Merge branch 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/gerg/m68knommu
[linux-2.6] / drivers / net / cxgb3 / cxgb3_main.c
1 /*
2  * Copyright (c) 2003-2008 Chelsio, Inc. All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and/or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  */
32 #include <linux/module.h>
33 #include <linux/moduleparam.h>
34 #include <linux/init.h>
35 #include <linux/pci.h>
36 #include <linux/dma-mapping.h>
37 #include <linux/netdevice.h>
38 #include <linux/etherdevice.h>
39 #include <linux/if_vlan.h>
40 #include <linux/mii.h>
41 #include <linux/sockios.h>
42 #include <linux/workqueue.h>
43 #include <linux/proc_fs.h>
44 #include <linux/rtnetlink.h>
45 #include <linux/firmware.h>
46 #include <linux/log2.h>
47 #include <asm/uaccess.h>
48
49 #include "common.h"
50 #include "cxgb3_ioctl.h"
51 #include "regs.h"
52 #include "cxgb3_offload.h"
53 #include "version.h"
54
55 #include "cxgb3_ctl_defs.h"
56 #include "t3_cpl.h"
57 #include "firmware_exports.h"
58
59 enum {
60         MAX_TXQ_ENTRIES = 16384,
61         MAX_CTRL_TXQ_ENTRIES = 1024,
62         MAX_RSPQ_ENTRIES = 16384,
63         MAX_RX_BUFFERS = 16384,
64         MAX_RX_JUMBO_BUFFERS = 16384,
65         MIN_TXQ_ENTRIES = 4,
66         MIN_CTRL_TXQ_ENTRIES = 4,
67         MIN_RSPQ_ENTRIES = 32,
68         MIN_FL_ENTRIES = 32
69 };
70
71 #define PORT_MASK ((1 << MAX_NPORTS) - 1)
72
73 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
74                          NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
75                          NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
76
77 #define EEPROM_MAGIC 0x38E2F10C
78
79 #define CH_DEVICE(devid, idx) \
80         { PCI_VENDOR_ID_CHELSIO, devid, PCI_ANY_ID, PCI_ANY_ID, 0, 0, idx }
81
82 static const struct pci_device_id cxgb3_pci_tbl[] = {
83         CH_DEVICE(0x20, 0),     /* PE9000 */
84         CH_DEVICE(0x21, 1),     /* T302E */
85         CH_DEVICE(0x22, 2),     /* T310E */
86         CH_DEVICE(0x23, 3),     /* T320X */
87         CH_DEVICE(0x24, 1),     /* T302X */
88         CH_DEVICE(0x25, 3),     /* T320E */
89         CH_DEVICE(0x26, 2),     /* T310X */
90         CH_DEVICE(0x30, 2),     /* T3B10 */
91         CH_DEVICE(0x31, 3),     /* T3B20 */
92         CH_DEVICE(0x32, 1),     /* T3B02 */
93         {0,}
94 };
95
96 MODULE_DESCRIPTION(DRV_DESC);
97 MODULE_AUTHOR("Chelsio Communications");
98 MODULE_LICENSE("Dual BSD/GPL");
99 MODULE_VERSION(DRV_VERSION);
100 MODULE_DEVICE_TABLE(pci, cxgb3_pci_tbl);
101
102 static int dflt_msg_enable = DFLT_MSG_ENABLE;
103
104 module_param(dflt_msg_enable, int, 0644);
105 MODULE_PARM_DESC(dflt_msg_enable, "Chelsio T3 default message enable bitmap");
106
107 /*
108  * The driver uses the best interrupt scheme available on a platform in the
109  * order MSI-X, MSI, legacy pin interrupts.  This parameter determines which
110  * of these schemes the driver may consider as follows:
111  *
112  * msi = 2: choose from among all three options
113  * msi = 1: only consider MSI and pin interrupts
114  * msi = 0: force pin interrupts
115  */
116 static int msi = 2;
117
118 module_param(msi, int, 0644);
119 MODULE_PARM_DESC(msi, "whether to use MSI or MSI-X");
120
121 /*
122  * The driver enables offload as a default.
123  * To disable it, use ofld_disable = 1.
124  */
125
126 static int ofld_disable = 0;
127
128 module_param(ofld_disable, int, 0644);
129 MODULE_PARM_DESC(ofld_disable, "whether to enable offload at init time or not");
130
131 /*
132  * We have work elements that we need to cancel when an interface is taken
133  * down.  Normally the work elements would be executed by keventd but that
134  * can deadlock because of linkwatch.  If our close method takes the rtnl
135  * lock and linkwatch is ahead of our work elements in keventd, linkwatch
136  * will block keventd as it needs the rtnl lock, and we'll deadlock waiting
137  * for our work to complete.  Get our own work queue to solve this.
138  */
139 static struct workqueue_struct *cxgb3_wq;
140
141 /**
142  *      link_report - show link status and link speed/duplex
143  *      @p: the port whose settings are to be reported
144  *
145  *      Shows the link status, speed, and duplex of a port.
146  */
147 static void link_report(struct net_device *dev)
148 {
149         if (!netif_carrier_ok(dev))
150                 printk(KERN_INFO "%s: link down\n", dev->name);
151         else {
152                 const char *s = "10Mbps";
153                 const struct port_info *p = netdev_priv(dev);
154
155                 switch (p->link_config.speed) {
156                 case SPEED_10000:
157                         s = "10Gbps";
158                         break;
159                 case SPEED_1000:
160                         s = "1000Mbps";
161                         break;
162                 case SPEED_100:
163                         s = "100Mbps";
164                         break;
165                 }
166
167                 printk(KERN_INFO "%s: link up, %s, %s-duplex\n", dev->name, s,
168                        p->link_config.duplex == DUPLEX_FULL ? "full" : "half");
169         }
170 }
171
172 /**
173  *      t3_os_link_changed - handle link status changes
174  *      @adapter: the adapter associated with the link change
175  *      @port_id: the port index whose limk status has changed
176  *      @link_stat: the new status of the link
177  *      @speed: the new speed setting
178  *      @duplex: the new duplex setting
179  *      @pause: the new flow-control setting
180  *
181  *      This is the OS-dependent handler for link status changes.  The OS
182  *      neutral handler takes care of most of the processing for these events,
183  *      then calls this handler for any OS-specific processing.
184  */
185 void t3_os_link_changed(struct adapter *adapter, int port_id, int link_stat,
186                         int speed, int duplex, int pause)
187 {
188         struct net_device *dev = adapter->port[port_id];
189         struct port_info *pi = netdev_priv(dev);
190         struct cmac *mac = &pi->mac;
191
192         /* Skip changes from disabled ports. */
193         if (!netif_running(dev))
194                 return;
195
196         if (link_stat != netif_carrier_ok(dev)) {
197                 if (link_stat) {
198                         t3_mac_enable(mac, MAC_DIRECTION_RX);
199                         netif_carrier_on(dev);
200                 } else {
201                         netif_carrier_off(dev);
202                         pi->phy.ops->power_down(&pi->phy, 1);
203                         t3_mac_disable(mac, MAC_DIRECTION_RX);
204                         t3_link_start(&pi->phy, mac, &pi->link_config);
205                 }
206
207                 link_report(dev);
208         }
209 }
210
211 /**
212  *      t3_os_phymod_changed - handle PHY module changes
213  *      @phy: the PHY reporting the module change
214  *      @mod_type: new module type
215  *
216  *      This is the OS-dependent handler for PHY module changes.  It is
217  *      invoked when a PHY module is removed or inserted for any OS-specific
218  *      processing.
219  */
220 void t3_os_phymod_changed(struct adapter *adap, int port_id)
221 {
222         static const char *mod_str[] = {
223                 NULL, "SR", "LR", "LRM", "TWINAX", "TWINAX", "unknown"
224         };
225
226         const struct net_device *dev = adap->port[port_id];
227         const struct port_info *pi = netdev_priv(dev);
228
229         if (pi->phy.modtype == phy_modtype_none)
230                 printk(KERN_INFO "%s: PHY module unplugged\n", dev->name);
231         else
232                 printk(KERN_INFO "%s: %s PHY module inserted\n", dev->name,
233                        mod_str[pi->phy.modtype]);
234 }
235
236 static void cxgb_set_rxmode(struct net_device *dev)
237 {
238         struct t3_rx_mode rm;
239         struct port_info *pi = netdev_priv(dev);
240
241         init_rx_mode(&rm, dev, dev->mc_list);
242         t3_mac_set_rx_mode(&pi->mac, &rm);
243 }
244
245 /**
246  *      link_start - enable a port
247  *      @dev: the device to enable
248  *
249  *      Performs the MAC and PHY actions needed to enable a port.
250  */
251 static void link_start(struct net_device *dev)
252 {
253         struct t3_rx_mode rm;
254         struct port_info *pi = netdev_priv(dev);
255         struct cmac *mac = &pi->mac;
256
257         init_rx_mode(&rm, dev, dev->mc_list);
258         t3_mac_reset(mac);
259         t3_mac_set_mtu(mac, dev->mtu);
260         t3_mac_set_address(mac, 0, dev->dev_addr);
261         t3_mac_set_rx_mode(mac, &rm);
262         t3_link_start(&pi->phy, mac, &pi->link_config);
263         t3_mac_enable(mac, MAC_DIRECTION_RX | MAC_DIRECTION_TX);
264 }
265
266 static inline void cxgb_disable_msi(struct adapter *adapter)
267 {
268         if (adapter->flags & USING_MSIX) {
269                 pci_disable_msix(adapter->pdev);
270                 adapter->flags &= ~USING_MSIX;
271         } else if (adapter->flags & USING_MSI) {
272                 pci_disable_msi(adapter->pdev);
273                 adapter->flags &= ~USING_MSI;
274         }
275 }
276
277 /*
278  * Interrupt handler for asynchronous events used with MSI-X.
279  */
280 static irqreturn_t t3_async_intr_handler(int irq, void *cookie)
281 {
282         t3_slow_intr_handler(cookie);
283         return IRQ_HANDLED;
284 }
285
286 /*
287  * Name the MSI-X interrupts.
288  */
289 static void name_msix_vecs(struct adapter *adap)
290 {
291         int i, j, msi_idx = 1, n = sizeof(adap->msix_info[0].desc) - 1;
292
293         snprintf(adap->msix_info[0].desc, n, "%s", adap->name);
294         adap->msix_info[0].desc[n] = 0;
295
296         for_each_port(adap, j) {
297                 struct net_device *d = adap->port[j];
298                 const struct port_info *pi = netdev_priv(d);
299
300                 for (i = 0; i < pi->nqsets; i++, msi_idx++) {
301                         snprintf(adap->msix_info[msi_idx].desc, n,
302                                  "%s-%d", d->name, pi->first_qset + i);
303                         adap->msix_info[msi_idx].desc[n] = 0;
304                 }
305         }
306 }
307
308 static int request_msix_data_irqs(struct adapter *adap)
309 {
310         int i, j, err, qidx = 0;
311
312         for_each_port(adap, i) {
313                 int nqsets = adap2pinfo(adap, i)->nqsets;
314
315                 for (j = 0; j < nqsets; ++j) {
316                         err = request_irq(adap->msix_info[qidx + 1].vec,
317                                           t3_intr_handler(adap,
318                                                           adap->sge.qs[qidx].
