW6_CFNAME,
3549 .fw_mod_name = FW6_FNAME,
3550 .fw_hdr = {
3551 .chip = FW_HDR_CHIP_T6,
3552 .fw_ver = __cpu_to_be32(FW_VERSION(T6)),
3553 .intfver_nic = FW_INTFVER(T6, NIC),
3554 .intfver_vnic = FW_INTFVER(T6, VNIC),
3555 .intfver_ofld = FW_INTFVER(T6, OFLD),
3556 .intfver_ri = FW_INTFVER(T6, RI),
3557 .intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
3558 .intfver_iscsi = FW_INTFVER(T6, ISCSI),
3559 .intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
3560 .intfver_fcoe = FW_INTFVER(T6, FCOE),
3561 },
16e47624 3562 }
3ccc6cf7 3563
16e47624
HS		 3564};
3565
3566static struct fw_info *find_fw_info(int chip)
3567{
3568 int i;
3569
3570 for (i = 0; i < ARRAY_SIZE(fw_info_array); i++) {
3571 if (fw_info_array[i].chip == chip)
3572 return &fw_info_array[i];
3573 }
3574 return NULL;
3575}
3576
b8ff05a9
DM		 3577/*
3578 * Phase 0 of initialization: contact FW, obtain config, perform basic init.
3579 */
3580static int adap_init0(struct adapter *adap)
3581{
3582 int ret;
3583 u32 v, port_vec;
3584 enum dev_state state;
3585 u32 params[7], val[7];
9a4da2cd 3586 struct fw_caps_config_cmd caps_cmd;
dcf7b6f5 3587 int reset = 1;
b8ff05a9 3588
ae469b68
HS		 3589 /* Grab Firmware Device Log parameters as early as possible so we have
3590 * access to it for debugging, etc.
3591 */
3592 ret = t4_init_devlog_params(adap);
3593 if (ret < 0)
3594 return ret;
3595
666224d4 3596 /* Contact FW, advertising Master capability */
c5a8c0f3
HS		 3597 ret = t4_fw_hello(adap, adap->mbox, adap->mbox,
3598 is_kdump_kernel() ? MASTER_MUST : MASTER_MAY, &state);
b8ff05a9
DM		 3599 if (ret < 0) {
3600 dev_err(adap->pdev_dev, "could not connect to FW, error %d\n",
3601 ret);
3602 return ret;
3603 }
636f9d37
VP		 3604 if (ret == adap->mbox)
3605 adap->flags |= MASTER_PF;
b8ff05a9 3606
636f9d37
VP		 3607 /*
3608 * If we're the Master PF Driver and the device is uninitialized,
3609 * then let's consider upgrading the firmware ... (We always want
3610 * to check the firmware version number in order to A. get it for
3611 * later reporting and B. to warn if the currently loaded firmware
3612 * is excessively mismatched relative to the driver.)
3613 */
0de72738 3614
760446f9 3615 t4_get_version_info(adap);
a69265e9
HS		 3616 ret = t4_check_fw_version(adap);
3617 /* If firmware is too old (not supported by driver) force an update. */
21d11bd6 3618 if (ret)
a69265e9 3619 state = DEV_STATE_UNINIT;
636f9d37 3620 if ((adap->flags & MASTER_PF) && state != DEV_STATE_INIT) {
16e47624
HS		 3621 struct fw_info *fw_info;
3622 struct fw_hdr *card_fw;
3623 const struct firmware *fw;
3624 const u8 *fw_data = NULL;
3625 unsigned int fw_size = 0;
3626
3627 /* This is the firmware whose headers the driver was compiled
3628 * against
3629 */
3630 fw_info = find_fw_info(CHELSIO_CHIP_VERSION(adap->params.chip));
3631 if (fw_info == NULL) {
3632 dev_err(adap->pdev_dev,
3633 "unable to get firmware info for chip %d.\n",
3634 CHELSIO_CHIP_VERSION(adap->params.chip));
3635 return -EINVAL;
636f9d37 3636 }
16e47624
HS		 3637
3638 /* allocate memory to read the header of the firmware on the
3639 * card
3640 */
752ade68 3641 card_fw = kvzalloc(sizeof(*card_fw), GFP_KERNEL);
16e47624
HS		 3642
3643 /* Get FW from from /lib/firmware/ */
3644 ret = request_firmware(&fw, fw_info->fw_mod_name,
3645 adap->pdev_dev);
3646 if (ret < 0) {
3647 dev_err(adap->pdev_dev,
3648 "unable to load firmware image %s, error %d\n",
3649 fw_info->fw_mod_name, ret);
3650 } else {
3651 fw_data = fw->data;
3652 fw_size = fw->size;
3653 }
3654
3655 /* upgrade FW logic */
3656 ret = t4_prep_fw(adap, fw_info, fw_data, fw_size, card_fw,
3657 state, &reset);
3658
3659 /* Cleaning up */
0b5b6bee 3660 release_firmware(fw);
752ade68 3661 kvfree(card_fw);
16e47624 3662
636f9d37 3663 if (ret < 0)
16e47624 3664 goto bye;
636f9d37 3665 }
b8ff05a9 3666
636f9d37
VP		 3667 /*
3668 * Grab VPD parameters. This should be done after we establish a
3669 * connection to the firmware since some of the VPD parameters
3670 * (notably the Core Clock frequency) are retrieved via requests to
3671 * the firmware. On the other hand, we need these fairly early on
3672 * so we do this right after getting ahold of the firmware.
3673 */
098ef6c2 3674 ret = t4_get_vpd_params(adap, &adap->params.vpd);
a0881cab
DM		 3675 if (ret < 0)
3676 goto bye;
a0881cab 3677
636f9d37 3678 /*
13ee15d3
VP		 3679 * Find out what ports are available to us. Note that we need to do
3680 * this before calling adap_init0_no_config() since it needs nports
3681 * and portvec ...
636f9d37
VP		 3682 */
3683 v =
5167865a
HS		 3684 FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
3685 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_PORTVEC);
b2612722 3686 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1, &v, &port_vec);
a0881cab
DM		 3687 if (ret < 0)
3688 goto bye;
3689
636f9d37
VP		 3690 adap->params.nports = hweight32(port_vec);
3691 adap->params.portvec = port_vec;
3692
06640310
HS		 3693 /* If the firmware is initialized already, emit a simply note to that
3694 * effect. Otherwise, it's time to try initializing the adapter.
636f9d37
VP		 3695 */
3696 if (state == DEV_STATE_INIT) {
3697 dev_info(adap->pdev_dev, "Coming up as %s: "\
3698 "Adapter already initialized\n",
3699 adap->flags & MASTER_PF ? "MASTER" : "SLAVE");
636f9d37
VP		 3700 } else {
3701 dev_info(adap->pdev_dev, "Coming up as MASTER: "\
3702 "Initializing adapter\n");
06640310
HS		 3703
3704 /* Find out whether we're dealing with a version of the
3705 * firmware which has configuration file support.
636f9d37 3706 */
06640310
HS		 3707 params[0] = (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
3708 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_CF));
b2612722 3709 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 1,
06640310 3710 params, val);
13ee15d3 3711
06640310
HS		 3712 /* If the firmware doesn't support Configuration Files,
3713 * return an error.
3714 */
3715 if (ret < 0) {
3716 dev_err(adap->pdev_dev, "firmware doesn't support "
3717 "Firmware Configuration Files\n");
3718 goto bye;
3719 }
3720
3721 /* The firmware provides us with a memory buffer where we can
3722 * load a Configuration File from the host if we want to
3723 * override the Configuration File in flash.
3724 */
3725 ret = adap_init0_config(adap, reset);
3726 if (ret == -ENOENT) {
3727 dev_err(adap->pdev_dev, "no Configuration File "
3728 "present on adapter.\n");
3729 goto bye;
636f9d37
VP		 3730 }
3731 if (ret < 0) {
06640310
HS		 3732 dev_err(adap->pdev_dev, "could not initialize "
3733 "adapter, error %d\n", -ret);
636f9d37
VP		 3734 goto bye;
3735 }
3736 }
3737
06640310
HS		 3738 /* Give the SGE code a chance to pull in anything that it needs ...
3739 * Note that this must be called after we retrieve our VPD parameters
3740 * in order to know how to convert core ticks to seconds, etc.
636f9d37 3741 */
06640310
HS		 3742 ret = t4_sge_init(adap);
3743 if (ret < 0)
3744 goto bye;
636f9d37 3745
9a4da2cd
VP		 3746 if (is_bypass_device(adap->pdev->device))
3747 adap->params.bypass = 1;
3748
636f9d37
VP		 3749 /*
3750 * Grab some of our basic fundamental operating parameters.
3751 */
3752#define FW_PARAM_DEV(param) \
5167865a
HS		 3753 (FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) | \
3754 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_##param))
636f9d37 3755
b8ff05a9 3756#define FW_PARAM_PFVF(param) \
5167865a
HS		 3757 FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_PFVF) | \
3758 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_PFVF_##param)| \
3759 FW_PARAMS_PARAM_Y_V(0) | \
3760 FW_PARAMS_PARAM_Z_V(0)
b8ff05a9 3761
636f9d37 3762 params[0] = FW_PARAM_PFVF(EQ_START);
b8ff05a9
DM		 3763 params[1] = FW_PARAM_PFVF(L2T_START);
3764 params[2] = FW_PARAM_PFVF(L2T_END);
3765 params[3] = FW_PARAM_PFVF(FILTER_START);
3766 params[4] = FW_PARAM_PFVF(FILTER_END);
e46dab4d 3767 params[5] = FW_PARAM_PFVF(IQFLINT_START);
b2612722 3768 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6, params, val);
b8ff05a9
DM		 3769 if (ret < 0)
3770 goto bye;
636f9d37
VP		 3771 adap->sge.egr_start = val[0];
3772 adap->l2t_start = val[1];
3773 adap->l2t_end = val[2];
b8ff05a9
DM		 3774 adap->tids.ftid_base = val[3];
3775 adap->tids.nftids = val[4] - val[3] + 1;
e46dab4d 3776 adap->sge.ingr_start = val[5];
b8ff05a9 3777
4b8e27a8
HS		 3778 /* qids (ingress/egress) returned from firmware can be anywhere
3779 * in the range from EQ(IQFLINT)_START to EQ(IQFLINT)_END.
3780 * Hence driver needs to allocate memory for this range to
3781 * store the queue info. Get the highest IQFLINT/EQ index returned
3782 * in FW_EQ_*_CMD.alloc command.