319                                                           rspq.polling), 0,
320                                           adap->msix_info[qidx + 1].desc,
321                                           &adap->sge.qs[qidx]);
322                         if (err) {
323                                 while (--qidx >= 0)
324                                         free_irq(adap->msix_info[qidx + 1].vec,
325                                                  &adap->sge.qs[qidx]);
326                                 return err;
327                         }
328                         qidx++;
329                 }
330         }
331         return 0;
332 }
333
334 static void free_irq_resources(struct adapter *adapter)
335 {
336         if (adapter->flags & USING_MSIX) {
337                 int i, n = 0;
338
339                 free_irq(adapter->msix_info[0].vec, adapter);
340                 for_each_port(adapter, i)
341                     n += adap2pinfo(adapter, i)->nqsets;
342
343                 for (i = 0; i < n; ++i)
344                         free_irq(adapter->msix_info[i + 1].vec,
345                                  &adapter->sge.qs[i]);
346         } else
347                 free_irq(adapter->pdev->irq, adapter);
348 }
349
350 static int await_mgmt_replies(struct adapter *adap, unsigned long init_cnt,
351                               unsigned long n)
352 {
353         int attempts = 5;
354
355         while (adap->sge.qs[0].rspq.offload_pkts < init_cnt + n) {
356                 if (!--attempts)
357                         return -ETIMEDOUT;
358                 msleep(10);
359         }
360         return 0;
361 }
362
363 static int init_tp_parity(struct adapter *adap)
364 {
365         int i;
366         struct sk_buff *skb;
367         struct cpl_set_tcb_field *greq;
368         unsigned long cnt = adap->sge.qs[0].rspq.offload_pkts;
369
370         t3_tp_set_offload_mode(adap, 1);
371
372         for (i = 0; i < 16; i++) {
373                 struct cpl_smt_write_req *req;
374
375                 skb = alloc_skb(sizeof(*req), GFP_KERNEL | __GFP_NOFAIL);
376                 req = (struct cpl_smt_write_req *)__skb_put(skb, sizeof(*req));
377                 memset(req, 0, sizeof(*req));
378                 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
379                 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SMT_WRITE_REQ, i));
380                 req->iff = i;
381                 t3_mgmt_tx(adap, skb);
382         }
383
384         for (i = 0; i < 2048; i++) {
385                 struct cpl_l2t_write_req *req;
386
387                 skb = alloc_skb(sizeof(*req), GFP_KERNEL | __GFP_NOFAIL);
388                 req = (struct cpl_l2t_write_req *)__skb_put(skb, sizeof(*req));
389                 memset(req, 0, sizeof(*req));
390                 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
391                 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_L2T_WRITE_REQ, i));
392                 req->params = htonl(V_L2T_W_IDX(i));
393                 t3_mgmt_tx(adap, skb);
394         }
395
396         for (i = 0; i < 2048; i++) {
397                 struct cpl_rte_write_req *req;
398
399                 skb = alloc_skb(sizeof(*req), GFP_KERNEL | __GFP_NOFAIL);
400                 req = (struct cpl_rte_write_req *)__skb_put(skb, sizeof(*req));
401                 memset(req, 0, sizeof(*req));
402                 req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
403                 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RTE_WRITE_REQ, i));
404                 req->l2t_idx = htonl(V_L2T_W_IDX(i));
405                 t3_mgmt_tx(adap, skb);
406         }
407
408         skb = alloc_skb(sizeof(*greq), GFP_KERNEL | __GFP_NOFAIL);
409         greq = (struct cpl_set_tcb_field *)__skb_put(skb, sizeof(*greq));
410         memset(greq, 0, sizeof(*greq));
411         greq->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
412         OPCODE_TID(greq) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, 0));
413         greq->mask = cpu_to_be64(1);
414         t3_mgmt_tx(adap, skb);
415
416         i = await_mgmt_replies(adap, cnt, 16 + 2048 + 2048 + 1);
417         t3_tp_set_offload_mode(adap, 0);
418         return i;
419 }
420
421 /**
422  *      setup_rss - configure RSS
423  *      @adap: the adapter
424  *
425  *      Sets up RSS to distribute packets to multiple receive queues.  We
426  *      configure the RSS CPU lookup table to distribute to the number of HW
427  *      receive queues, and the response queue lookup table to narrow that
428  *      down to the response queues actually configured for each port.
429  *      We always configure the RSS mapping for two ports since the mapping
430  *      table has plenty of entries.
431  */
432 static void setup_rss(struct adapter *adap)
433 {
434         int i;
435         unsigned int nq0 = adap2pinfo(adap, 0)->nqsets;
436         unsigned int nq1 = adap->port[1] ? adap2pinfo(adap, 1)->nqsets : 1;
437         u8 cpus[SGE_QSETS + 1];
438         u16 rspq_map[RSS_TABLE_SIZE];
439
440         for (i = 0; i < SGE_QSETS; ++i)
441                 cpus[i] = i;
442         cpus[SGE_QSETS] = 0xff; /* terminator */
443
444         for (i = 0; i < RSS_TABLE_SIZE / 2; ++i) {
445                 rspq_map[i] = i % nq0;
446                 rspq_map[i + RSS_TABLE_SIZE / 2] = (i % nq1) + nq0;
447         }
448
449         t3_config_rss(adap, F_RQFEEDBACKENABLE | F_TNLLKPEN | F_TNLMAPEN |
450                       F_TNLPRTEN | F_TNL2TUPEN | F_TNL4TUPEN |
451                       V_RRCPLCPUSIZE(6) | F_HASHTOEPLITZ, cpus, rspq_map);
452 }
453
454 static void init_napi(struct adapter *adap)
455 {
456         int i;
457
458         for (i = 0; i < SGE_QSETS; i++) {
459                 struct sge_qset *qs = &adap->sge.qs[i];
460
461                 if (qs->adap)
462                         netif_napi_add(qs->netdev, &qs->napi, qs->napi.poll,
463                                        64);
464         }
465
466         /*
467          * netif_napi_add() can be called only once per napi_struct because it
468          * adds each new napi_struct to a list.  Be careful not to call it a
469          * second time, e.g., during EEH recovery, by making a note of it.
470          */
471         adap->flags |= NAPI_INIT;
472 }
473
474 /*
475  * Wait until all NAPI handlers are descheduled.  This includes the handlers of
476  * both netdevices representing interfaces and the dummy ones for the extra
477  * queues.
478  */
479 static void quiesce_rx(struct adapter *adap)
480 {
481         int i;
482
483         for (i = 0; i < SGE_QSETS; i++)
484                 if (adap->sge.qs[i].adap)
485                         napi_disable(&adap->sge.qs[i].napi);
486 }
487
488 static void enable_all_napi(struct adapter *adap)
489 {
490         int i;
491         for (i = 0; i < SGE_QSETS; i++)
492                 if (adap->sge.qs[i].adap)
493                         napi_enable(&adap->sge.qs[i].napi);
494 }
495
496 /**
497  *      set_qset_lro - Turn a queue set's LRO capability on and off
498  *      @dev: the device the qset is attached to
499  *      @qset_idx: the queue set index
500  *      @val: the LRO switch
501  *
502  *      Sets LRO on or off for a particular queue set.
503  *      the device's features flag is updated to reflect the LRO
504  *      capability when all queues belonging to the device are
505  *      in the same state.
506  */
507 static void set_qset_lro(struct net_device *dev, int qset_idx, int val)
508 {
509         struct port_info *pi = netdev_priv(dev);
510         struct adapter *adapter = pi->adapter;
511         int i, lro_on = 1;
512
513         adapter->params.sge.qset[qset_idx].lro = !!val;
514         adapter->sge.qs[qset_idx].lro_enabled = !!val;
515
516         /* let ethtool report LRO on only if all queues are LRO enabled */
517         for (i = pi->first_qset; i < pi->first_qset + pi->nqsets; ++i)
518                 lro_on &= adapter->params.sge.qset[i].lro;
519
520         if (lro_on)
521                 dev->features |= NETIF_F_LRO;
522         else
523                 dev->features &= ~NETIF_F_LRO;
524 }
525
526 /**
527  *      setup_sge_qsets - configure SGE Tx/Rx/response queues
528  *      @adap: the adapter
529  *
530  *      Determines how many sets of SGE queues to use and initializes them.
531  *      We support multiple queue sets per port if we have MSI-X, otherwise
532  *      just one queue set per port.