3783 */
3784 params[0] = FW_PARAM_PFVF(EQ_END);
3785 params[1] = FW_PARAM_PFVF(IQFLINT_END);
b2612722 3786 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
4b8e27a8
HS		 3787 if (ret < 0)
3788 goto bye;
3789 adap->sge.egr_sz = val[0] - adap->sge.egr_start + 1;
3790 adap->sge.ingr_sz = val[1] - adap->sge.ingr_start + 1;
3791
3792 adap->sge.egr_map = kcalloc(adap->sge.egr_sz,
3793 sizeof(*adap->sge.egr_map), GFP_KERNEL);
3794 if (!adap->sge.egr_map) {
3795 ret = -ENOMEM;
3796 goto bye;
3797 }
3798
3799 adap->sge.ingr_map = kcalloc(adap->sge.ingr_sz,
3800 sizeof(*adap->sge.ingr_map), GFP_KERNEL);
3801 if (!adap->sge.ingr_map) {
3802 ret = -ENOMEM;
3803 goto bye;
3804 }
3805
3806 /* Allocate the memory for the vaious egress queue bitmaps
5b377d11 3807 * ie starving_fl, txq_maperr and blocked_fl.
4b8e27a8
HS		 3808 */
3809 adap->sge.starving_fl = kcalloc(BITS_TO_LONGS(adap->sge.egr_sz),
3810 sizeof(long), GFP_KERNEL);
3811 if (!adap->sge.starving_fl) {
3812 ret = -ENOMEM;
3813 goto bye;
3814 }
3815
3816 adap->sge.txq_maperr = kcalloc(BITS_TO_LONGS(adap->sge.egr_sz),
3817 sizeof(long), GFP_KERNEL);
3818 if (!adap->sge.txq_maperr) {
3819 ret = -ENOMEM;
3820 goto bye;
3821 }
3822
5b377d11
HS		 3823#ifdef CONFIG_DEBUG_FS
3824 adap->sge.blocked_fl = kcalloc(BITS_TO_LONGS(adap->sge.egr_sz),
3825 sizeof(long), GFP_KERNEL);
3826 if (!adap->sge.blocked_fl) {
3827 ret = -ENOMEM;
3828 goto bye;
3829 }
3830#endif
3831
b5a02f50
AB		 3832 params[0] = FW_PARAM_PFVF(CLIP_START);
3833 params[1] = FW_PARAM_PFVF(CLIP_END);
b2612722 3834 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
b5a02f50
AB		 3835 if (ret < 0)
3836 goto bye;
3837 adap->clipt_start = val[0];
3838 adap->clipt_end = val[1];
3839
b72a32da
RL		 3840 /* We don't yet have a PARAMs calls to retrieve the number of Traffic
3841 * Classes supported by the hardware/firmware so we hard code it here
3842 * for now.
3843 */
3844 adap->params.nsched_cls = is_t4(adap->params.chip) ? 15 : 16;
3845
636f9d37
VP		 3846 /* query params related to active filter region */
3847 params[0] = FW_PARAM_PFVF(ACTIVE_FILTER_START);
3848 params[1] = FW_PARAM_PFVF(ACTIVE_FILTER_END);
b2612722 3849 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params, val);
636f9d37
VP		 3850 /* If Active filter size is set we enable establishing
3851 * offload connection through firmware work request
3852 */
3853 if ((val[0] != val[1]) && (ret >= 0)) {
3854 adap->flags |= FW_OFLD_CONN;
3855 adap->tids.aftid_base = val[0];
3856 adap->tids.aftid_end = val[1];
3857 }
3858
b407a4a9
VP		 3859 /* If we're running on newer firmware, let it know that we're
3860 * prepared to deal with encapsulated CPL messages. Older
3861 * firmware won't understand this and we'll just get
3862 * unencapsulated messages ...
3863 */
3864 params[0] = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
3865 val[0] = 1;
b2612722 3866 (void)t4_set_params(adap, adap->mbox, adap->pf, 0, 1, params, val);
b407a4a9 3867
1ac0f095
KS		 3868 /*
3869 * Find out whether we're allowed to use the T5+ ULPTX MEMWRITE DSGL
3870 * capability. Earlier versions of the firmware didn't have the
3871 * ULPTX_MEMWRITE_DSGL so we'll interpret a query failure as no
3872 * permission to use ULPTX MEMWRITE DSGL.
3873 */
3874 if (is_t4(adap->params.chip)) {
3875 adap->params.ulptx_memwrite_dsgl = false;
3876 } else {
3877 params[0] = FW_PARAM_DEV(ULPTX_MEMWRITE_DSGL);
b2612722 3878 ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
1ac0f095
KS		 3879 1, params, val);
3880 adap->params.ulptx_memwrite_dsgl = (ret == 0 && val[0] != 0);
3881 }
3882
086de575
SW		 3883 /* See if FW supports FW_RI_FR_NSMR_TPTE_WR work request */
3884 params[0] = FW_PARAM_DEV(RI_FR_NSMR_TPTE_WR);
3885 ret = t4_query_params(adap, adap->mbox, adap->pf, 0,
3886 1, params, val);
3887 adap->params.fr_nsmr_tpte_wr_support = (ret == 0 && val[0] != 0);
3888
636f9d37
VP		 3889 /*
3890 * Get device capabilities so we can determine what resources we need
3891 * to manage.
3892 */
3893 memset(&caps_cmd, 0, sizeof(caps_cmd));
e2ac9628
HS		 3894 caps_cmd.op_to_write = htonl(FW_CMD_OP_V(FW_CAPS_CONFIG_CMD) |
3895 FW_CMD_REQUEST_F | FW_CMD_READ_F);
ce91a923 3896 caps_cmd.cfvalid_to_len16 = htonl(FW_LEN16(caps_cmd));
636f9d37
VP		 3897 ret = t4_wr_mbox(adap, adap->mbox, &caps_cmd, sizeof(caps_cmd),
3898 &caps_cmd);
3899 if (ret < 0)
3900 goto bye;
3901
13ee15d3 3902 if (caps_cmd.ofldcaps) {
b8ff05a9
DM		 3903 /* query offload-related parameters */
3904 params[0] = FW_PARAM_DEV(NTID);
3905 params[1] = FW_PARAM_PFVF(SERVER_START);
3906 params[2] = FW_PARAM_PFVF(SERVER_END);
3907 params[3] = FW_PARAM_PFVF(TDDP_START);
3908 params[4] = FW_PARAM_PFVF(TDDP_END);
3909 params[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
b2612722 3910 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6,
636f9d37 3911 params, val);
b8ff05a9
DM		 3912 if (ret < 0)
3913 goto bye;
3914 adap->tids.ntids = val[0];
3915 adap->tids.natids = min(adap->tids.ntids / 2, MAX_ATIDS);
3916 adap->tids.stid_base = val[1];
3917 adap->tids.nstids = val[2] - val[1] + 1;
636f9d37 3918 /*
dbedd44e 3919 * Setup server filter region. Divide the available filter
636f9d37
VP		 3920 * region into two parts. Regular filters get 1/3rd and server
3921 * filters get 2/3rd part. This is only enabled if workarond
3922 * path is enabled.
3923 * 1. For regular filters.
3924 * 2. Server filter: This are special filters which are used
3925 * to redirect SYN packets to offload queue.
3926 */
3927 if (adap->flags & FW_OFLD_CONN && !is_bypass(adap)) {
3928 adap->tids.sftid_base = adap->tids.ftid_base +
3929 DIV_ROUND_UP(adap->tids.nftids, 3);
3930 adap->tids.nsftids = adap->tids.nftids -
3931 DIV_ROUND_UP(adap->tids.nftids, 3);
3932 adap->tids.nftids = adap->tids.sftid_base -
3933 adap->tids.ftid_base;
3934 }
b8ff05a9
DM		 3935 adap->vres.ddp.start = val[3];
3936 adap->vres.ddp.size = val[4] - val[3] + 1;
3937 adap->params.ofldq_wr_cred = val[5];
636f9d37 3938
b8ff05a9 3939 adap->params.offload = 1;
0fbc81b3 3940 adap->num_ofld_uld += 1;
b8ff05a9 3941 }
636f9d37 3942 if (caps_cmd.rdmacaps) {
b8ff05a9
DM		 3943 params[0] = FW_PARAM_PFVF(STAG_START);
3944 params[1] = FW_PARAM_PFVF(STAG_END);
3945 params[2] = FW_PARAM_PFVF(RQ_START);
3946 params[3] = FW_PARAM_PFVF(RQ_END);
3947 params[4] = FW_PARAM_PFVF(PBL_START);
3948 params[5] = FW_PARAM_PFVF(PBL_END);
b2612722 3949 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6,
636f9d37 3950 params, val);
b8ff05a9
DM		 3951 if (ret < 0)
3952 goto bye;
3953 adap->vres.stag.start = val[0];
3954 adap->vres.stag.size = val[1] - val[0] + 1;
3955 adap->vres.rq.start = val[2];
3956 adap->vres.rq.size = val[3] - val[2] + 1;
3957 adap->vres.pbl.start = val[4];
3958 adap->vres.pbl.size = val[5] - val[4] + 1;
a0881cab
DM		 3959
3960 params[0] = FW_PARAM_PFVF(SQRQ_START);
3961 params[1] = FW_PARAM_PFVF(SQRQ_END);
3962 params[2] = FW_PARAM_PFVF(CQ_START);
3963 params[3] = FW_PARAM_PFVF(CQ_END);
1ae970e0
DM		 3964 params[4] = FW_PARAM_PFVF(OCQ_START);
3965 params[5] = FW_PARAM_PFVF(OCQ_END);
b2612722 3966 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 6, params,
5c937dd3 3967 val);
a0881cab
DM		 3968 if (ret < 0)
3969 goto bye;
3970 adap->vres.qp.start = val[0];
3971 adap->vres.qp.size = val[1] - val[0] + 1;
3972 adap->vres.cq.start = val[2];
3973 adap->vres.cq.size = val[3] - val[2] + 1;
1ae970e0
DM		 3974 adap->vres.ocq.start = val[4];
3975 adap->vres.ocq.size = val[5] - val[4] + 1;
4c2c5763
HS		 3976
3977 params[0] = FW_PARAM_DEV(MAXORDIRD_QP);
3978 params[1] = FW_PARAM_DEV(MAXIRD_ADAPTER);
b2612722 3979 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2, params,
5c937dd3 3980 val);
4c2c5763
HS		 3981 if (ret < 0) {
3982 adap->params.max_ordird_qp = 8;
3983 adap->params.max_ird_adapter = 32 * adap->tids.ntids;
3984 ret = 0;
3985 } else {
3986 adap->params.max_ordird_qp = val[0];
3987 adap->params.max_ird_adapter = val[1];
3988 }
3989 dev_info(adap->pdev_dev,
3990 "max_ordird_qp %d max_ird_adapter %d\n",
3991 adap->params.max_ordird_qp,
3992 adap->params.max_ird_adapter);
0fbc81b3 3993 adap->num_ofld_uld += 2;
b8ff05a9 3994 }
636f9d37 3995 if (caps_cmd.iscsicaps) {
b8ff05a9
DM		 3996 params[0] = FW_PARAM_PFVF(ISCSI_START);
3997 params[1] = FW_PARAM_PFVF(ISCSI_END);
b2612722 3998 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
636f9d37 3999 params, val);
b8ff05a9
DM		 4000 if (ret < 0)
4001 goto bye;
4002 adap->vres.iscsi.start = val[0];
4003 adap->vres.iscsi.size = val[1] - val[0] + 1;
0fbc81b3
HS		 4004 /* LIO target and cxgb4i initiaitor */
4005 adap->num_ofld_uld += 2;
b8ff05a9 4006 }
94cdb8bb
HS		 4007 if (caps_cmd.cryptocaps) {
4008 /* Should query params here...TODO */
72a56ca9
HJ		 4009 params[0] = FW_PARAM_PFVF(NCRYPTO_LOOKASIDE);
4010 ret = t4_query_params(adap, adap->mbox, adap->pf, 0, 2,
4011 params, val);
4012 if (ret < 0) {
4013 if (ret != -EINVAL)
4014 goto bye;
4015 } else {
4016 adap->vres.ncrypto_fc = val[0];
4017 }
94cdb8bb
HS		 4018 adap->params.crypto |= ULP_CRYPTO_LOOKASIDE;
4019 adap->num_uld += 1;
4020 }
b8ff05a9
DM		 4021#undef FW_PARAM_PFVF
4022#undef FW_PARAM_DEV
4023
92e7ae71
HS		 4024 /* The MTU/MSS Table is initialized by now, so load their values. If
4025 * we're initializing the adapter, then we'll make any modifications
4026 * we want to the MTU/MSS Table and also initialize the congestion
4027 * parameters.
636f9d37 4028 */
b8ff05a9 4029 t4_read_mtu_tbl(adap, adap->params.mtus, NULL);
92e7ae71
HS		 4030 if (state != DEV_STATE_INIT) {
4031 int i;
4032
4033 /* The default MTU Table contains values 1492 and 1500.
4034 * However, for TCP, it's better to have two values which are
4035 * a multiple of 8 +/- 4 bytes apart near this popular MTU.
4036 * This allows us to have a TCP Data Payload which is a
4037 * multiple of 8 regardless of what combination of TCP Options
4038 * are in use (always a multiple of 4 bytes) which is
4039 * important for performance reasons. For instance, if no
4040 * options are in use, then we have a 20-byte IP header and a
4041 * 20-byte TCP header. In this case, a 1500-byte MSS would
4042 * result in a TCP Data Payload of 1500 - 40 == 1460 bytes
4043 * which is not a multiple of 8. So using an MSS of 1488 in
4044 * this case results in a TCP Data Payload of 1448 bytes which
4045 * is a multiple of 8. On the other hand, if 12-byte TCP Time
4046 * Stamps have been negotiated, then an MTU of 1500 bytes
4047 * results in a TCP Data Payload of 1448 bytes which, as
4048 * above, is a multiple of 8 bytes ...
4049 */
4050 for (i = 0; i < NMTUS; i++)
4051 if (adap->params.mtus[i] == 1492) {
4052 adap->params.mtus[i] = 1488;
4053 break;
4054 }
7ee9ff94 4055
92e7ae71
HS		 4056 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
4057 adap->params.b_wnd);
4058 }
df64e4d3 4059 t4_init_sge_params(adap);
636f9d37 4060 adap->flags |= FW_OK;
c1e9af0c 4061 t4_init_tp_params(adap);
b8ff05a9
DM		 4062 return 0;
4063
4064 /*
636f9d37
VP		 4065 * Something bad happened. If a command timed out or failed with EIO
4066 * FW does not operate within its spec or something catastrophic
4067 * happened to HW/FW, stop issuing commands.
b8ff05a9 4068 */
636f9d37 4069bye:
4b8e27a8
HS		 4070 kfree(adap->sge.egr_map);
4071 kfree(adap->sge.ingr_map);
4072 kfree(adap->sge.starving_fl);
4073 kfree(adap->sge.txq_maperr);
5b377d11
HS		 4074#ifdef CONFIG_DEBUG_FS
4075 kfree(adap->sge.blocked_fl);
4076#endif
636f9d37
VP		 4077 if (ret != -ETIMEDOUT && ret != -EIO)
4078 t4_fw_bye(adap, adap->mbox);
b8ff05a9
DM		 4079 return ret;
4080}
4081
204dc3c0
DM		 4082/* EEH callbacks */
4083
4084static pci_ers_result_t eeh_err_detected(struct pci_dev *pdev,
4085 pci_channel_state_t state)
4086{
4087 int i;
4088 struct adapter *adap = pci_get_drvdata(pdev);
4089
4090 if (!adap)
4091 goto out;
4092
4093 rtnl_lock();
4094 adap->flags &= ~FW_OK;
4095 notify_ulds(adap, CXGB4_STATE_START_RECOVERY);
9fe6cb58 4096 spin_lock(&adap->stats_lock);
204dc3c0
DM		 4097 for_each_port(adap, i) {
4098 struct net_device *dev = adap->port[i];
025d0973
GP		 4099 if (dev) {
4100 netif_device_detach(dev);
4101 netif_carrier_off(dev);
4102 }
204dc3c0 4103 }
9fe6cb58 4104 spin_unlock(&adap->stats_lock);
b37987e8 4105 disable_interrupts(adap);
204dc3c0
DM		 4106 if (adap->flags & FULL_INIT_DONE)
4107 cxgb_down(adap);
4108 rtnl_unlock();
144be3d9
GS		 4109 if ((adap->flags & DEV_ENABLED)) {
4110 pci_disable_device(pdev);
4111 adap->flags &= ~DEV_ENABLED;
4112 }
204dc3c0
DM		 4113out: return state == pci_channel_io_perm_failure ?
4114 PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
4115}
4116
4117static pci_ers_result_t eeh_slot_reset(struct pci_dev *pdev)
4118{
4119 int i, ret;
4120 struct fw_caps_config_cmd c;
4121 struct adapter *adap = pci_get_drvdata(pdev);
4122
4123 if (!adap) {
4124 pci_restore_state(pdev);
4125 pci_save_state(pdev);
4126 return PCI_ERS_RESULT_RECOVERED;
4127 }
4128
144be3d9
GS		 4129 if (!(adap->flags & DEV_ENABLED)) {
4130 if (pci_enable_device(pdev)) {
4131 dev_err(&pdev->dev, "Cannot reenable PCI "
4132 "device after reset\n");
4133 return PCI_ERS_RESULT_DISCONNECT;
4134 }
4135 adap->flags |= DEV_ENABLED;
204dc3c0
DM		 4136 }
4137
4138 pci_set_master(pdev);
4139 pci_restore_state(pdev);
4140 pci_save_state(pdev);
4141 pci_cleanup_aer_uncorrect_error_status(pdev);
4142
8203b509 4143 if (t4_wait_dev_ready(adap->regs) < 0)
204dc3c0 4144 return PCI_ERS_RESULT_DISCONNECT;
b2612722 4145 if (t4_fw_hello(adap, adap->mbox, adap->pf, MASTER_MUST, NULL) < 0)
204dc3c0
DM		 4146 return PCI_ERS_RESULT_DISCONNECT;
4147 adap->flags |= FW_OK;
4148 if (adap_init1(adap, &c))
4149 return PCI_ERS_RESULT_DISCONNECT;
4150
4151 for_each_port(adap, i) {
4152 struct port_info *p = adap2pinfo(adap, i);
4153
b2612722 4154 ret = t4_alloc_vi(adap, adap->mbox, p->tx_chan, adap->pf, 0, 1,
060e0c75 4155 NULL, NULL);
204dc3c0
DM		 4156 if (ret < 0)
4157 return PCI_ERS_RESULT_DISCONNECT;
4158 p->viid = ret;
4159 p->xact_addr_filt = -1;
4160 }
4161
4162 t4_load_mtus(adap, adap->params.mtus, adap->params.a_wnd,
4163 adap->params.b_wnd);
1ae970e0 4164 setup_memwin(adap);
204dc3c0
DM		 4165 if (cxgb_up(adap))
4166 return PCI_ERS_RESULT_DISCONNECT;
4167 return PCI_ERS_RESULT_RECOVERED;
4168}
4169
4170static void eeh_resume(struct pci_dev *pdev)
4171{
4172 int i;
4173 struct adapter *adap = pci_get_drvdata(pdev);
4174
4175 if (!adap)
4176 return;
4177
4178 rtnl_lock();
4179 for_each_port(adap, i) {
4180 struct net_device *dev = adap->port[i];
025d0973
GP		 4181 if (dev) {
4182 if (netif_running(dev)) {
4183 link_start(dev);
4184 cxgb_set_rxmode(dev);
4185 }
4186 netif_device_attach(dev);
204dc3c0 4187 }
204dc3c0
DM		 4188 }
4189 rtnl_unlock();
4190}
4191
3646f0e5 4192static const struct pci_error_handlers cxgb4_eeh = {
204dc3c0
DM		 4193 .error_detected = eeh_err_detected,
4194 .slot_reset = eeh_slot_reset,
4195 .resume = eeh_resume,
4196};
4197
9b86a8d1
HS		 4198/* Return true if the Link Configuration supports "High Speeds" (those greater
4199 * than 1Gb/s).
4200 */
57d8b764 4201static inline bool is_x_10g_port(const struct link_config *lc)
b8ff05a9 4202{
9b86a8d1
HS		 4203 unsigned int speeds, high_speeds;
4204
4205 speeds = FW_PORT_CAP_SPEED_V(FW_PORT_CAP_SPEED_G(lc->supported));
4206 high_speeds = speeds & ~(FW_PORT_CAP_SPEED_100M | FW_PORT_CAP_SPEED_1G);
4207
4208 return high_speeds != 0;
b8ff05a9
DM		 4209}
4210
b8ff05a9
DM		 4211/*
4212 * Perform default configuration of DMA queues depending on the number and type
4213 * of ports we found and the number of available CPUs. Most settings can be
4214 * modified by the admin prior to actual use.
4215 */
91744948 4216static void cfg_queues(struct adapter *adap)
b8ff05a9
DM		 4217{
4218 struct sge *s = &adap->sge;
ab677ff4 4219 int i = 0, n10g = 0, qidx = 0;
688848b1
AB		 4220#ifndef CONFIG_CHELSIO_T4_DCB
4221 int q10g = 0;
4222#endif
b8ff05a9 4223
94cdb8bb
HS		 4224 /* Reduce memory usage in kdump environment, disable all offload.
4225 */
85eacf3f 4226 if (is_kdump_kernel() || (is_uld(adap) && t4_uld_mem_alloc(adap))) {
0fbc81b3 4227 adap->params.offload = 0;
94cdb8bb
HS		 4228 adap->params.crypto = 0;
4229 }
4230
ab677ff4 4231 n10g += is_x_10g_port(&adap2pinfo(adap, i)->link_cfg);
688848b1
AB		 4232#ifdef CONFIG_CHELSIO_T4_DCB
4233 /* For Data Center Bridging support we need to be able to support up
4234 * to 8 Traffic Priorities; each of which will be assigned to its
4235 * own TX Queue in order to prevent Head-Of-Line Blocking.