533  */
534 static int setup_sge_qsets(struct adapter *adap)
535 {
536         int i, j, err, irq_idx = 0, qset_idx = 0;
537         unsigned int ntxq = SGE_TXQ_PER_SET;
538
539         if (adap->params.rev > 0 && !(adap->flags & USING_MSI))
540                 irq_idx = -1;
541
542         for_each_port(adap, i) {
543                 struct net_device *dev = adap->port[i];
544                 struct port_info *pi = netdev_priv(dev);
545
546                 pi->qs = &adap->sge.qs[pi->first_qset];
547                 for (j = pi->first_qset; j < pi->first_qset + pi->nqsets;
548                      ++j, ++qset_idx) {
549                         set_qset_lro(dev, qset_idx, pi->rx_offload & T3_LRO);
550                         err = t3_sge_alloc_qset(adap, qset_idx, 1,
551                                 (adap->flags & USING_MSIX) ? qset_idx + 1 :
552                                                              irq_idx,
553                                 &adap->params.sge.qset[qset_idx], ntxq, dev,
554                                 netdev_get_tx_queue(dev, j));
555                         if (err) {
556                                 t3_stop_sge_timers(adap);
557                                 t3_free_sge_resources(adap);
558                                 return err;
559                         }
560                 }
561         }
562
563         return 0;
564 }
565
566 static ssize_t attr_show(struct device *d, char *buf,
567                          ssize_t(*format) (struct net_device *, char *))
568 {
569         ssize_t len;
570
571         /* Synchronize with ioctls that may shut down the device */
572         rtnl_lock();
573         len = (*format) (to_net_dev(d), buf);
574         rtnl_unlock();
575         return len;
576 }
577
578 static ssize_t attr_store(struct device *d,
579                           const char *buf, size_t len,
580                           ssize_t(*set) (struct net_device *, unsigned int),
581                           unsigned int min_val, unsigned int max_val)
582 {
583         char *endp;
584         ssize_t ret;
585         unsigned int val;
586
587         if (!capable(CAP_NET_ADMIN))
588                 return -EPERM;
589
590         val = simple_strtoul(buf, &endp, 0);
591         if (endp == buf || val < min_val || val > max_val)
592                 return -EINVAL;
593
594         rtnl_lock();
595         ret = (*set) (to_net_dev(d), val);
596         if (!ret)
597                 ret = len;
598         rtnl_unlock();
599         return ret;
600 }
601
602 #define CXGB3_SHOW(name, val_expr) \
603 static ssize_t format_##name(struct net_device *dev, char *buf) \
604 { \
605         struct port_info *pi = netdev_priv(dev); \
606         struct adapter *adap = pi->adapter; \
607         return sprintf(buf, "%u\n", val_expr); \
608 } \
609 static ssize_t show_##name(struct device *d, struct device_attribute *attr, \
610                            char *buf) \
611 { \
612         return attr_show(d, buf, format_##name); \
613 }
614
615 static ssize_t set_nfilters(struct net_device *dev, unsigned int val)
616 {
617         struct port_info *pi = netdev_priv(dev);
618         struct adapter *adap = pi->adapter;
619         int min_tids = is_offload(adap) ? MC5_MIN_TIDS : 0;
620
621         if (adap->flags & FULL_INIT_DONE)
622                 return -EBUSY;
623         if (val && adap->params.rev == 0)
624                 return -EINVAL;
625         if (val > t3_mc5_size(&adap->mc5) - adap->params.mc5.nservers -
626             min_tids)
627                 return -EINVAL;
628         adap->params.mc5.nfilters = val;
629         return 0;
630 }
631
632 static ssize_t store_nfilters(struct device *d, struct device_attribute *attr,
633                               const char *buf, size_t len)
634 {
635         return attr_store(d, buf, len, set_nfilters, 0, ~0);
636 }
637
638 static ssize_t set_nservers(struct net_device *dev, unsigned int val)
639 {
640         struct port_info *pi = netdev_priv(dev);
641         struct adapter *adap = pi->adapter;
642
643         if (adap->flags & FULL_INIT_DONE)
644                 return -EBUSY;
645         if (val > t3_mc5_size(&adap->mc5) - adap->params.mc5.nfilters -
646             MC5_MIN_TIDS)
647                 return -EINVAL;
648         adap->params.mc5.nservers = val;
649         return 0;
650 }
651
652 static ssize_t store_nservers(struct device *d, struct device_attribute *attr,
653                               const char *buf, size_t len)
654 {
655         return attr_store(d, buf, len, set_nservers, 0, ~0);
656 }
657
658 #define CXGB3_ATTR_R(name, val_expr) \
659 CXGB3_SHOW(name, val_expr) \
660 static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
661
662 #define CXGB3_ATTR_RW(name, val_expr, store_method) \
663 CXGB3_SHOW(name, val_expr) \
664 static DEVICE_ATTR(name, S_IRUGO | S_IWUSR, show_##name, store_method)
665
666 CXGB3_ATTR_R(cam_size, t3_mc5_size(&adap->mc5));
667 CXGB3_ATTR_RW(nfilters, adap->params.mc5.nfilters, store_nfilters);
668 CXGB3_ATTR_RW(nservers, adap->params.mc5.nservers, store_nservers);
669
670 static struct attribute *cxgb3_attrs[] = {
671         &dev_attr_cam_size.attr,
672         &dev_attr_nfilters.attr,
673         &dev_attr_nservers.attr,
674         NULL
675 };
676
677 static struct attribute_group cxgb3_attr_group = {.attrs = cxgb3_attrs };
678
679 static ssize_t tm_attr_show(struct device *d,
680                             char *buf, int sched)
681 {
682         struct port_info *pi = netdev_priv(to_net_dev(d));
683         struct adapter *adap = pi->adapter;
684         unsigned int v, addr, bpt, cpt;
685         ssize_t len;
686
687         addr = A_TP_TX_MOD_Q1_Q0_RATE_LIMIT - sched / 2;
688         rtnl_lock();
689         t3_write_reg(adap, A_TP_TM_PIO_ADDR, addr);
690         v = t3_read_reg(adap, A_TP_TM_PIO_DATA);
691         if (sched & 1)
692                 v >>= 16;
693         bpt = (v >> 8) & 0xff;
694         cpt = v & 0xff;
695         if (!cpt)
696                 len = sprintf(buf, "disabled\n");
697         else {
698                 v = (adap->params.vpd.cclk * 1000) / cpt;
699                 len = sprintf(buf, "%u Kbps\n", (v * bpt) / 125);
700         }
701         rtnl_unlock();
702         return len;
703 }
704
705 static ssize_t tm_attr_store(struct device *d,
706                              const char *buf, size_t len, int sched)
707 {
708         struct port_info *pi = netdev_priv(to_net_dev(d));
709         struct adapter *adap = pi->adapter;
710         unsigned int val;
711         char *endp;
712         ssize_t ret;
713
714         if (!capable(CAP_NET_ADMIN))
715                 return -EPERM;
716
717         val = simple_strtoul(buf, &endp, 0);
718         if (endp == buf || val > 10000000)
719                 return -EINVAL;
720
721         rtnl_lock();
722         ret = t3_config_sched(adap, val, sched);
723         if (!ret)
724                 ret = len;
725         rtnl_unlock();
726         return ret;
727 }
728
729 #define TM_ATTR(name, sched) \
730 static ssize_t show_##name(struct device *d, struct device_attribute *attr, \
731                            char *buf) \
732 { \
733         return tm_attr_show(d, buf, sched); \
734 } \
735 static ssize_t store_##name(struct device *d, struct device_attribute *attr, \
736                             const char *buf, size_t len) \
737 { \
738         return tm_attr_store(d, buf, len, sched); \
739 } \
740 static DEVICE_ATTR(name, S_IRUGO | S_IWUSR, show_##name, store_##name)
741
742 TM_ATTR(sched0, 0);
743 TM_ATTR(sched1, 1);
744 TM_ATTR(sched2, 2);
745 TM_ATTR(sched3, 3);
746 TM_ATTR(sched4, 4);
747 TM_ATTR(sched5, 5);
748 TM_ATTR(sched6, 6);
749 TM_ATTR(sched7, 7);
750
751 static struct attribute *offload_attrs[] = {
752         &dev_attr_sched0.attr,
753         &dev_attr_sched1.attr,
754         &dev_attr_sched2.attr,
755         &dev_attr_sched3.attr,
756         &dev_attr_sched4.attr,
757         &dev_attr_sched5.attr,
758         &dev_attr_sched6.attr,
759         &dev_attr_sched7.attr,
760         NULL
761 };
762
763 static struct attribute_group offload_attr_group = {.attrs = offload_attrs };
764
765 /*
766  * Sends an sk_buff to an offload queue driver
767  * after dealing with any active network taps.
768  */
769 static inline int offload_tx(struct t3cdev *tdev, struct sk_buff *skb)
770 {
771         int ret;
772
773         local_bh_disable();
774         ret = t3_offload_tx(tdev, skb);
775         local_bh_enable();
776         return ret;
777 }
778
779 static int write_smt_entry(struct adapter *adapter, int idx)
780 {
781         struct cpl_smt_write_req *req;
782         struct sk_buff *skb = alloc_skb(sizeof(*req), GFP_KERNEL);
783
784         if (!skb)
785                 return -ENOMEM;
786
787         req = (struct cpl_smt_write_req *)__skb_put(skb, sizeof(*req));
788         req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
789         OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SMT_WRITE_REQ, idx));
790         req->mtu_idx = NMTUS - 1;       /* should be 0 but there's a T3 bug */
791         req->iff = idx;
792         memset(req->src_mac1, 0, sizeof(req->src_mac1));
793         memcpy(req->src_mac0, adapter->port[idx]->dev_addr, ETH_ALEN);
794         skb->priority = 1;
795         offload_tx(&adapter->tdev, skb);
796         return 0;
797 }
798
799 static int init_smt(struct adapter *adapter)
800 {
801         int i;
802
803         for_each_port(adapter, i)
804             write_smt_entry(adapter, i);
805         return 0;
806 }
807
808 static void init_port_mtus(struct adapter *adapter)
809 {
810         unsigned int mtus = adapter->port[0]->mtu;
811
812         if (adapter->port[1])
813                 mtus |= adapter->port[1]->mtu << 16;
814         t3_write_reg(adapter, A_TP_MTU_PORT_TABLE, mtus);
815 }
816
817 static int send_pktsched_cmd(struct adapter *adap, int sched, int qidx, int lo,
818                               int hi, int port)
819 {
820         struct sk_buff *skb;
821         struct mngt_pktsched_wr *req;
822         int ret;
823
824         skb = alloc_skb(sizeof(*req), GFP_KERNEL | __GFP_NOFAIL);
825         req = (struct mngt_pktsched_wr *)skb_put(skb, sizeof(*req));
826         req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_MNGT));
827         req->mngt_opcode = FW_MNGTOPCODE_PKTSCHED_SET;
828         req->sched = sched;
829         req->idx = qidx;
830         req->min = lo;
831         req->max = hi;
832         req->binding = port;
833         ret = t3_mgmt_tx(adap, skb);
834
835         return ret;
836 }
837
838 static int bind_qsets(struct adapter *adap)
839 {
840         int i, j, err = 0;
841
842         for_each_port(adap, i) {
843                 const struct port_info *pi = adap2pinfo(adap, i);
844
845                 for (j = 0; j < pi->nqsets; ++j) {
846                         int ret = send_pktsched_cmd(adap, 1,
847                                                     pi->first_qset + j, -1,
848                                                     -1, i);
849                         if (ret)
850                                 err = ret;
851                 }
852         }
853
854         return err;
855 }
856
857 #define FW_FNAME "cxgb3/t3fw-%d.%d.%d.bin"
858 #define TPSRAM_NAME "cxgb3/t3%c_psram-%d.%d.%d.bin"
859
860 static int upgrade_fw(struct adapter *adap)
861 {
862         int ret;
863         char buf[64];
864         const struct firmware *fw;
865         struct device *dev = &adap->pdev->dev;
866
867         snprintf(buf, sizeof(buf), FW_FNAME, FW_VERSION_MAJOR,
868                  FW_VERSION_MINOR, FW_VERSION_MICRO);
869         ret = request_firmware(&fw, buf, dev);
870         if (ret < 0) {
871                 dev_err(dev, "could not upgrade firmware: unable to load %s\n",
872                         buf);
873                 return ret;
874         }
875         ret = t3_load_fw(adap, fw->data, fw->size);
876         release_firmware(fw);
877
878         if (ret == 0)
879                 dev_info(dev, "successful upgrade to firmware %d.%d.%d\n",
880                          FW_VERSION_MAJOR, FW_VERSION_MINOR, FW_VERSION_MICRO);
881         else
882                 dev_err(dev, "failed to upgrade to firmware %d.%d.%d\n",
883                         FW_VERSION_MAJOR, FW_VERSION_MINOR, FW_VERSION_MICRO);
884
885         return ret;
886 }
887
888 static inline char t3rev2char(struct adapter *adapter)
889 {
890         char rev = 0;
891
892         switch(adapter->params.rev) {
893         case T3_REV_B:
894         case T3_REV_B2:
895                 rev = 'b';
896                 break;
897         case T3_REV_C:
898                 rev = 'c';
899                 break;
900         }
901         return rev;
902 }
903
904 static int update_tpsram(struct adapter *adap)
905 {
906         const struct firmware *tpsram;
907         char buf[64];
908         struct device *dev = &adap->pdev->dev;
909         int ret;
910         char rev;
911
912         rev = t3rev2char(adap);
913         if (!rev)
914                 return 0;
915
916         snprintf(buf, sizeof(buf), TPSRAM_NAME, rev,
917                  TP_VERSION_MAJOR, TP_VERSION_MINOR, TP_VERSION_MICRO);
918
919         ret = request_firmware(&tpsram, buf, dev);
920         if (ret < 0) {
921                 dev_err(dev, "could not load TP SRAM: unable to load %s\n",
922                         buf);
923                 return ret;
924         }
925
926         ret = t3_check_tpsram(adap, tpsram->data, tpsram->size);
927         if (ret)
928                 goto release_tpsram;
929
930         ret = t3_set_proto_sram(adap, tpsram->data);
931         if (ret == 0)
932                 dev_info(dev,
933                          "successful update of protocol engine "
934                          "to %d.%d.%d\n",
935                          TP_VERSION_MAJOR, TP_VERSION_MINOR, TP_VERSION_MICRO);
936         else
937                 dev_err(dev, "failed to update of protocol engine %d.%d.%d\n",
938                         TP_VERSION_MAJOR, TP_VERSION_MINOR, TP_VERSION_MICRO);
939         if (ret)
940                 dev_err(dev, "loading protocol SRAM failed\n");
941
942 release_tpsram:
943         release_firmware(tpsram);
944
945         return ret;
946 }
947
948 /**
949  *      cxgb_up - enable the adapter
950  *      @adapter: adapter being enabled
951  *
952  *      Called when the first port is enabled, this function performs the
953  *      actions necessary to make an adapter operational, such as completing
954  *      the initialization of HW modules, and enabling interrupts.