4236 */
4237 if (adap->params.nports * 8 > MAX_ETH_QSETS) {
4238 dev_err(adap->pdev_dev, "MAX_ETH_QSETS=%d < %d!\n",
4239 MAX_ETH_QSETS, adap->params.nports * 8);
4240 BUG_ON(1);
4241 }
b8ff05a9 4242
688848b1
AB		 4243 for_each_port(adap, i) {
4244 struct port_info *pi = adap2pinfo(adap, i);
4245
4246 pi->first_qset = qidx;
85eacf3f 4247 pi->nqsets = is_kdump_kernel() ? 1 : 8;
688848b1
AB		 4248 qidx += pi->nqsets;
4249 }
4250#else /* !CONFIG_CHELSIO_T4_DCB */
b8ff05a9
DM		 4251 /*
4252 * We default to 1 queue per non-10G port and up to # of cores queues
4253 * per 10G port.
4254 */
4255 if (n10g)
4256 q10g = (MAX_ETH_QSETS - (adap->params.nports - n10g)) / n10g;
5952dde7
YM		 4257 if (q10g > netif_get_num_default_rss_queues())
4258 q10g = netif_get_num_default_rss_queues();
b8ff05a9 4259
85eacf3f
GG		 4260 if (is_kdump_kernel())
4261 q10g = 1;
4262
b8ff05a9
DM		 4263 for_each_port(adap, i) {
4264 struct port_info *pi = adap2pinfo(adap, i);
4265
4266 pi->first_qset = qidx;
57d8b764 4267 pi->nqsets = is_x_10g_port(&pi->link_cfg) ? q10g : 1;
b8ff05a9
DM		 4268 qidx += pi->nqsets;
4269 }
688848b1 4270#endif /* !CONFIG_CHELSIO_T4_DCB */
b8ff05a9
DM		 4271
4272 s->ethqsets = qidx;
4273 s->max_ethqsets = qidx; /* MSI-X may lower it later */
4274
0fbc81b3 4275 if (is_uld(adap)) {
b8ff05a9
DM		 4276 /*
4277 * For offload we use 1 queue/channel if all ports are up to 1G,
4278 * otherwise we divide all available queues amongst the channels
4279 * capped by the number of available cores.
4280 */
4281 if (n10g) {
a56177e1 4282 i = min_t(int, MAX_OFLD_QSETS, num_online_cpus());
0fbc81b3
HS		 4283 s->ofldqsets = roundup(i, adap->params.nports);
4284 } else {
4285 s->ofldqsets = adap->params.nports;
4286 }
b8ff05a9
DM		 4287 }
4288
4289 for (i = 0; i < ARRAY_SIZE(s->ethrxq); i++) {
4290 struct sge_eth_rxq *r = &s->ethrxq[i];
4291
c887ad0e 4292 init_rspq(adap, &r->rspq, 5, 10, 1024, 64);
b8ff05a9
DM		 4293 r->fl.size = 72;
4294 }
4295
4296 for (i = 0; i < ARRAY_SIZE(s->ethtxq); i++)
4297 s->ethtxq[i].q.size = 1024;
4298
4299 for (i = 0; i < ARRAY_SIZE(s->ctrlq); i++)
4300 s->ctrlq[i].q.size = 512;
4301
a4569504
AG		 4302 if (!is_t4(adap->params.chip))
4303 s->ptptxq.q.size = 8;
4304
c887ad0e 4305 init_rspq(adap, &s->fw_evtq, 0, 1, 1024, 64);
0fbc81b3 4306 init_rspq(adap, &s->intrq, 0, 1, 512, 64);
b8ff05a9
DM		 4307}
4308
4309/*
4310 * Reduce the number of Ethernet queues across all ports to at most n.
4311 * n provides at least one queue per port.
4312 */
91744948 4313static void reduce_ethqs(struct adapter *adap, int n)
b8ff05a9
DM		 4314{
4315 int i;
4316 struct port_info *pi;
4317
4318 while (n < adap->sge.ethqsets)
4319 for_each_port(adap, i) {
4320 pi = adap2pinfo(adap, i);
4321 if (pi->nqsets > 1) {
4322 pi->nqsets--;
4323 adap->sge.ethqsets--;
4324 if (adap->sge.ethqsets <= n)
4325 break;
4326 }
4327 }
4328
4329 n = 0;
4330 for_each_port(adap, i) {
4331 pi = adap2pinfo(adap, i);
4332 pi->first_qset = n;
4333 n += pi->nqsets;
4334 }
4335}
4336
94cdb8bb
HS		 4337static int get_msix_info(struct adapter *adap)
4338{
4339 struct uld_msix_info *msix_info;
0fbc81b3
HS		 4340 unsigned int max_ingq = 0;
4341
4342 if (is_offload(adap))
4343 max_ingq += MAX_OFLD_QSETS * adap->num_ofld_uld;
4344 if (is_pci_uld(adap))
4345 max_ingq += MAX_OFLD_QSETS * adap->num_uld;
4346
4347 if (!max_ingq)
4348 goto out;
94cdb8bb
HS		 4349
4350 msix_info = kcalloc(max_ingq, sizeof(*msix_info), GFP_KERNEL);
4351 if (!msix_info)
4352 return -ENOMEM;
4353
4354 adap->msix_bmap_ulds.msix_bmap = kcalloc(BITS_TO_LONGS(max_ingq),
4355 sizeof(long), GFP_KERNEL);
4356 if (!adap->msix_bmap_ulds.msix_bmap) {
4357 kfree(msix_info);
4358 return -ENOMEM;
4359 }
4360 spin_lock_init(&adap->msix_bmap_ulds.lock);
4361 adap->msix_info_ulds = msix_info;
0fbc81b3 4362out:
94cdb8bb
HS		 4363 return 0;
4364}
4365
4366static void free_msix_info(struct adapter *adap)
4367{
0fbc81b3 4368 if (!(adap->num_uld && adap->num_ofld_uld))
94cdb8bb
HS		 4369 return;
4370
4371 kfree(adap->msix_info_ulds);
4372 kfree(adap->msix_bmap_ulds.msix_bmap);
4373}
4374
b8ff05a9
DM		 4375/* 2 MSI-X vectors needed for the FW queue and non-data interrupts */
4376#define EXTRA_VECS 2
4377
91744948 4378static int enable_msix(struct adapter *adap)
b8ff05a9 4379{
94cdb8bb
HS		 4380 int ofld_need = 0, uld_need = 0;
4381 int i, j, want, need, allocated;
b8ff05a9
DM		 4382 struct sge *s = &adap->sge;
4383 unsigned int nchan = adap->params.nports;
f36e58e5 4384 struct msix_entry *entries;
94cdb8bb 4385 int max_ingq = MAX_INGQ;
f36e58e5 4386
0fbc81b3
HS		 4387 if (is_pci_uld(adap))
4388 max_ingq += (MAX_OFLD_QSETS * adap->num_uld);
4389 if (is_offload(adap))
4390 max_ingq += (MAX_OFLD_QSETS * adap->num_ofld_uld);
94cdb8bb 4391 entries = kmalloc(sizeof(*entries) * (max_ingq + 1),
f36e58e5
HS		 4392 GFP_KERNEL);
4393 if (!entries)
4394 return -ENOMEM;
b8ff05a9 4395
94cdb8bb 4396 /* map for msix */
0fbc81b3
HS		 4397 if (get_msix_info(adap)) {
4398 adap->params.offload = 0;
94cdb8bb 4399 adap->params.crypto = 0;
0fbc81b3 4400 }
94cdb8bb
HS		 4401
4402 for (i = 0; i < max_ingq + 1; ++i)
b8ff05a9
DM		 4403 entries[i].entry = i;
4404
4405 want = s->max_ethqsets + EXTRA_VECS;
4406 if (is_offload(adap)) {
0fbc81b3
HS		 4407 want += adap->num_ofld_uld * s->ofldqsets;
4408 ofld_need = adap->num_ofld_uld * nchan;
b8ff05a9 4409 }
94cdb8bb 4410 if (is_pci_uld(adap)) {
0fbc81b3
HS		 4411 want += adap->num_uld * s->ofldqsets;
4412 uld_need = adap->num_uld * nchan;
94cdb8bb 4413 }
688848b1
AB		 4414#ifdef CONFIG_CHELSIO_T4_DCB
4415 /* For Data Center Bridging we need 8 Ethernet TX Priority Queues for
4416 * each port.
4417 */
94cdb8bb 4418 need = 8 * adap->params.nports + EXTRA_VECS + ofld_need + uld_need;
688848b1 4419#else
94cdb8bb 4420 need = adap->params.nports + EXTRA_VECS + ofld_need + uld_need;
688848b1 4421#endif
f36e58e5
HS		 4422 allocated = pci_enable_msix_range(adap->pdev, entries, need, want);
4423 if (allocated < 0) {
4424 dev_info(adap->pdev_dev, "not enough MSI-X vectors left,"
4425 " not using MSI-X\n");
4426 kfree(entries);
4427 return allocated;
4428 }
b8ff05a9 4429
f36e58e5 4430 /* Distribute available vectors to the various queue groups.
c32ad224
AG		 4431 * Every group gets its minimum requirement and NIC gets top
4432 * priority for leftovers.
4433 */
94cdb8bb 4434 i = allocated - EXTRA_VECS - ofld_need - uld_need;
c32ad224
AG		 4435 if (i < s->max_ethqsets) {
4436 s->max_ethqsets = i;
4437 if (i < s->ethqsets)
4438 reduce_ethqs(adap, i);
4439 }
0fbc81b3 4440 if (is_uld(adap)) {
94cdb8bb
HS		 4441 if (allocated < want)
4442 s->nqs_per_uld = nchan;
4443 else
0fbc81b3 4444 s->nqs_per_uld = s->ofldqsets;
94cdb8bb
HS		 4445 }
4446
0fbc81b3 4447 for (i = 0; i < (s->max_ethqsets + EXTRA_VECS); ++i)
c32ad224 4448 adap->msix_info[i].vec = entries[i].vector;
0fbc81b3
HS		 4449 if (is_uld(adap)) {
4450 for (j = 0 ; i < allocated; ++i, j++) {
94cdb8bb 4451 adap->msix_info_ulds[j].vec = entries[i].vector;
0fbc81b3
HS		 4452 adap->msix_info_ulds[j].idx = i;
4453 }
94cdb8bb
HS		 4454 adap->msix_bmap_ulds.mapsize = j;
4455 }
43eb4e82 4456 dev_info(adap->pdev_dev, "%d MSI-X vectors allocated, "
0fbc81b3
HS		 4457 "nic %d per uld %d\n",
4458 allocated, s->max_ethqsets, s->nqs_per_uld);
c32ad224 4459
f36e58e5 4460 kfree(entries);
c32ad224 4461 return 0;
b8ff05a9
DM		 4462}
4463
4464#undef EXTRA_VECS
4465
91744948 4466static int init_rss(struct adapter *adap)
671b0060 4467{
c035e183
HS		 4468 unsigned int i;
4469 int err;
4470
4471 err = t4_init_rss_mode(adap, adap->mbox);
4472 if (err)
4473 return err;
671b0060
DM		 4474
4475 for_each_port(adap, i) {
4476 struct port_info *pi = adap2pinfo(adap, i);
4477
4478 pi->rss = kcalloc(pi->rss_size, sizeof(u16), GFP_KERNEL);
4479 if (!pi->rss)
4480 return -ENOMEM;
671b0060
DM		 4481 }
4482 return 0;
4483}
4484
547fd272
HS		 4485static int cxgb4_get_pcie_dev_link_caps(struct adapter *adap,
4486 enum pci_bus_speed *speed,
4487 enum pcie_link_width *width)
4488{
4489 u32 lnkcap1, lnkcap2;
4490 int err1, err2;
4491
4492#define PCIE_MLW_CAP_SHIFT 4 /* start of MLW mask in link capabilities */
4493
4494 *speed = PCI_SPEED_UNKNOWN;
4495 *width = PCIE_LNK_WIDTH_UNKNOWN;
4496
4497 err1 = pcie_capability_read_dword(adap->pdev, PCI_EXP_LNKCAP,
4498 &lnkcap1);
4499 err2 = pcie_capability_read_dword(adap->pdev, PCI_EXP_LNKCAP2,
4500 &lnkcap2);
4501 if (!err2 && lnkcap2) { /* PCIe r3.0-compliant */
4502 if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_8_0GB)
4503 *speed = PCIE_SPEED_8_0GT;
4504 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_5_0GB)
4505 *speed = PCIE_SPEED_5_0GT;
4506 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_2_5GB)
4507 *speed = PCIE_SPEED_2_5GT;
4508 }
4509 if (!err1) {
4510 *width = (lnkcap1 & PCI_EXP_LNKCAP_MLW) >> PCIE_MLW_CAP_SHIFT;
4511 if (!lnkcap2) { /* pre-r3.0 */
4512 if (lnkcap1 & PCI_EXP_LNKCAP_SLS_5_0GB)
4513 *speed = PCIE_SPEED_5_0GT;
4514 else if (lnkcap1 & PCI_EXP_LNKCAP_SLS_2_5GB)
4515 *speed = PCIE_SPEED_2_5GT;
4516 }
4517 }
4518
4519 if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN)
4520 return err1 ? err1 : err2 ? err2 : -EINVAL;
4521 return 0;
4522}
4523
4524static void cxgb4_check_pcie_caps(struct adapter *adap)
4525{
4526 enum pcie_link_width width, width_cap;
4527 enum pci_bus_speed speed, speed_cap;
4528
4529#define PCIE_SPEED_STR(speed) \
4530 (speed == PCIE_SPEED_8_0GT ? "8.0GT/s" : \
4531 speed == PCIE_SPEED_5_0GT ? "5.0GT/s" : \
4532 speed == PCIE_SPEED_2_5GT ? "2.5GT/s" : \
4533 "Unknown")
4534
4535 if (cxgb4_get_pcie_dev_link_caps(adap, &speed_cap, &width_cap)) {
4536 dev_warn(adap->pdev_dev,
4537 "Unable to determine PCIe device BW capabilities\n");
4538 return;
4539 }
4540
4541 if (pcie_get_minimum_link(adap->pdev, &speed, &width) ||
4542 speed == PCI_SPEED_UNKNOWN || width == PCIE_LNK_WIDTH_UNKNOWN) {
4543 dev_warn(adap->pdev_dev,
4544 "Unable to determine PCI Express bandwidth.\n");
4545 return;
4546 }
4547
4548 dev_info(adap->pdev_dev, "PCIe link speed is %s, device supports %s\n",
4549 PCIE_SPEED_STR(speed), PCIE_SPEED_STR(speed_cap));
4550 dev_info(adap->pdev_dev, "PCIe link width is x%d, device supports x%d\n",
4551 width, width_cap);
4552 if (speed < speed_cap || width < width_cap)
4553 dev_info(adap->pdev_dev,
4554 "A slot with more lanes and/or higher speed is "
4555 "suggested for optimal performance.\n");
4556}
4557
0de72738
HS		 4558/* Dump basic information about the adapter */
4559static void print_adapter_info(struct adapter *adapter)
4560{
760446f9
GG		 4561 /* Hardware/Firmware/etc. Version/Revision IDs */
4562 t4_dump_version_info(adapter);
0de72738
HS		 4563
4564 /* Software/Hardware configuration */
4565 dev_info(adapter->pdev_dev, "Configuration: %sNIC %s, %s capable\n",
4566 is_offload(adapter) ? "R" : "",
4567 ((adapter->flags & USING_MSIX) ? "MSI-X" :
4568 (adapter->flags & USING_MSI) ? "MSI" : ""),
4569 is_offload(adapter) ? "Offload" : "non-Offload");
4570}
4571
91744948 4572static void print_port_info(const struct net_device *dev)
b8ff05a9 4573{
b8ff05a9 4574 char buf[80];
118969ed 4575 char *bufp = buf;
f1a051b9 4576 const char *spd = "";
118969ed
DM		 4577 const struct port_info *pi = netdev_priv(dev);
4578 const struct adapter *adap = pi->adapter;
f1a051b9
DM		 4579
4580 if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_2_5GB)
4581 spd = " 2.5 GT/s";
4582 else if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_5_0GB)
4583 spd = " 5 GT/s";
d2e752db
RD		 4584 else if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_8_0GB)
4585 spd = " 8 GT/s";
b8ff05a9 4586
118969ed 4587 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_100M)
5e78f7fd 4588 bufp += sprintf(bufp, "100M/");
118969ed 4589 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_1G)
5e78f7fd 4590 bufp += sprintf(bufp, "1G/");
118969ed
DM		 4591 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_10G)
4592 bufp += sprintf(bufp, "10G/");
9b86a8d1
HS		 4593 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_25G)
4594 bufp += sprintf(bufp, "25G/");
72aca4bf
KS		 4595 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_40G)
4596 bufp += sprintf(bufp, "40G/");
9b86a8d1
HS		 4597 if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_100G)
4598 bufp += sprintf(bufp, "100G/");
118969ed
DM		 4599 if (bufp != buf)
4600 --bufp;
72aca4bf 4601 sprintf(bufp, "BASE-%s", t4_get_port_type_description(pi->port_type));
118969ed 4602
0de72738
HS		 4603 netdev_info(dev, "%s: Chelsio %s (%s) %s\n",
4604 dev->name, adap->params.vpd.id, adap->name, buf);
b8ff05a9
DM		 4605}
4606
91744948 4607static void enable_pcie_relaxed_ordering(struct pci_dev *dev)
ef306b50 4608{
e5c8ae5f 4609 pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_RELAX_EN);
ef306b50
DM		 4610}
4611
06546391
DM		 4612/*
4613 * Free the following resources:
4614 * - memory used for tables
4615 * - MSI/MSI-X
4616 * - net devices
4617 * - resources FW is holding for us
4618 */
4619static void free_some_resources(struct adapter *adapter)
4620{
4621 unsigned int i;
4622
752ade68 4623 kvfree(adapter->l2t);
b72a32da 4624 t4_cleanup_sched(adapter);
752ade68 4625 kvfree(adapter->tids.tid_tab);
d8931847 4626 cxgb4_cleanup_tc_u32(adapter);
4b8e27a8
HS		 4627 kfree(adapter->sge.egr_map);
4628 kfree(adapter->sge.ingr_map);
4629 kfree(adapter->sge.starving_fl);
4630 kfree(adapter->sge.txq_maperr);
5b377d11
HS		 4631#ifdef CONFIG_DEBUG_FS
4632 kfree(adapter->sge.blocked_fl);
4633#endif
06546391
DM		 4634 disable_msi(adapter);
4635
4636 for_each_port(adapter, i)
671b0060 4637 if (adapter->port[i]) {
4f3a0fcf
HS		 4638 struct port_info *pi = adap2pinfo(adapter, i);
4639
4640 if (pi->viid != 0)
4641 t4_free_vi(adapter, adapter->mbox, adapter->pf,
4642 0, pi->viid);
671b0060 4643 kfree(adap2pinfo(adapter, i)->rss);
06546391 4644 free_netdev(adapter->port[i]);
671b0060 4645 }
06546391 4646 if (adapter->flags & FW_OK)
b2612722 4647 t4_fw_bye(adapter, adapter->pf);
06546391
DM		 4648}
4649
2ed28baa 4650#define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
35d35682 4651#define VLAN_FEAT (NETIF_F_SG | NETIF_F_IP_CSUM | TSO_FLAGS | \
b8ff05a9 4652 NETIF_F_IPV6_CSUM | NETIF_F_HIGHDMA)
22adfe0a 4653#define SEGMENT_SIZE 128
b8ff05a9 4654
d86bd29e
HS		 4655static int get_chip_type(struct pci_dev *pdev, u32 pl_rev)
4656{
d86bd29e
HS		 4657 u16 device_id;
4658
4659 /* Retrieve adapter's device ID */
4660 pci_read_config_word(pdev, PCI_DEVICE_ID, &device_id);
46cdc9be 4661
4662 switch (device_id >> 12) {
d86bd29e 4663 case CHELSIO_T4:
46cdc9be 4664 return CHELSIO_CHIP_CODE(CHELSIO_T4, pl_rev);
d86bd29e 4665 case CHELSIO_T5:
46cdc9be 4666 return CHELSIO_CHIP_CODE(CHELSIO_T5, pl_rev);
d86bd29e 4667 case CHELSIO_T6:
46cdc9be 4668 return CHELSIO_CHIP_CODE(CHELSIO_T6, pl_rev);
d86bd29e
HS		 4669 default:
4670 dev_err(&pdev->dev, "Device %d is not supported\n",
4671 device_id);
d86bd29e 4672 }
46cdc9be 4673 return -EINVAL;
d86bd29e
HS		 4674}
4675
b6244201 4676#ifdef CONFIG_PCI_IOV
e7b48a32
HS		 4677static void dummy_setup(struct net_device *dev)
4678{
4679 dev->type = ARPHRD_NONE;
4680 dev->mtu = 0;
4681 dev->hard_header_len = 0;
4682 dev->addr_len = 0;
4683 dev->tx_queue_len = 0;
4684 dev->flags |= IFF_NOARP;
4685 dev->priv_flags |= IFF_NO_QUEUE;
4686
4687 /* Initialize the device structure. */
4688 dev->netdev_ops = &cxgb4_mgmt_netdev_ops;
4689 dev->ethtool_ops = &cxgb4_mgmt_ethtool_ops;
cf124db5 4690 dev->needs_free_netdev = true;
e7b48a32
HS		 4691}
4692
4693static int config_mgmt_dev(struct pci_dev *pdev)
4694{
4695 struct adapter *adap = pci_get_drvdata(pdev);
4696 struct net_device *netdev;
4697 struct port_info *pi;
4698 char name[IFNAMSIZ];
4699 int err;
4700
4701 snprintf(name, IFNAMSIZ, "mgmtpf%d%d", adap->adap_idx, adap->pf);
038c35a8
GG		 4702 netdev = alloc_netdev(sizeof(struct port_info), name, NET_NAME_UNKNOWN,
4703 dummy_setup);
e7b48a32
HS		 4704 if (!netdev)
4705 return -ENOMEM;
4706
4707 pi = netdev_priv(netdev);
4708 pi->adapter = adap;
96fe11f2 4709 pi->port_id = adap->pf % adap->params.nports;
e7b48a32
HS		 4710 SET_NETDEV_DEV(netdev, &pdev->dev);
4711
4712 adap->port[0] = netdev;
4713
4714 err = register_netdev(adap->port[0]);
4715 if (err) {
4716 pr_info("Unable to register VF mgmt netdev %s\n", name);
4717 free_netdev(adap->port[0]);
4718 adap->port[0] = NULL;
4719 return err;
4720 }
4721 return 0;
4722}
4723
b6244201
HS		 4724static int cxgb4_iov_configure(struct pci_dev *pdev, int num_vfs)
4725{
7829451c 4726 struct adapter *adap = pci_get_drvdata(pdev);
b6244201
HS		 4727 int err = 0;
4728 int current_vfs = pci_num_vf(pdev);
4729 u32 pcie_fw;
b6244201 4730
7829451c 4731 pcie_fw = readl(adap->regs + PCIE_FW_A);
b6244201
HS		 4732 /* Check if cxgb4 is the MASTER and fw is initialized */
4733 if (!(pcie_fw & PCIE_FW_INIT_F) ||
4734 !(pcie_fw & PCIE_FW_MASTER_VLD_F) ||
4735 PCIE_FW_MASTER_G(pcie_fw) != 4) {
4736 dev_warn(&pdev->dev,
4737 "cxgb4 driver needs to be MASTER to support SRIOV\n");
4738 return -EOPNOTSUPP;
4739 }
4740
4741 /* If any of the VF's is already assigned to Guest OS, then
4742 * SRIOV for the same cannot be modified
4743 */
4744 if (current_vfs && pci_vfs_assigned(pdev)) {
4745 dev_err(&pdev->dev,
4746 "Cannot modify SR-IOV while VFs are assigned\n");
4747 num_vfs = current_vfs;
4748 return num_vfs;
4749 }
4750
4751 /* Disable SRIOV when zero is passed.