955  *
956  *      Must be called with the rtnl lock held.
957  */
958 static int cxgb_up(struct adapter *adap)
959 {
960         int err;
961
962         if (!(adap->flags & FULL_INIT_DONE)) {
963                 err = t3_check_fw_version(adap);
964                 if (err == -EINVAL) {
965                         err = upgrade_fw(adap);
966                         CH_WARN(adap, "FW upgrade to %d.%d.%d %s\n",
967                                 FW_VERSION_MAJOR, FW_VERSION_MINOR,
968                                 FW_VERSION_MICRO, err ? "failed" : "succeeded");
969                 }
970
971                 err = t3_check_tpsram_version(adap);
972                 if (err == -EINVAL) {
973                         err = update_tpsram(adap);
974                         CH_WARN(adap, "TP upgrade to %d.%d.%d %s\n",
975                                 TP_VERSION_MAJOR, TP_VERSION_MINOR,
976                                 TP_VERSION_MICRO, err ? "failed" : "succeeded");
977                 }
978
979                 /*
980                  * Clear interrupts now to catch errors if t3_init_hw fails.
981                  * We clear them again later as initialization may trigger
982                  * conditions that can interrupt.
983                  */
984                 t3_intr_clear(adap);
985
986                 err = t3_init_hw(adap, 0);
987                 if (err)
988                         goto out;
989
990                 t3_set_reg_field(adap, A_TP_PARA_REG5, 0, F_RXDDPOFFINIT);
991                 t3_write_reg(adap, A_ULPRX_TDDP_PSZ, V_HPZ0(PAGE_SHIFT - 12));
992
993                 err = setup_sge_qsets(adap);
994                 if (err)
995                         goto out;
996
997                 setup_rss(adap);
998                 if (!(adap->flags & NAPI_INIT))
999                         init_napi(adap);
1000                 adap->flags |= FULL_INIT_DONE;
1001         }
1002
1003         t3_intr_clear(adap);
1004
1005         if (adap->flags & USING_MSIX) {
1006                 name_msix_vecs(adap);
1007                 err = request_irq(adap->msix_info[0].vec,
1008                                   t3_async_intr_handler, 0,
1009                                   adap->msix_info[0].desc, adap);
1010                 if (err)
1011                         goto irq_err;
1012
1013                 err = request_msix_data_irqs(adap);
1014                 if (err) {
1015                         free_irq(adap->msix_info[0].vec, adap);
1016                         goto irq_err;
1017                 }
1018         } else if ((err = request_irq(adap->pdev->irq,
1019                                       t3_intr_handler(adap,
1020                                                       adap->sge.qs[0].rspq.
1021                                                       polling),
1022                                       (adap->flags & USING_MSI) ?
1023                                        0 : IRQF_SHARED,
1024                                       adap->name, adap)))
1025                 goto irq_err;
1026
1027         enable_all_napi(adap);
1028         t3_sge_start(adap);
1029         t3_intr_enable(adap);
1030
1031         if (adap->params.rev >= T3_REV_C && !(adap->flags & TP_PARITY_INIT) &&
1032             is_offload(adap) && init_tp_parity(adap) == 0)
1033                 adap->flags |= TP_PARITY_INIT;
1034
1035         if (adap->flags & TP_PARITY_INIT) {
1036                 t3_write_reg(adap, A_TP_INT_CAUSE,
1037                              F_CMCACHEPERR | F_ARPLUTPERR);
1038                 t3_write_reg(adap, A_TP_INT_ENABLE, 0x7fbfffff);
1039         }
1040
1041         if (!(adap->flags & QUEUES_BOUND)) {
1042                 err = bind_qsets(adap);
1043                 if (err) {
1044                         CH_ERR(adap, "failed to bind qsets, err %d\n", err);
1045                         t3_intr_disable(adap);
1046                         free_irq_resources(adap);
1047                         goto out;
1048                 }
1049                 adap->flags |= QUEUES_BOUND;
1050         }
1051
1052 out:
1053         return err;
1054 irq_err:
1055         CH_ERR(adap, "request_irq failed, err %d\n", err);
1056         goto out;
1057 }
1058
1059 /*
1060  * Release resources when all the ports and offloading have been stopped.
1061  */
1062 static void cxgb_down(struct adapter *adapter)
1063 {
1064         t3_sge_stop(adapter);
1065         spin_lock_irq(&adapter->work_lock);     /* sync with PHY intr task */
1066         t3_intr_disable(adapter);
1067         spin_unlock_irq(&adapter->work_lock);
1068
1069         free_irq_resources(adapter);
1070         flush_workqueue(cxgb3_wq);      /* wait for external IRQ handler */
1071         quiesce_rx(adapter);
1072 }
1073
1074 static void schedule_chk_task(struct adapter *adap)
1075 {
1076         unsigned int timeo;
1077
1078         timeo = adap->params.linkpoll_period ?
1079             (HZ * adap->params.linkpoll_period) / 10 :
1080             adap->params.stats_update_period * HZ;
1081         if (timeo)
1082                 queue_delayed_work(cxgb3_wq, &adap->adap_check_task, timeo);
1083 }
1084
1085 static int offload_open(struct net_device *dev)
1086 {
1087         struct port_info *pi = netdev_priv(dev);
1088         struct adapter *adapter = pi->adapter;
1089         struct t3cdev *tdev = dev2t3cdev(dev);
1090         int adap_up = adapter->open_device_map & PORT_MASK;
1091         int err;
1092
1093         if (test_and_set_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map))
1094                 return 0;
1095
1096         if (!adap_up && (err = cxgb_up(adapter)) < 0)
1097                 goto out;
1098
1099         t3_tp_set_offload_mode(adapter, 1);
1100         tdev->lldev = adapter->port[0];
1101         err = cxgb3_offload_activate(adapter);
1102         if (err)
1103                 goto out;
1104
1105         init_port_mtus(adapter);
1106         t3_load_mtus(adapter, adapter->params.mtus, adapter->params.a_wnd,
1107                      adapter->params.b_wnd,
1108                      adapter->params.rev == 0 ?