4752 * One needs to disable SRIOV before modifying it, else
4753 * stack throws the below warning:
4754 * " 'n' VFs already enabled. Disable before enabling 'm' VFs."
4755 */
4756 if (!num_vfs) {
4757 pci_disable_sriov(pdev);
e7b48a32 4758 if (adap->port[0]) {
7829451c 4759 unregister_netdev(adap->port[0]);
e7b48a32
HS		 4760 adap->port[0] = NULL;
4761 }
661dbeb9
HS		 4762 /* free VF resources */
4763 kfree(adap->vfinfo);
4764 adap->vfinfo = NULL;
4765 adap->num_vfs = 0;
b6244201
HS		 4766 return num_vfs;
4767 }
4768
4769 if (num_vfs != current_vfs) {
4770 err = pci_enable_sriov(pdev, num_vfs);
4771 if (err)
4772 return err;
7829451c 4773
661dbeb9 4774 adap->num_vfs = num_vfs;
e7b48a32
HS		 4775 err = config_mgmt_dev(pdev);
4776 if (err)
4777 return err;
b6244201 4778 }
661dbeb9
HS		 4779
4780 adap->vfinfo = kcalloc(adap->num_vfs,
4781 sizeof(struct vf_info), GFP_KERNEL);
4782 if (adap->vfinfo)
4783 fill_vf_station_mac_addr(adap);
b6244201
HS		 4784 return num_vfs;
4785}
4786#endif
4787
1dd06ae8 4788static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
b8ff05a9 4789{
22adfe0a 4790 int func, i, err, s_qpp, qpp, num_seg;
b8ff05a9 4791 struct port_info *pi;
c8f44aff 4792 bool highdma = false;
b8ff05a9 4793 struct adapter *adapter = NULL;
7829451c 4794 struct net_device *netdev;
d6ce2628 4795 void __iomem *regs;
d86bd29e
HS		 4796 u32 whoami, pl_rev;
4797 enum chip_type chip;
7829451c 4798 static int adap_idx = 1;
0a327889 4799#ifdef CONFIG_PCI_IOV
96fe11f2 4800 u32 v, port_vec;
0a327889 4801#endif
b8ff05a9
DM		 4802
4803 printk_once(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
4804
4805 err = pci_request_regions(pdev, KBUILD_MODNAME);
4806 if (err) {
4807 /* Just info, some other driver may have claimed the device. */
4808 dev_info(&pdev->dev, "cannot obtain PCI resources\n");
4809 return err;
4810 }
4811
b8ff05a9
DM		 4812 err = pci_enable_device(pdev);
4813 if (err) {
4814 dev_err(&pdev->dev, "cannot enable PCI device\n");
4815 goto out_release_regions;
4816 }
4817
d6ce2628
HS		 4818 regs = pci_ioremap_bar(pdev, 0);
4819 if (!regs) {
4820 dev_err(&pdev->dev, "cannot map device registers\n");
4821 err = -ENOMEM;
4822 goto out_disable_device;
4823 }
4824
8203b509
HS		 4825 err = t4_wait_dev_ready(regs);
4826 if (err < 0)
4827 goto out_unmap_bar0;
4828
d6ce2628 4829 /* We control everything through one PF */
d86bd29e
HS		 4830 whoami = readl(regs + PL_WHOAMI_A);
4831 pl_rev = REV_G(readl(regs + PL_REV_A));
4832 chip = get_chip_type(pdev, pl_rev);
4833 func = CHELSIO_CHIP_VERSION(chip) <= CHELSIO_T5 ?
4834 SOURCEPF_G(whoami) : T6_SOURCEPF_G(whoami);
d6ce2628 4835 if (func != ent->driver_data) {
7829451c 4836#ifndef CONFIG_PCI_IOV
d6ce2628 4837 iounmap(regs);
7829451c 4838#endif
d6ce2628
HS		 4839 pci_disable_device(pdev);
4840 pci_save_state(pdev); /* to restore SR-IOV later */
4841 goto sriov;
4842 }
4843
b8ff05a9 4844 if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
c8f44aff 4845 highdma = true;
b8ff05a9
DM		 4846 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
4847 if (err) {
4848 dev_err(&pdev->dev, "unable to obtain 64-bit DMA for "
4849 "coherent allocations\n");
d6ce2628 4850 goto out_unmap_bar0;
b8ff05a9
DM		 4851 }
4852 } else {
4853 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
4854 if (err) {
4855 dev_err(&pdev->dev, "no usable DMA configuration\n");
d6ce2628 4856 goto out_unmap_bar0;
b8ff05a9
DM		 4857 }
4858 }
4859
4860 pci_enable_pcie_error_reporting(pdev);
ef306b50 4861 enable_pcie_relaxed_ordering(pdev);
b8ff05a9
DM		 4862 pci_set_master(pdev);
4863 pci_save_state(pdev);
4864
4865 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
4866 if (!adapter) {
4867 err = -ENOMEM;
d6ce2628 4868 goto out_unmap_bar0;
b8ff05a9 4869 }
7829451c 4870 adap_idx++;
b8ff05a9 4871
29aaee65
AB		 4872 adapter->workq = create_singlethread_workqueue("cxgb4");
4873 if (!adapter->workq) {
4874 err = -ENOMEM;
4875 goto out_free_adapter;
4876 }
4877
7f080c3f
HS		 4878 adapter->mbox_log = kzalloc(sizeof(*adapter->mbox_log) +
4879 (sizeof(struct mbox_cmd) *
4880 T4_OS_LOG_MBOX_CMDS),
4881 GFP_KERNEL);
4882 if (!adapter->mbox_log) {
4883 err = -ENOMEM;
4884 goto out_free_adapter;
4885 }
4886 adapter->mbox_log->size = T4_OS_LOG_MBOX_CMDS;
4887
144be3d9
GS		 4888 /* PCI device has been enabled */
4889 adapter->flags |= DEV_ENABLED;
4890
d6ce2628 4891 adapter->regs = regs;
b8ff05a9
DM		 4892 adapter->pdev = pdev;
4893 adapter->pdev_dev = &pdev->dev;
0de72738 4894 adapter->name = pci_name(pdev);
3069ee9b 4895 adapter->mbox = func;
b2612722 4896 adapter->pf = func;
ea1e76f7 4897 adapter->msg_enable = DFLT_MSG_ENABLE;
b8ff05a9
DM		 4898 memset(adapter->chan_map, 0xff, sizeof(adapter->chan_map));
4899
4900 spin_lock_init(&adapter->stats_lock);
4901 spin_lock_init(&adapter->tid_release_lock);
e327c225 4902 spin_lock_init(&adapter->win0_lock);
4055ae5e
HS		 4903 spin_lock_init(&adapter->mbox_lock);
4904
4905 INIT_LIST_HEAD(&adapter->mlist.list);
b8ff05a9
DM		 4906
4907 INIT_WORK(&adapter->tid_release_task, process_tid_release_list);
881806bc
VP		 4908 INIT_WORK(&adapter->db_full_task, process_db_full);
4909 INIT_WORK(&adapter->db_drop_task, process_db_drop);
b8ff05a9
DM		 4910
4911 err = t4_prep_adapter(adapter);
4912 if (err)
d6ce2628
HS		 4913 goto out_free_adapter;
4914
22adfe0a 4915
d14807dd 4916 if (!is_t4(adapter->params.chip)) {
f612b815
HS		 4917 s_qpp = (QUEUESPERPAGEPF0_S +
4918 (QUEUESPERPAGEPF1_S - QUEUESPERPAGEPF0_S) *
b2612722 4919 adapter->pf);
f612b815
HS		 4920 qpp = 1 << QUEUESPERPAGEPF0_G(t4_read_reg(adapter,
4921 SGE_EGRESS_QUEUES_PER_PAGE_PF_A) >> s_qpp);
22adfe0a
SR		 4922 num_seg = PAGE_SIZE / SEGMENT_SIZE;
4923
4924 /* Each segment size is 128B. Write coalescing is enabled only
4925 * when SGE_EGRESS_QUEUES_PER_PAGE_PF reg value for the
4926 * queue is less no of segments that can be accommodated in
4927 * a page size.