1109                      adapter->port[0]->mtu : 0xffff);
1110         init_smt(adapter);
1111
1112         if (sysfs_create_group(&tdev->lldev->dev.kobj, &offload_attr_group))
1113                 dev_dbg(&dev->dev, "cannot create sysfs group\n");
1114
1115         /* Call back all registered clients */
1116         cxgb3_add_clients(tdev);
1117
1118 out:
1119         /* restore them in case the offload module has changed them */
1120         if (err) {
1121                 t3_tp_set_offload_mode(adapter, 0);
1122                 clear_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map);
1123                 cxgb3_set_dummy_ops(tdev);
1124         }
1125         return err;
1126 }
1127
1128 static int offload_close(struct t3cdev *tdev)
1129 {
1130         struct adapter *adapter = tdev2adap(tdev);
1131
1132         if (!test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map))
1133                 return 0;
1134
1135         /* Call back all registered clients */
1136         cxgb3_remove_clients(tdev);
1137
1138         sysfs_remove_group(&tdev->lldev->dev.kobj, &offload_attr_group);
1139
1140         tdev->lldev = NULL;
1141         cxgb3_set_dummy_ops(tdev);
1142         t3_tp_set_offload_mode(adapter, 0);
1143         clear_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map);
1144
1145         if (!adapter->open_device_map)
1146                 cxgb_down(adapter);
1147
1148         cxgb3_offload_deactivate(adapter);
1149         return 0;
1150 }
1151
1152 static int cxgb_open(struct net_device *dev)
1153 {
1154         struct port_info *pi = netdev_priv(dev);
1155         struct adapter *adapter = pi->adapter;
1156         int other_ports = adapter->open_device_map & PORT_MASK;
1157         int err;
1158
1159         if (!adapter->open_device_map && (err = cxgb_up(adapter)) < 0)
1160                 return err;
1161
1162         set_bit(pi->port_id, &adapter->open_device_map);
1163         if (is_offload(adapter) && !ofld_disable) {
1164                 err = offload_open(dev);
1165                 if (err)
1166                         printk(KERN_WARNING
1167                                "Could not initialize offload capabilities\n");
1168         }
1169
1170         dev->real_num_tx_queues = pi->nqsets;
1171         link_start(dev);
1172         t3_port_intr_enable(adapter, pi->port_id);
1173         netif_tx_start_all_queues(dev);
1174         if (!other_ports)
1175                 schedule_chk_task(adapter);
1176
1177         return 0;
1178 }
1179
1180 static int cxgb_close(struct net_device *dev)
1181 {
1182         struct port_info *pi = netdev_priv(dev);
1183         struct adapter *adapter = pi->adapter;
1184
1185         t3_port_intr_disable(adapter, pi->port_id);
1186         netif_tx_stop_all_queues(dev);
1187         pi->phy.ops->power_down(&pi->phy, 1);
1188         netif_carrier_off(dev);
1189         t3_mac_disable(&pi->mac, MAC_DIRECTION_TX | MAC_DIRECTION_RX);
1190
1191         spin_lock_irq(&adapter->work_lock);     /* sync with update task */
1192         clear_bit(pi->port_id, &adapter->open_device_map);
1193         spin_unlock_irq(&adapter->work_lock);
1194
1195         if (!(adapter->open_device_map & PORT_MASK))
1196                 cancel_rearming_delayed_workqueue(cxgb3_wq,
1197                                                   &adapter->adap_check_task);
1198
1199         if (!adapter->open_device_map)
1200                 cxgb_down(adapter);
1201
1202         return 0;
1203 }
1204
1205 static struct net_device_stats *cxgb_get_stats(struct net_device *dev)
1206 {
1207         struct port_info *pi = netdev_priv(dev);
1208         struct adapter *adapter = pi->adapter;
1209         struct net_device_stats *ns = &pi->netstats;
1210         const struct mac_stats *pstats;
1211
1212         spin_lock(&adapter->stats_lock);
1213         pstats = t3_mac_update_stats(&pi->mac);
1214         spin_unlock(&adapter->stats_lock);
1215
1216         ns->tx_bytes = pstats->tx_octets;
1217         ns->tx_packets = pstats->tx_frames;
1218         ns->rx_bytes = pstats->rx_octets;
1219         ns->rx_packets = pstats->rx_frames;
1220         ns->multicast = pstats->rx_mcast_frames;
1221
1222         ns->tx_errors = pstats->tx_underrun;
1223         ns->rx_errors = pstats->rx_symbol_errs + pstats->rx_fcs_errs +
1224             pstats->rx_too_long + pstats->rx_jabber + pstats->rx_short +
1225             pstats->rx_fifo_ovfl;
1226
1227         /* detailed rx_errors */
1228         ns->rx_length_errors = pstats->rx_jabber + pstats->rx_too_long;
1229         ns->rx_over_errors = 0;
1230         ns->rx_crc_errors = pstats->rx_fcs_errs;
1231         ns->rx_frame_errors = pstats->rx_symbol_errs;
1232         ns->rx_fifo_errors = pstats->rx_fifo_ovfl;
1233         ns->rx_missed_errors = pstats->rx_cong_drops;
1234
1235         /* detailed tx_errors */
1236         ns->tx_aborted_errors = 0;
1237         ns->tx_carrier_errors = 0;
1238         ns->tx_fifo_errors = pstats->tx_underrun;
1239         ns->tx_heartbeat_errors = 0;
1240         ns->tx_window_errors = 0;
1241         return ns;
1242 }
1243
1244 static u32 get_msglevel(struct net_device *dev)
1245 {
1246         struct port_info *pi = netdev_priv(dev);
1247         struct adapter *adapter = pi->adapter;
1248
1249         return adapter->msg_enable;
1250 }
1251
1252 static void set_msglevel(struct net_device *dev, u32 val)
1253 {
1254         struct port_info *pi = netdev_priv(dev);
1255         struct adapter *adapter = pi->adapter;
1256
1257         adapter->msg_enable = val;
1258 }
1259
1260 static char stats_strings[][ETH_GSTRING_LEN] = {
1261         "TxOctetsOK         ",
1262         "TxFramesOK         ",
1263         "TxMulticastFramesOK",
1264         "TxBroadcastFramesOK",
1265         "TxPauseFrames      ",
1266         "TxUnderrun         ",
1267         "TxExtUnderrun      ",
1268
1269         "TxFrames64         ",
1270         "TxFrames65To127    ",
1271         "TxFrames128To255   ",
1272         "TxFrames256To511   ",
1273         "TxFrames512To1023  ",
1274         "TxFrames1024To1518 ",
1275         "TxFrames1519ToMax  ",
1276
1277         "RxOctetsOK         ",
1278         "RxFramesOK         ",
1279         "RxMulticastFramesOK",
1280         "RxBroadcastFramesOK",
1281         "RxPauseFrames      ",
1282         "RxFCSErrors        ",
1283         "RxSymbolErrors     ",
1284         "RxShortErrors      ",
1285         "RxJabberErrors     ",
1286         "RxLengthErrors     ",
1287         "RxFIFOoverflow     ",
1288
1289         "RxFrames64         ",
1290         "RxFrames65To127    ",
1291         "RxFrames128To255   ",
1292         "RxFrames256To511   ",
1293         "RxFrames512To1023  ",
1294         "RxFrames1024To1518 ",
1295         "RxFrames1519ToMax  ",
1296
1297         "PhyFIFOErrors      ",
1298         "TSO                ",
1299         "VLANextractions    ",
1300         "VLANinsertions     ",
1301         "TxCsumOffload      ",
1302         "RxCsumGood         ",
1303         "LroAggregated      ",
1304         "LroFlushed         ",
1305         "LroNoDesc          ",
1306         "RxDrops            ",
1307
1308         "CheckTXEnToggled   ",
1309         "CheckResets        ",
1310
1311 };
1312
1313 static int get_sset_count(struct net_device *dev, int sset)
1314 {
1315         switch (sset) {
1316         case ETH_SS_STATS:
1317                 return ARRAY_SIZE(stats_strings);
1318         default:
1319                 return -EOPNOTSUPP;
1320         }
1321 }
1322
1323 #define T3_REGMAP_SIZE (3 * 1024)
1324
1325 static int get_regs_len(struct net_device *dev)
1326 {
1327         return T3_REGMAP_SIZE;
1328 }
1329
1330 static int get_eeprom_len(struct net_device *dev)
1331 {
1332         return EEPROMSIZE;
1333 }
1334
1335 static void get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1336 {
1337         struct port_info *pi = netdev_priv(dev);
1338         struct adapter *adapter = pi->adapter;
1339         u32 fw_vers = 0;
1340         u32 tp_vers = 0;
1341
1342         spin_lock(&adapter->stats_lock);
1343         t3_get_fw_version(adapter, &fw_vers);
1344         t3_get_tp_version(adapter, &tp_vers);
1345         spin_unlock(&adapter->stats_lock);
1346
1347         strcpy(info->driver, DRV_NAME);
1348         strcpy(info->version, DRV_VERSION);
1349         strcpy(info->bus_info, pci_name(adapter->pdev));
1350         if (!fw_vers)
1351                 strcpy(info->fw_version, "N/A");
1352         else {
1353                 snprintf(info->fw_version, sizeof(info->fw_version),
1354                          "%s %u.%u.%u TP %u.%u.%u",
1355                          G_FW_VERSION_TYPE(fw_vers) ? "T" : "N",
1356                          G_FW_VERSION_MAJOR(fw_vers),
1357                          G_FW_VERSION_MINOR(fw_vers),
1358                          G_FW_VERSION_MICRO(fw_vers),
1359                          G_TP_VERSION_MAJOR(tp_vers),
1360                          G_TP_VERSION_MINOR(tp_vers),
1361                          G_TP_VERSION_MICRO(tp_vers));
1362         }
1363 }
1364
1365 static void get_strings(struct net_device *dev, u32 stringset, u8 * data)
1366 {
1367         if (stringset == ETH_SS_STATS)
1368                 memcpy(data, stats_strings, sizeof(stats_strings));
1369 }
1370
1371 static unsigned long collect_sge_port_stats(struct adapter *adapter,
1372                                             struct port_info *p, int idx)
1373 {
1374         int i;
1375         unsigned long tot = 0;
1376
1377         for (i = p->first_qset; i < p->first_qset + p->nqsets; ++i)
1378                 tot += adapter->sge.qs[i].port_stats[idx];
1379         return tot;
1380 }
1381
1382 static void get_stats(struct net_device *dev, struct ethtool_stats *stats,
1383                       u64 *data)
1384 {
1385         struct port_info *pi = netdev_priv(dev);
1386         struct adapter *adapter = pi->adapter;
1387         const struct mac_stats *s;
1388
1389         spin_lock(&adapter->stats_lock);
1390         s = t3_mac_update_stats(&pi->mac);
1391         spin_unlock(&adapter->stats_lock);
1392
1393         *data++ = s->tx_octets;
1394         *data++ = s->tx_frames;
1395         *data++ = s->tx_mcast_frames;
1396         *data++ = s->tx_bcast_frames;
1397         *data++ = s->tx_pause;
1398         *data++ = s->tx_underrun;
1399         *data++ = s->tx_fifo_urun;
1400
1401         *data++ = s->tx_frames_64;
1402         *data++ = s->tx_frames_65_127;
1403         *data++ = s->tx_frames_128_255;
1404         *data++ = s->tx_frames_256_511;
1405         *data++ = s->tx_frames_512_1023;
1406         *data++ = s->tx_frames_1024_1518;
1407         *data++ = s->tx_frames_1519_max;
1408
1409         *data++ = s->rx_octets;
1410         *data++ = s->rx_frames;
1411         *data++ = s->rx_mcast_frames;
1412         *data++ = s->rx_bcast_frames;
1413         *data++ = s->rx_pause;
1414         *data++ = s->rx_fcs_errs;
1415         *data++ = s->rx_symbol_errs;
1416         *data++ = s->rx_short;
1417         *data++ = s->rx_jabber;
1418         *data++ = s->rx_too_long;
1419         *data++ = s->rx_fifo_ovfl;
1420
1421         *data++ = s->rx_frames_64;
1422         *data++ = s->rx_frames_65_127;
1423         *data++ = s->rx_frames_128_255;
1424         *data++ = s->rx_frames_256_511;
1425         *data++ = s->rx_frames_512_1023;
1426         *data++ = s->rx_frames_1024_1518;
1427         *data++ = s->rx_frames_1519_max;
1428
1429         *data++ = pi->phy.fifo_errors;
1430
1431         *data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_TSO);
1432         *data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_VLANEX);
1433         *data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_VLANINS);
1434         *data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_TX_CSUM);
1435         *data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_RX_CSUM_GOOD);
1436         *data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_LRO_AGGR);
1437         *data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_LRO_FLUSHED);
1438         *data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_LRO_NO_DESC);
1439         *data++ = s->rx_cong_drops;
1440
1441         *data++ = s->num_toggled;
1442         *data++ = s->num_resets;
1443 }
1444
1445 static inline void reg_block_dump(struct adapter *ap, void *buf,
1446                                   unsigned int start, unsigned int end)
1447 {
1448         u32 *p = buf + start;
1449
1450         for (; start <= end; start += sizeof(u32))
1451                 *p++ = t3_read_reg(ap, start);
1452 }
1453
1454 static void get_regs(struct net_device *dev, struct ethtool_regs *regs,
1455                      void *buf)
1456 {
1457         struct port_info *pi = netdev_priv(dev);
1458         struct adapter *ap = pi->adapter;
1459
1460         /*
1461          * Version scheme:
1462          * bits 0..9: chip version
1463          * bits 10..15: chip revision
1464          * bit 31: set for PCIe cards
1465          */
1466         regs->version = 3 | (ap->params.rev << 10) | (is_pcie(ap) << 31);
1467
1468         /*
1469          * We skip the MAC statistics registers because they are clear-on-read.
1470          * Also reading multi-register stats would need to synchronize with the
1471          * periodic mac stats accumulation.  Hard to justify the complexity.
1472          */
1473         memset(buf, 0, T3_REGMAP_SIZE);
1474         reg_block_dump(ap, buf, 0, A_SG_RSPQ_CREDIT_RETURN);
1475         reg_block_dump(ap, buf, A_SG_HI_DRB_HI_THRSH, A_ULPRX_PBL_ULIMIT);
1476         reg_block_dump(ap, buf, A_ULPTX_CONFIG, A_MPS_INT_CAUSE);
1477         reg_block_dump(ap, buf, A_CPL_SWITCH_CNTRL, A_CPL_MAP_TBL_DATA);
1478         reg_block_dump(ap, buf, A_SMB_GLOBAL_TIME_CFG, A_XGM_SERDES_STAT3);
1479         reg_block_dump(ap, buf, A_XGM_SERDES_STATUS0,
1480                        XGM_REG(A_XGM_SERDES_STAT3, 1));
1481         reg_block_dump(ap, buf, XGM_REG(A_XGM_SERDES_STATUS0, 1),
1482                        XGM_REG(A_XGM_RX_SPI4_SOP_EOP_CNT, 1));
1483 }
1484
1485 static int restart_autoneg(struct net_device *dev)
1486 {
1487         struct port_info *p = netdev_priv(dev);
1488
1489         if (!netif_running(dev))
1490                 return -EAGAIN;
1491         if (p->link_config.autoneg != AUTONEG_ENABLE)
1492                 return -EINVAL;
1493         p->phy.ops->autoneg_restart(&p->phy);
1494         return 0;
1495 }
1496
1497 static int cxgb3_phys_id(struct net_device *dev, u32 data)
1498 {
1499         struct port_info *pi = netdev_priv(dev);
1500         struct adapter *adapter = pi->adapter;
1501         int i;
1502
1503         if (data == 0)
1504                 data = 2;
1505
1506         for (i = 0; i < data * 2; i++) {
1507                 t3_set_reg_field(adapter, A_T3DBG_GPIO_EN, F_GPIO0_OUT_VAL,
1508                                  (i & 1) ? F_GPIO0_OUT_VAL : 0);
1509                 if (msleep_interruptible(500))
1510                         break;
1511         }
1512         t3_set_reg_field(adapter, A_T3DBG_GPIO_EN, F_GPIO0_OUT_VAL,
1513                          F_GPIO0_OUT_VAL);
1514         return 0;
1515 }
1516
1517 static int get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1518 {
1519         struct port_info *p = netdev_priv(dev);
1520
1521         cmd->supported = p->link_config.supported;
1522         cmd->advertising = p->link_config.advertising;
1523
1524         if (netif_carrier_ok(dev)) {
1525                 cmd->speed = p->link_config.speed;
1526                 cmd->duplex = p->link_config.duplex;
1527         } else {
1528                 cmd->speed = -1;
1529                 cmd->duplex = -1;
1530         }
1531
1532         cmd->port = (cmd->supported & SUPPORTED_TP) ? PORT_TP : PORT_FIBRE;
1533         cmd->phy_address = p->phy.addr;
1534         cmd->transceiver = XCVR_EXTERNAL;
1535         cmd->autoneg = p->link_config.autoneg;
1536         cmd->maxtxpkt = 0;
1537         cmd->maxrxpkt = 0;
1538         return 0;
1539 }
1540
1541 static int speed_duplex_to_caps(int speed, int duplex)
1542 {
1543         int cap = 0;
1544
1545         switch (speed) {
1546         case SPEED_10:
1547                 if (duplex == DUPLEX_FULL)
1548                         cap = SUPPORTED_10baseT_Full;
1549                 else
1550                         cap = SUPPORTED_10baseT_Half;
1551                 break;
1552         case SPEED_100:
1553                 if (duplex == DUPLEX_FULL)
1554                         cap = SUPPORTED_100baseT_Full;
1555                 else
1556                         cap = SUPPORTED_100baseT_Half;
1557                 break;
1558         case SPEED_1000:
1559                 if (duplex == DUPLEX_FULL)
1560                         cap = SUPPORTED_1000baseT_Full;
1561                 else
1562                         cap = SUPPORTED_1000baseT_Half;
1563                 break;
1564         case SPEED_10000:
1565                 if (duplex == DUPLEX_FULL)
1566                         cap = SUPPORTED_10000baseT_Full;
1567         }
1568         return cap;
1569 }
1570
1571 #define ADVERTISED_MASK (ADVERTISED_10baseT_Half | ADVERTISED_10baseT_Full | \
1572                       ADVERTISED_100baseT_Half | ADVERTISED_100baseT_Full | \
1573                       ADVERTISED_1000baseT_Half | ADVERTISED_1000baseT_Full | \
1574                       ADVERTISED_10000baseT_Full)
1575
1576 static int set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1577 {
1578         int cap;
1579         struct port_info *p = netdev_priv(dev);
1580         struct link_config *lc = &p->link_config;
1581
1582         if (!(lc->supported & SUPPORTED_Autoneg)) {
1583                 /*
1584                  * PHY offers a single speed/duplex.  See if that's what's
1585                  * being requested.
1586                  */
1587                 if (cmd->autoneg == AUTONEG_DISABLE) {
1588                         cap = speed_duplex_to_caps(cmd->speed, cmd->duplex);
1589                         if (lc->supported & cap)
1590                                 return 0;
1591                 }
1592                 return -EINVAL;
1593         }
1594
1595         if (cmd->autoneg == AUTONEG_DISABLE) {
1596                 int cap = speed_duplex_to_caps(cmd->speed, cmd->duplex);
1597
1598                 if (!(lc->supported & cap) || cmd->speed == SPEED_1000)
1599                         return -EINVAL;
1600                 lc->requested_speed = cmd->speed;
1601                 lc->requested_duplex = cmd->duplex;
1602                 lc->advertising = 0;
1603         } else {
1604                 cmd->advertising &= ADVERTISED_MASK;
1605                 cmd->advertising &= lc->supported;
1606                 if (!cmd->advertising)
1607                         return -EINVAL;
1608                 lc->requested_speed = SPEED_INVALID;
1609                 lc->requested_duplex = DUPLEX_INVALID;
1610                 lc->advertising = cmd->advertising | ADVERTISED_Autoneg;
1611         }
1612         lc->autoneg = cmd->autoneg;
1613         if (netif_running(dev))
1614                 t3_link_start(&p->phy, &p->mac, lc);
1615         return 0;
1616 }
1617
1618 static void get_pauseparam(struct net_device *dev,
1619                            struct ethtool_pauseparam *epause)
1620 {
1621         struct port_info *p = netdev_priv(dev);
1622
1623         epause->autoneg = (p->link_config.requested_fc & PAUSE_AUTONEG) != 0;
1624         epause->rx_pause = (p->link_config.fc & PAUSE_RX) != 0;
1625         epause->tx_pause = (p->link_config.fc & PAUSE_TX) != 0;
1626 }
1627
1628 static int set_pauseparam(struct net_device *dev,
1629                           struct ethtool_pauseparam *epause)
1630 {
1631         struct port_info *p = netdev_priv(dev);
1632         struct link_config *lc = &p->link_config;
1633
1634         if (epause->autoneg == AUTONEG_DISABLE)
1635                 lc->requested_fc = 0;
1636         else if (lc->supported & SUPPORTED_Autoneg)
1637                 lc->requested_fc = PAUSE_AUTONEG;
1638         else
1639                 return -EINVAL;
1640
1641         if (epause->rx_pause)
1642                 lc->requested_fc |= PAUSE_RX;
1643         if (epause->tx_pause)
1644                 lc->requested_fc |= PAUSE_TX;
1645         if (lc->autoneg == AUTONEG_ENABLE) {
1646                 if (netif_running(dev))
1647                         t3_link_start(&p->phy, &p->mac, lc);
1648         } else {
1649                 lc->fc = lc->requested_fc & (PAUSE_RX | PAUSE_TX);
1650                 if (netif_running(dev))
1651                         t3_mac_set_speed_duplex_fc(&p->mac, -1, -1, lc->fc);
1652         }
1653         return 0;
1654 }
1655
1656 static u32 get_rx_csum(struct net_device *dev)
1657 {
1658         struct port_info *p = netdev_priv(dev);
1659
1660         return p->rx_offload & T3_RX_CSUM;
1661 }
1662
1663 static int set_rx_csum(struct net_device *dev, u32 data)
1664 {
1665         struct port_info *p = netdev_priv(dev);
1666
1667         if (data) {
1668                 p->rx_offload |= T3_RX_CSUM;
1669         } else {
1670                 int i;
1671
1672                 p->rx_offload &= ~(T3_RX_CSUM | T3_LRO);
1673                 for (i = p->first_qset; i < p->first_qset + p->nqsets; i++)
1674                         set_qset_lro(dev, i, 0);
1675         }
1676         return 0;
1677 }
1678
1679 static void get_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
1680 {
1681         struct port_info *pi = netdev_priv(dev);
1682         struct adapter *adapter = pi->adapter;
1683         const struct qset_params *q = &adapter->params.sge.qset[pi->first_qset];
1684
1685         e->rx_max_pending = MAX_RX_BUFFERS;
1686         e->rx_mini_max_pending = 0;
1687         e->rx_jumbo_max_pending = MAX_RX_JUMBO_BUFFERS;
1688         e->tx_max_pending = MAX_TXQ_ENTRIES;
1689
1690         e->rx_pending = q->fl_size;
1691         e->rx_mini_pending = q->rspq_size;
1692         e->rx_jumbo_pending = q->jumbo_size;
1693         e->tx_pending = q->txq_size[0];
1694 }
1695
1696 static int set_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
1697 {
1698         struct port_info *pi = netdev_priv(dev);
1699         struct adapter *adapter = pi->adapter;
1700         struct qset_params *q;
1701         int i;
1702
1703         if (e->rx_pending > MAX_RX_BUFFERS ||
1704             e->rx_jumbo_pending > MAX_RX_JUMBO_BUFFERS ||
1705             e->tx_pending > MAX_TXQ_ENTRIES ||
1706             e->rx_mini_pending > MAX_RSPQ_ENTRIES ||
1707             e->rx_mini_pending < MIN_RSPQ_ENTRIES ||
1708             e->rx_pending < MIN_FL_ENTRIES ||
1709             e->rx_jumbo_pending < MIN_FL_ENTRIES ||
1710             e->tx_pending < adapter->params.nports * MIN_TXQ_ENTRIES)
1711                 return -EINVAL;
1712
1713         if (adapter->flags & FULL_INIT_DONE)
1714                 return -EBUSY;
1715
1716         q = &adapter->params.sge.qset[pi->first_qset];
1717         for (i = 0; i < pi->nqsets; ++i, ++q) {
1718                 q->rspq_size = e->rx_mini_pending;
1719                 q->fl_size = e->rx_pending;
1720                 q->jumbo_size = e->rx_jumbo_pending;
1721                 q->txq_size[0] = e->tx_pending;
1722                 q->txq_size[1] = e->tx_pending;
1723                 q->txq_size[2] = e->tx_pending;
1724         }
1725         return 0;
1726 }
1727
1728 static int set_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
1729 {
1730         struct port_info *pi = netdev_priv(dev);
1731         struct adapter *adapter = pi->adapter;
1732         struct qset_params *qsp = &adapter->params.sge.qset[0];
1733         struct sge_qset *qs = &adapter->sge.qs[0];
1734
1735         if (c->rx_coalesce_usecs * 10 > M_NEWTIMER)
1736                 return -EINVAL;
1737
1738         qsp->coalesce_usecs = c->rx_coalesce_usecs;