4928 */
4929 if (qpp > num_seg) {
4930 dev_err(&pdev->dev,
4931 "Incorrect number of egress queues per page\n");
4932 err = -EINVAL;
d6ce2628 4933 goto out_free_adapter;
22adfe0a
SR		 4934 }
4935 adapter->bar2 = ioremap_wc(pci_resource_start(pdev, 2),
4936 pci_resource_len(pdev, 2));
4937 if (!adapter->bar2) {
4938 dev_err(&pdev->dev, "cannot map device bar2 region\n");
4939 err = -ENOMEM;
d6ce2628 4940 goto out_free_adapter;
22adfe0a
SR		 4941 }
4942 }
4943
636f9d37 4944 setup_memwin(adapter);
b8ff05a9 4945 err = adap_init0(adapter);
5b377d11
HS		 4946#ifdef CONFIG_DEBUG_FS
4947 bitmap_zero(adapter->sge.blocked_fl, adapter->sge.egr_sz);
4948#endif
636f9d37 4949 setup_memwin_rdma(adapter);
b8ff05a9
DM		 4950 if (err)
4951 goto out_unmap_bar;
4952
2a485cf7
HS		 4953 /* configure SGE_STAT_CFG_A to read WC stats */
4954 if (!is_t4(adapter->params.chip))
676d6a75
HS		 4955 t4_write_reg(adapter, SGE_STAT_CFG_A, STATSOURCE_T5_V(7) |
4956 (is_t5(adapter->params.chip) ? STATMODE_V(0) :
4957 T6_STATMODE_V(0)));
2a485cf7 4958
b8ff05a9 4959 for_each_port(adapter, i) {
b8ff05a9
DM		 4960 netdev = alloc_etherdev_mq(sizeof(struct port_info),
4961 MAX_ETH_QSETS);
4962 if (!netdev) {
4963 err = -ENOMEM;
4964 goto out_free_dev;
4965 }
4966
4967 SET_NETDEV_DEV(netdev, &pdev->dev);
4968
4969 adapter->port[i] = netdev;
4970 pi = netdev_priv(netdev);
4971 pi->adapter = adapter;
4972 pi->xact_addr_filt = -1;
b8ff05a9 4973 pi->port_id = i;
b8ff05a9
DM		 4974 netdev->irq = pdev->irq;
4975
2ed28baa
MM		 4976 netdev->hw_features = NETIF_F_SG | TSO_FLAGS |
4977 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
4978 NETIF_F_RXCSUM | NETIF_F_RXHASH |
d8931847
RL		 4979 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX |
4980 NETIF_F_HW_TC;
c8f44aff
MM		 4981 if (highdma)
4982 netdev->hw_features |= NETIF_F_HIGHDMA;
4983 netdev->features |= netdev->hw_features;
b8ff05a9
DM		 4984 netdev->vlan_features = netdev->features & VLAN_FEAT;
4985
01789349
JP		 4986 netdev->priv_flags |= IFF_UNICAST_FLT;
4987
d894be57
JW		 4988 /* MTU range: 81 - 9600 */
4989 netdev->min_mtu = 81;
4990 netdev->max_mtu = MAX_MTU;
4991
b8ff05a9 4992 netdev->netdev_ops = &cxgb4_netdev_ops;
688848b1
AB		 4993#ifdef CONFIG_CHELSIO_T4_DCB
4994 netdev->dcbnl_ops = &cxgb4_dcb_ops;
4995 cxgb4_dcb_state_init(netdev);
4996#endif
812034f1 4997 cxgb4_set_ethtool_ops(netdev);
b8ff05a9
DM		 4998 }
4999
5000 pci_set_drvdata(pdev, adapter);
5001
5002 if (adapter->flags & FW_OK) {
060e0c75 5003 err = t4_port_init(adapter, func, func, 0);
b8ff05a9
DM		 5004 if (err)
5005 goto out_free_dev;
098ef6c2
HS		 5006 } else if (adapter->params.nports == 1) {
5007 /* If we don't have a connection to the firmware -- possibly
5008 * because of an error -- grab the raw VPD parameters so we
5009 * can set the proper MAC Address on the debug network
5010 * interface that we've created.
5011 */
5012 u8 hw_addr[ETH_ALEN];
5013 u8 *na = adapter->params.vpd.na;
5014
5015 err = t4_get_raw_vpd_params(adapter, &adapter->params.vpd);
5016 if (!err) {
5017 for (i = 0; i < ETH_ALEN; i++)
5018 hw_addr[i] = (hex2val(na[2 * i + 0]) * 16 +
5019 hex2val(na[2 * i + 1]));
5020 t4_set_hw_addr(adapter, 0, hw_addr);
5021 }
b8ff05a9
DM		 5022 }
5023
098ef6c2 5024 /* Configure queues and allocate tables now, they can be needed as
b8ff05a9
DM		 5025 * soon as the first register_netdev completes.
5026 */
5027 cfg_queues(adapter);
5028
5be9ed8d 5029 adapter->l2t = t4_init_l2t(adapter->l2t_start, adapter->l2t_end);
b8ff05a9
DM		 5030 if (!adapter->l2t) {
5031 /* We tolerate a lack of L2T, giving up some functionality */
5032 dev_warn(&pdev->dev, "could not allocate L2T, continuing\n");
5033 adapter->params.offload = 0;
5034 }
5035
b5a02f50 5036#if IS_ENABLED(CONFIG_IPV6)
eb72f74f
HS		 5037 if ((CHELSIO_CHIP_VERSION(adapter->params.chip) <= CHELSIO_T5) &&
5038 (!(t4_read_reg(adapter, LE_DB_CONFIG_A) & ASLIPCOMPEN_F))) {
5039 /* CLIP functionality is not present in hardware,
5040 * hence disable all offload features
b5a02f50
AB		 5041 */
5042 dev_warn(&pdev->dev,
eb72f74f 5043 "CLIP not enabled in hardware, continuing\n");
b5a02f50 5044 adapter->params.offload = 0;
eb72f74f
HS		 5045 } else {
5046 adapter->clipt = t4_init_clip_tbl(adapter->clipt_start,
5047 adapter->clipt_end);
5048 if (!adapter->clipt) {
5049 /* We tolerate a lack of clip_table, giving up
5050 * some functionality
5051 */
5052 dev_warn(&pdev->dev,
5053 "could not allocate Clip table, continuing\n");
5054 adapter->params.offload = 0;
5055 }
b5a02f50
AB		 5056 }
5057#endif
b72a32da
RL		 5058
5059 for_each_port(adapter, i) {
5060 pi = adap2pinfo(adapter, i);
5061 pi->sched_tbl = t4_init_sched(adapter->params.nsched_cls);
5062 if (!pi->sched_tbl)
5063 dev_warn(&pdev->dev,
5064 "could not activate scheduling on port %d\n",
5065 i);
5066 }
5067
578b46b9 5068 if (tid_init(&adapter->tids) < 0) {
b8ff05a9
DM		 5069 dev_warn(&pdev->dev, "could not allocate TID table, "
5070 "continuing\n");
5071 adapter->params.offload = 0;
d8931847 5072 } else {
45da1ca2 5073 adapter->tc_u32 = cxgb4_init_tc_u32(adapter);
d8931847
RL		 5074 if (!adapter->tc_u32)
5075 dev_warn(&pdev->dev,
5076 "could not offload tc u32, continuing\n");
b8ff05a9
DM		 5077 }
5078
9a1bb9f6
HS		 5079 if (is_offload(adapter)) {
5080 if (t4_read_reg(adapter, LE_DB_CONFIG_A) & HASHEN_F) {
5081 u32 hash_base, hash_reg;
5082
5083 if (chip <= CHELSIO_T5) {
5084 hash_reg = LE_DB_TID_HASHBASE_A;
5085 hash_base = t4_read_reg(adapter, hash_reg);
5086 adapter->tids.hash_base = hash_base / 4;
5087 } else {
5088 hash_reg = T6_LE_DB_HASH_TID_BASE_A;
5089 hash_base = t4_read_reg(adapter, hash_reg);
5090 adapter->tids.hash_base = hash_base;
5091 }
5092 }
5093 }
5094
f7cabcdd
DM		 5095 /* See what interrupts we'll be using */
5096 if (msi > 1 && enable_msix(adapter) == 0)
5097 adapter->flags |= USING_MSIX;
94cdb8bb 5098 else if (msi > 0 && pci_enable_msi(pdev) == 0) {
f7cabcdd 5099 adapter->flags |= USING_MSI;
94cdb8bb
HS		 5100 if (msi > 1)
5101 free_msix_info(adapter);
5102 }
f7cabcdd 5103
547fd272
HS		 5104 /* check for PCI Express bandwidth capabiltites */
5105 cxgb4_check_pcie_caps(adapter);
5106
671b0060
DM		 5107 err = init_rss(adapter);
5108 if (err)
5109 goto out_free_dev;
5110
b8ff05a9
DM		 5111 /*
5112 * The card is now ready to go. If any errors occur during device
5113 * registration we do not fail the whole card but rather proceed only
5114 * with the ports we manage to register successfully. However we must
5115 * register at least one net device.
5116 */
5117 for_each_port(adapter, i) {
a57cabe0 5118 pi = adap2pinfo(adapter, i);
d2a007ab 5119 adapter->port[i]->dev_port = pi->lport;
a57cabe0
DM		 5120 netif_set_real_num_tx_queues(adapter->port[i], pi->nqsets);
5121 netif_set_real_num_rx_queues(adapter->port[i], pi->nqsets);
5122
b1a73af9
SM		 5123 netif_carrier_off(adapter->port[i]);
5124
b8ff05a9
DM		 5125 err = register_netdev(adapter->port[i]);
5126 if (err)
b1a3c2b6 5127 break;
b1a3c2b6
DM		 5128 adapter->chan_map[pi->tx_chan] = i;
5129 print_port_info(adapter->port[i]);
b8ff05a9 5130 }
b1a3c2b6 5131 if (i == 0) {
b8ff05a9
DM		 5132 dev_err(&pdev->dev, "could not register any net devices\n");
5133 goto out_free_dev;
5134 }
b1a3c2b6
DM		 5135 if (err) {
5136 dev_warn(&pdev->dev, "only %d net devices registered\n", i);
5137 err = 0;
6403eab1 5138 }
b8ff05a9
DM		 5139
5140 if (cxgb4_debugfs_root) {
5141 adapter->debugfs_root = debugfs_create_dir(pci_name(pdev),
5142 cxgb4_debugfs_root);
5143 setup_debugfs(adapter);
5144 }
5145
6482aa7c
DLR		 5146 /* PCIe EEH recovery on powerpc platforms needs fundamental reset */
5147 pdev->needs_freset = 1;
5148
0fbc81b3
HS		 5149 if (is_uld(adapter)) {
5150 mutex_lock(&uld_mutex);
5151 list_add_tail(&adapter->list_node, &adapter_list);
5152 mutex_unlock(&uld_mutex);
5153 }
b8ff05a9 5154
9c33e420
AG		 5155 if (!is_t4(adapter->params.chip))
5156 cxgb4_ptp_init(adapter);
5157
0de72738 5158 print_adapter_info(adapter);
0fbc81b3 5159 setup_fw_sge_queues(adapter);
7829451c 5160 return 0;
0de72738 5161
8e1e6059 5162sriov:
b8ff05a9 5163#ifdef CONFIG_PCI_IOV
7829451c
HS		 5164 adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
5165 if (!adapter) {
5166 err = -ENOMEM;
5167 goto free_pci_region;
5168 }
5169
7829451c
HS		 5170 adapter->pdev = pdev;
5171 adapter->pdev_dev = &pdev->dev;
5172 adapter->name = pci_name(pdev);
5173 adapter->mbox = func;
5174 adapter->pf = func;
5175 adapter->regs = regs;
e7b48a32 5176 adapter->adap_idx = adap_idx;
7829451c
HS		 5177 adapter->mbox_log = kzalloc(sizeof(*adapter->mbox_log) +
5178 (sizeof(struct mbox_cmd) *
5179 T4_OS_LOG_MBOX_CMDS),
5180 GFP_KERNEL);
5181 if (!adapter->mbox_log) {
5182 err = -ENOMEM;
e7b48a32 5183 goto free_adapter;
7829451c 5184 }
038c35a8
GG		 5185 spin_lock_init(&adapter->mbox_lock);
5186 INIT_LIST_HEAD(&adapter->mlist.list);
96fe11f2
GG		 5187
5188 v = FW_PARAMS_MNEM_V(FW_PARAMS_MNEM_DEV) |
5189 FW_PARAMS_PARAM_X_V(FW_PARAMS_PARAM_DEV_PORTVEC);
5190 err = t4_query_params(adapter, adapter->mbox, adapter->pf, 0, 1,
5191 &v, &port_vec);
5192 if (err < 0) {
5193 dev_err(adapter->pdev_dev, "Could not fetch port params\n");
d0417849 5194 goto free_mbox_log;
96fe11f2
GG		 5195 }
5196
5197 adapter->params.nports = hweight32(port_vec);
7829451c 5198 pci_set_drvdata(pdev, adapter);
7829451c
HS		 5199 return 0;
5200
d0417849
GG		 5201free_mbox_log:
5202 kfree(adapter->mbox_log);
7829451c
HS		 5203 free_adapter:
5204 kfree(adapter);
5205 free_pci_region:
5206 iounmap(regs);
5207 pci_disable_sriov(pdev);
5208 pci_release_regions(pdev);
5209 return err;
5210#else
b8ff05a9 5211 return 0;
7829451c 5212#endif
b8ff05a9
DM		 5213
5214 out_free_dev:
06546391 5215 free_some_resources(adapter);
94cdb8bb
HS		 5216 if (adapter->flags & USING_MSIX)
5217 free_msix_info(adapter);
0fbc81b3
HS		 5218 if (adapter->num_uld || adapter->num_ofld_uld)
5219 t4_uld_mem_free(adapter);
b8ff05a9 5220 out_unmap_bar:
d14807dd 5221 if (!is_t4(adapter->params.chip))
22adfe0a 5222 iounmap(adapter->bar2);
b8ff05a9 5223 out_free_adapter:
29aaee65
AB		 5224 if (adapter->workq)
5225 destroy_workqueue(adapter->workq);
5226
7f080c3f 5227 kfree(adapter->mbox_log);
b8ff05a9 5228 kfree(adapter);
d6ce2628
HS		 5229 out_unmap_bar0:
5230 iounmap(regs);
b8ff05a9
DM		 5231 out_disable_device:
5232 pci_disable_pcie_error_reporting(pdev);
5233 pci_disable_device(pdev);
5234 out_release_regions:
5235 pci_release_regions(pdev);
b8ff05a9
DM		 5236 return err;
5237}
5238
91744948 5239static void remove_one(struct pci_dev *pdev)
b8ff05a9
DM		 5240{
5241 struct adapter *adapter = pci_get_drvdata(pdev);
5242
7829451c
HS		 5243 if (!adapter) {
5244 pci_release_regions(pdev);
5245 return;
5246 }
636f9d37 5247
7829451c 5248 if (adapter->pf == 4) {
b8ff05a9
DM		 5249 int i;
5250
29aaee65
AB		 5251 /* Tear down per-adapter Work Queue first since it can contain
5252 * references to our adapter data structure.
5253 */
5254 destroy_workqueue(adapter->workq);
5255
6a146f3a 5256 if (is_uld(adapter)) {
b8ff05a9 5257 detach_ulds(adapter);
6a146f3a
GP		 5258 t4_uld_clean_up(adapter);
5259 }
b8ff05a9 5260
b37987e8
HS		 5261 disable_interrupts(adapter);
5262
b8ff05a9 5263 for_each_port(adapter, i)
8f3a7676 5264 if (adapter->port[i]->reg_state == NETREG_REGISTERED)
b8ff05a9
DM		 5265 unregister_netdev(adapter->port[i]);
5266
9f16dc2e 5267 debugfs_remove_recursive(adapter->debugfs_root);
b8ff05a9 5268
9c33e420
AG		 5269 if (!is_t4(adapter->params.chip))
5270 cxgb4_ptp_stop(adapter);
5271
f2b7e78d
VP		 5272 /* If we allocated filters, free up state associated with any
5273 * valid filters ...
5274 */
578b46b9 5275 clear_all_filters(adapter);
f2b7e78d 5276
aaefae9b
DM		 5277 if (adapter->flags & FULL_INIT_DONE)
5278 cxgb_down(adapter);
b8ff05a9 5279
94cdb8bb
HS		 5280 if (adapter->flags & USING_MSIX)
5281 free_msix_info(adapter);
0fbc81b3
HS		 5282 if (adapter->num_uld || adapter->num_ofld_uld)
5283 t4_uld_mem_free(adapter);
06546391 5284 free_some_resources(adapter);
b5a02f50
AB		 5285#if IS_ENABLED(CONFIG_IPV6)
5286 t4_cleanup_clip_tbl(adapter);
5287#endif
b8ff05a9 5288 iounmap(adapter->regs);
d14807dd 5289 if (!is_t4(adapter->params.chip))
22adfe0a 5290 iounmap(adapter->bar2);
b8ff05a9 5291 pci_disable_pcie_error_reporting(pdev);
144be3d9
GS		 5292 if ((adapter->flags & DEV_ENABLED)) {
5293 pci_disable_device(pdev);
5294 adapter->flags &= ~DEV_ENABLED;
5295 }
b8ff05a9 5296 pci_release_regions(pdev);
7f080c3f 5297 kfree(adapter->mbox_log);
ee9a33b2 5298 synchronize_rcu();
8b662fe7 5299 kfree(adapter);
7829451c
HS		 5300 }
5301#ifdef CONFIG_PCI_IOV
5302 else {
e7b48a32 5303 if (adapter->port[0])
7829451c 5304 unregister_netdev(adapter->port[0]);
7829451c 5305 iounmap(adapter->regs);
661dbeb9 5306 kfree(adapter->vfinfo);
d0417849 5307 kfree(adapter->mbox_log);
7829451c
HS		 5308 kfree(adapter);
5309 pci_disable_sriov(pdev);
b8ff05a9 5310 pci_release_regions(pdev);
7829451c
HS		 5311 }
5312#endif
b8ff05a9
DM		 5313}
5314
0fbc81b3
HS		 5315/* "Shutdown" quiesces the device, stopping Ingress Packet and Interrupt
5316 * delivery. This is essentially a stripped down version of the PCI remove()
5317 * function where we do the minimal amount of work necessary to shutdown any
5318 * further activity.
5319 */
5320static void shutdown_one(struct pci_dev *pdev)
5321{
5322 struct adapter *adapter = pci_get_drvdata(pdev);
5323
5324 /* As with remove_one() above (see extended comment), we only want do
5325 * do cleanup on PCI Devices which went all the way through init_one()
5326 * ...
5327 */
5328 if (!adapter) {
5329 pci_release_regions(pdev);
5330 return;
5331 }
5332
5333 if (adapter->pf == 4) {
5334 int i;
5335
5336 for_each_port(adapter, i)
5337 if (adapter->port[i]->reg_state == NETREG_REGISTERED)
5338 cxgb_close(adapter->port[i]);
5339
6a146f3a
GP		 5340 if (is_uld(adapter)) {
5341 detach_ulds(adapter);
5342 t4_uld_clean_up(adapter);
5343 }
5344
0fbc81b3
HS		 5345 disable_interrupts(adapter);
5346 disable_msi(adapter);
5347
5348 t4_sge_stop(adapter);
5349 if (adapter->flags & FW_OK)
5350 t4_fw_bye(adapter, adapter->mbox);
5351 }
5352#ifdef CONFIG_PCI_IOV
5353 else {
5354 if (adapter->port[0])
5355 unregister_netdev(adapter->port[0]);
5356 iounmap(adapter->regs);
5357 kfree(adapter->vfinfo);
d0417849 5358 kfree(adapter->mbox_log);
0fbc81b3
HS		 5359 kfree(adapter);
5360 pci_disable_sriov(pdev);
5361 pci_release_regions(pdev);
5362 }
5363#endif
5364}
5365
b8ff05a9
DM		 5366static struct pci_driver cxgb4_driver = {
5367 .name = KBUILD_MODNAME,
5368 .id_table = cxgb4_pci_tbl,
5369 .probe = init_one,
91744948 5370 .remove = remove_one,
0fbc81b3 5371 .shutdown = shutdown_one,
b6244201
HS		 5372#ifdef CONFIG_PCI_IOV
5373 .sriov_configure = cxgb4_iov_configure,
5374#endif
204dc3c0 5375 .err_handler = &cxgb4_eeh,
b8ff05a9
DM		 5376};
5377
5378static int __init cxgb4_init_module(void)
5379{
5380 int ret;
5381
5382 /* Debugfs support is optional, just warn if this fails */
5383 cxgb4_debugfs_root = debugfs_create_dir(KBUILD_MODNAME, NULL);
5384 if (!cxgb4_debugfs_root)
428ac43f 5385 pr_warn("could not create debugfs entry, continuing\n");
b8ff05a9
DM		 5386
5387 ret = pci_register_driver(&cxgb4_driver);
29aaee65 5388 if (ret < 0)
b8ff05a9 5389 debugfs_remove(cxgb4_debugfs_root);
01bcca68 5390
1bb60376 5391#if IS_ENABLED(CONFIG_IPV6)
b5a02f50
AB		 5392 if (!inet6addr_registered) {
5393 register_inet6addr_notifier(&cxgb4_inet6addr_notifier);
5394 inet6addr_registered = true;
5395 }
1bb60376 5396#endif
01bcca68 5397
b8ff05a9
DM		 5398 return ret;
5399}
5400
5401static void __exit cxgb4_cleanup_module(void)
5402{
1bb60376 5403#if IS_ENABLED(CONFIG_IPV6)
1793c798 5404 if (inet6addr_registered) {
b5a02f50
AB		 5405 unregister_inet6addr_notifier(&cxgb4_inet6addr_notifier);
5406 inet6addr_registered = false;
5407 }
1bb60376 5408#endif
b8ff05a9
DM		 5409 pci_unregister_driver(&cxgb4_driver);
5410 debugfs_remove(cxgb4_debugfs_root); /* NULL ok */
5411}
5412
5413module_init(cxgb4_init_module);
5414module_exit(cxgb4_cleanup_module);
A mirror of Linus' kernel repository