1739         t3_update_qset_coalesce(qs, qsp);
1740         return 0;
1741 }
1742
1743 static int get_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
1744 {
1745         struct port_info *pi = netdev_priv(dev);
1746         struct adapter *adapter = pi->adapter;
1747         struct qset_params *q = adapter->params.sge.qset;
1748
1749         c->rx_coalesce_usecs = q->coalesce_usecs;
1750         return 0;
1751 }
1752
1753 static int get_eeprom(struct net_device *dev, struct ethtool_eeprom *e,
1754                       u8 * data)
1755 {
1756         struct port_info *pi = netdev_priv(dev);
1757         struct adapter *adapter = pi->adapter;
1758         int i, err = 0;
1759
1760         u8 *buf = kmalloc(EEPROMSIZE, GFP_KERNEL);
1761         if (!buf)
1762                 return -ENOMEM;
1763
1764         e->magic = EEPROM_MAGIC;
1765         for (i = e->offset & ~3; !err && i < e->offset + e->len; i += 4)
1766                 err = t3_seeprom_read(adapter, i, (__le32 *) & buf[i]);
1767
1768         if (!err)
1769                 memcpy(data, buf + e->offset, e->len);
1770         kfree(buf);
1771         return err;
1772 }
1773
1774 static int set_eeprom(struct net_device *dev, struct ethtool_eeprom *eeprom,
1775                       u8 * data)
1776 {
1777         struct port_info *pi = netdev_priv(dev);
1778         struct adapter *adapter = pi->adapter;
1779         u32 aligned_offset, aligned_len;
1780         __le32 *p;
1781         u8 *buf;
1782         int err;
1783
1784         if (eeprom->magic != EEPROM_MAGIC)
1785                 return -EINVAL;
1786
1787         aligned_offset = eeprom->offset & ~3;
1788         aligned_len = (eeprom->len + (eeprom->offset & 3) + 3) & ~3;
1789
1790         if (aligned_offset != eeprom->offset || aligned_len != eeprom->len) {
1791                 buf = kmalloc(aligned_len, GFP_KERNEL);
1792                 if (!buf)
1793                         return -ENOMEM;
1794                 err = t3_seeprom_read(adapter, aligned_offset, (__le32 *) buf);
1795                 if (!err && aligned_len > 4)
1796                         err = t3_seeprom_read(adapter,
1797                                               aligned_offset + aligned_len - 4,
1798                                               (__le32 *) & buf[aligned_len - 4]);
1799                 if (err)
1800                         goto out;
1801                 memcpy(buf + (eeprom->offset & 3), data, eeprom->len);
1802         } else
1803                 buf = data;
1804
1805         err = t3_seeprom_wp(adapter, 0);
1806         if (err)
1807                 goto out;
1808
1809         for (p = (__le32 *) buf; !err && aligned_len; aligned_len -= 4, p++) {
1810                 err = t3_seeprom_write(adapter, aligned_offset, *p);
1811                 aligned_offset += 4;
1812         }
1813
1814         if (!err)
1815                 err = t3_seeprom_wp(adapter, 1);
1816 out:
1817         if (buf != data)
1818                 kfree(buf);
1819         return err;
1820 }
1821
1822 static void get_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
1823 {
1824         wol->supported = 0;
1825         wol->wolopts = 0;
1826         memset(&wol->sopass, 0, sizeof(wol->sopass));
1827 }
1828
1829 static int cxgb3_set_flags(struct net_device *dev, u32 data)
1830 {
1831         struct port_info *pi = netdev_priv(dev);
1832         int i;
1833
1834         if (data & ETH_FLAG_LRO) {
1835                 if (!(pi->rx_offload & T3_RX_CSUM))
1836                         return -EINVAL;
1837
1838                 pi->rx_offload |= T3_LRO;
1839                 for (i = pi->first_qset; i < pi->first_qset + pi->nqsets; i++)
1840                         set_qset_lro(dev, i, 1);
1841
1842         } else {
1843                 pi->rx_offload &= ~T3_LRO;
1844                 for (i = pi->first_qset; i < pi->first_qset + pi->nqsets; i++)
1845                         set_qset_lro(dev, i, 0);
1846         }
1847
1848         return 0;
1849 }
1850
1851 static const struct ethtool_ops cxgb_ethtool_ops = {
1852         .get_settings = get_settings,
1853         .set_settings = set_settings,
1854         .get_drvinfo = get_drvinfo,
1855         .get_msglevel = get_msglevel,
1856         .set_msglevel = set_msglevel,
1857         .get_ringparam = get_sge_param,
1858         .set_ringparam = set_sge_param,
1859         .get_coalesce = get_coalesce,
1860         .set_coalesce = set_coalesce,
1861         .get_eeprom_len = get_eeprom_len,
1862         .get_eeprom = get_eeprom,
1863         .set_eeprom = set_eeprom,
1864         .get_pauseparam = get_pauseparam,
1865         .set_pauseparam = set_pauseparam,
1866         .get_rx_csum = get_rx_csum,
1867         .set_rx_csum = set_rx_csum,
1868         .set_tx_csum = ethtool_op_set_tx_csum,
1869         .set_sg = ethtool_op_set_sg,
1870         .get_link = ethtool_op_get_link,
1871         .get_strings = get_strings,
1872         .phys_id = cxgb3_phys_id,
1873         .nway_reset = restart_autoneg,
1874         .get_sset_count = get_sset_count,
1875         .get_ethtool_stats = get_stats,
1876         .get_regs_len = get_regs_len,
1877         .get_regs = get_regs,
1878         .get_wol = get_wol,
1879         .set_tso = ethtool_op_set_tso,
1880         .get_flags = ethtool_op_get_flags,
1881         .set_flags = cxgb3_set_flags,
1882 };
1883
1884 static int in_range(int val, int lo, int hi)
1885 {
1886         return val < 0 || (val <= hi && val >= lo);
1887 }
1888
1889 static int cxgb_extension_ioctl(struct net_device *dev, void __user *useraddr)
1890 {
1891         struct port_info *pi = netdev_priv(dev);
1892         struct adapter *adapter = pi->adapter;
1893         u32 cmd;
1894         int ret;
1895
1896         if (copy_from_user(&cmd, useraddr, sizeof(cmd)))
1897                 return -EFAULT;
1898
1899         switch (cmd) {
1900         case CHELSIO_SET_QSET_PARAMS:{
1901                 int i;
1902                 struct qset_params *q;
1903                 struct ch_qset_params t;
1904                 int q1 = pi->first_qset;
1905                 int nqsets = pi->nqsets;
1906
1907                 if (!capable(CAP_NET_ADMIN))
1908                         return -EPERM;
1909                 if (copy_from_user(&t, useraddr, sizeof(t)))
1910                         return -EFAULT;
1911                 if (t.qset_idx >= SGE_QSETS)
1912                         return -EINVAL;
1913                 if (!in_range(t.intr_lat, 0, M_NEWTIMER) ||
1914                         !in_range(t.cong_thres, 0, 255) ||
1915                         !in_range(t.txq_size[0], MIN_TXQ_ENTRIES,
1916                                 MAX_TXQ_ENTRIES) ||
1917                         !in_range(t.txq_size[1], MIN_TXQ_ENTRIES,
1918                                 MAX_TXQ_ENTRIES) ||
1919                         !in_range(t.txq_size[2], MIN_CTRL_TXQ_ENTRIES,
1920                                 MAX_CTRL_TXQ_ENTRIES) ||
1921                         !in_range(t.fl_size[0], MIN_FL_ENTRIES,
1922                                 MAX_RX_BUFFERS)
1923                         || !in_range(t.fl_size[1], MIN_FL_ENTRIES,
1924                                         MAX_RX_JUMBO_BUFFERS)
1925                         || !in_range(t.rspq_size, MIN_RSPQ_ENTRIES,
1926                                         MAX_RSPQ_ENTRIES))
1927                         return -EINVAL;
1928
1929                 if ((adapter->flags & FULL_INIT_DONE) && t.lro > 0)
1930                         for_each_port(adapter, i) {
1931                                 pi = adap2pinfo(adapter, i);
1932                                 if (t.qset_idx >= pi->first_qset &&
1933                                     t.qset_idx < pi->first_qset + pi->nqsets &&
1934                                     !(pi->rx_offload & T3_RX_CSUM))
1935                                         return -EINVAL;
1936                         }
1937
1938                 if ((adapter->flags & FULL_INIT_DONE) &&
1939                         (t.rspq_size >= 0 || t.fl_size[0] >= 0 ||
1940                         t.fl_size[1] >= 0 || t.txq_size[0] >= 0 ||
1941                         t.txq_size[1] >= 0 || t.txq_size[2] >= 0 ||
1942                         t.polling >= 0 || t.cong_thres >= 0))
1943                         return -EBUSY;
1944
1945                 /* Allow setting of any available qset when offload enabled */
1946                 if (test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map)) {
1947                         q1 = 0;
1948                         for_each_port(adapter, i) {
1949                                 pi = adap2pinfo(adapter, i);
1950                                 nqsets += pi->first_qset + pi->nqsets;
1951                         }
1952                 }
1953
1954                 if (t.qset_idx < q1)
1955                         return -EINVAL;
1956                 if (t.qset_idx > q1 + nqsets - 1)
1957                         return -EINVAL;
1958
1959                 q = &adapter->params.sge.qset[t.qset_idx];
1960
1961                 if (t.rspq_size >= 0)
1962                         q->rspq_size = t.rspq_size;
1963                 if (t.fl_size[0] >= 0)
1964                         q->fl_size = t.fl_size[0];
1965                 if (t.fl_size[1] >= 0)
1966                         q->jumbo_size = t.fl_size[1];
1967                 if (t.txq_size[0] >= 0)
1968                         q->txq_size[0] = t.txq_size[0];
1969                 if (t.txq_size[1] >= 0)
1970                         q->txq_size[1] = t.txq_size[1];
1971                 if (t.txq_size[2] >= 0)
1972                         q->txq_size[2] = t.txq_size[2];
1973                 if (t.cong_thres >= 0)
1974                         q->cong_thres = t.cong_thres;
1975                 if (t.intr_lat >= 0) {
1976                         struct sge_qset *qs =
1977                                 &adapter->sge.qs[t.qset_idx];
1978
1979                         q->coalesce_usecs = t.intr_lat;
1980                         t3_update_qset_coalesce(qs, q);
1981                 }
1982                 if (t.polling >= 0) {
1983                         if (adapter->flags & USING_MSIX)
1984                                 q->polling = t.polling;
1985                         else {
1986                                 /* No polling with INTx for T3A */
1987                                 if (adapter->params.rev == 0 &&
1988                                         !(adapter->flags & USING_MSI))
1989                                         t.polling = 0;
1990
1991                                 for (i = 0; i < SGE_QSETS; i++) {
1992                                         q = &adapter->params.sge.
1993                                                 qset[i];
1994                                         q->polling = t.polling;
1995                                 }
1996                         }
1997                 }
1998                 if (t.lro >= 0)
1999                         set_qset_lro(dev, t.qset_idx, t.lro);
2000
2001                 break;
2002         }
2003         case CHELSIO_GET_QSET_PARAMS:{
2004                 struct qset_params *q;
2005                 struct ch_qset_params t;
2006                 int q1 = pi->first_qset;
2007                 int nqsets = pi->nqsets;
2008                 int i;
2009
2010                 if (copy_from_user(&t, useraddr, sizeof(t)))
2011                         return -EFAULT;
2012
2013                 /* Display qsets for all ports when offload enabled */
2014                 if (test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map)) {
2015                         q1 = 0;
2016                         for_each_port(adapter, i) {
2017                                 pi = adap2pinfo(adapter, i);
2018                                 nqsets = pi->first_qset + pi->nqsets;
2019                         }
2020                 }
2021
2022                 if (t.qset_idx >= nqsets)
2023                         return -EINVAL;
2024
2025                 q = &adapter->params.sge.qset[q1 + t.qset_idx];
2026                 t.rspq_size = q->rspq_size;
2027                 t.txq_size[0] = q->txq_size[0];
2028                 t.txq_size[1] = q->txq_size[1];
2029                 t.txq_size[2] = q->txq_size[2];
2030                 t.fl_size[0] = q->fl_size;
2031                 t.fl_size[1] = q->jumbo_size;
2032                 t.polling = q->polling;
2033                 t.lro = q->lro;
2034                 t.intr_lat = q->coalesce_usecs;
2035                 t.cong_thres = q->cong_thres;
2036                 t.qnum = q1;
2037
2038                 if (adapter->flags & USING_MSIX)
2039                         t.vector = adapter->msix_info[q1 + t.qset_idx + 1].vec;
2040                 else
2041                         t.vector = adapter->pdev->irq;
2042
2043                 if (copy_to_user(useraddr, &t, sizeof(t)))
2044                         return -EFAULT;
2045                 break;
2046         }
2047         case CHELSIO_SET_QSET_NUM:{
2048                 struct ch_reg edata;
2049                 unsigned int i, first_qset = 0, other_qsets = 0;
2050
2051                 if (!capable(CAP_NET_ADMIN))
2052                         return -EPERM;
2053                 if (adapter->flags & FULL_INIT_DONE)
2054                         return -EBUSY;
2055                 if (copy_from_user(&edata, useraddr, sizeof(edata)))
2056                         return -EFAULT;
2057                 if (edata.val < 1 ||
2058                         (edata.val > 1 && !(adapter->flags & USING_MSIX)))
2059                         return -EINVAL;
2060
2061                 for_each_port(adapter, i)
2062                         if (adapter->port[i] && adapter->port[i] != dev)
2063                                 other_qsets += adap2pinfo(adapter, i)->nqsets;
2064
2065                 if (edata.val + other_qsets > SGE_QSETS)
2066                         return -EINVAL;
2067
2068                 pi->nqsets = edata.val;
2069
2070                 for_each_port(adapter, i)
2071                         if (adapter->port[i]) {
2072                                 pi = adap2pinfo(adapter, i);
2073                                 pi->first_qset = first_qset;
2074                                 first_qset += pi->nqsets;
2075                         }
2076                 break;
2077         }
2078         case CHELSIO_GET_QSET_NUM:{
2079                 struct ch_reg edata;
2080
2081                 edata.cmd = CHELSIO_GET_QSET_NUM;
2082                 edata.val = pi->nqsets;
2083                 if (copy_to_user(useraddr, &edata, sizeof(edata)))
2084                         return -EFAULT;
2085                 break;
2086         }
2087         case CHELSIO_LOAD_FW:{
2088                 u8 *fw_data;
2089                 struct ch_mem_range t;
2090
2091                 if (!capable(CAP_SYS_RAWIO))
2092                         return -EPERM;
2093                 if (copy_from_user(&t, useraddr, sizeof(t)))
2094                         return -EFAULT;
2095                 /* Check t.len sanity ? */
2096                 fw_data = kmalloc(t.len, GFP_KERNEL);
2097                 if (!fw_data)
2098                         return -ENOMEM;
2099
2100                 if (copy_from_user
2101                         (fw_data, useraddr + sizeof(t), t.len)) {
2102                         kfree(fw_data);
2103                         return -EFAULT;
2104                 }
2105
2106                 ret = t3_load_fw(adapter, fw_data, t.len);
2107                 kfree(fw_data);
2108                 if (ret)
2109                         return ret;
2110                 break;
2111         }
2112         case CHELSIO_SETMTUTAB:{
2113                 struct ch_mtus m;
2114                 int i;
2115
2116                 if (!is_offload(adapter))
2117                         return -EOPNOTSUPP;
2118                 if (!capable(CAP_NET_ADMIN))
2119                         return -EPERM;
2120                 if (offload_running(adapter))
2121                         return -EBUSY;
2122                 if (copy_from_user(&m, useraddr, sizeof(m)))
2123                         return -EFAULT;
2124                 if (m.nmtus != NMTUS)
2125                         return -EINVAL;
2126                 if (m.mtus[0] < 81)     /* accommodate SACK */
2127                         return -EINVAL;
2128
2129                 /* MTUs must be in ascending order */
2130                 for (i = 1; i < NMTUS; ++i)
2131                         if (m.mtus[i] < m.mtus[i - 1])
2132                                 return -EINVAL;
2133
2134                 memcpy(adapter->params.mtus, m.mtus,
2135                         sizeof(adapter->params.mtus));
2136                 break;
2137         }
2138         case CHELSIO_GET_PM:{
2139                 struct tp_params *p = &adapter->params.tp;
2140                 struct ch_pm m = {.cmd = CHELSIO_GET_PM };
2141
2142                 if (!is_offload(adapter))
2143                         return -EOPNOTSUPP;
2144                 m.tx_pg_sz = p->tx_pg_size;
2145                 m.tx_num_pg = p->tx_num_pgs;
2146                 m.rx_pg_sz = p->rx_pg_size;
2147                 m.rx_num_pg = p->rx_num_pgs;
2148                 m.pm_total = p->pmtx_size + p->chan_rx_size * p->nchan;
2149                 if (copy_to_user(useraddr, &m, sizeof(m)))
2150                         return -EFAULT;
2151                 break;
2152         }
2153         case CHELSIO_SET_PM:{
2154                 struct ch_pm m;
2155                 struct tp_params *p = &adapter->params.tp;
2156
2157                 if (!is_offload(adapter))
2158                         return -EOPNOTSUPP;
2159                 if (!capable(CAP_NET_ADMIN))
2160                         return -EPERM;
2161                 if (adapter->flags & FULL_INIT_DONE)
2162                         return -EBUSY;
2163                 if (copy_from_user(&m, useraddr, sizeof(m)))
2164                         return -EFAULT;
2165                 if (!is_power_of_2(m.rx_pg_sz) ||
2166                         !is_power_of_2(m.tx_pg_sz))
2167                         return -EINVAL; /* not power of 2 */
2168                 if (!(m.rx_pg_sz & 0x14000))
2169                         return -EINVAL; /* not 16KB or 64KB */
2170                 if (!(m.tx_pg_sz & 0x1554000))
2171                         return -EINVAL;
2172                 if (m.tx_num_pg == -1)
2173                         m.tx_num_pg = p->tx_num_pgs;
2174                 if (m.rx_num_pg == -1)
2175                         m.rx_num_pg = p->rx_num_pgs;
2176                 if (m.tx_num_pg % 24 || m.rx_num_pg % 24)
2177                         return -EINVAL;
2178                 if (m.rx_num_pg * m.rx_pg_sz > p->chan_rx_size ||
2179                         m.tx_num_pg * m.tx_pg_sz > p->chan_tx_size)
2180                         return -EINVAL;
2181                 p->rx_pg_size = m.rx_pg_sz;
2182                 p->tx_pg_size = m.tx_pg_sz;
2183                 p->rx_num_pgs = m.rx_num_pg;
2184                 p->tx_num_pgs = m.tx_num_pg;
2185                 break;
2186         }
2187         case CHELSIO_GET_MEM:{
2188                 struct ch_mem_range t;
2189                 struct mc7 *mem;
2190                 u64 buf[32];
2191
2192                 if (!is_offload(adapter))
2193                         return -EOPNOTSUPP;
2194                 if (!(adapter->flags & FULL_INIT_DONE))
2195                         return -EIO;    /* need the memory controllers */
2196                 if (copy_from_user(&t, useraddr, sizeof(t)))
2197                         return -EFAULT;
2198                 if ((t.addr & 7) || (t.len & 7))
2199                         return -EINVAL;
2200                 if (t.mem_id == MEM_CM)
2201                         mem = &adapter->cm;
2202                 else if (t.mem_id == MEM_PMRX)
2203                         mem = &adapter->pmrx;
2204                 else if (t.mem_id == MEM_PMTX)
2205                         mem = &adapter->pmtx;
2206                 else
2207                         return -EINVAL;
2208
2209                 /*
2210                  * Version scheme:
2211                  * bits 0..9: chip version
2212                  * bits 10..15: chip revision
2213                  */
2214                 t.version = 3 | (adapter->params.rev << 10);
2215                 if (copy_to_user(useraddr, &t, sizeof(t)))
2216                         return -EFAULT;
2217
2218                 /*
2219                  * Read 256 bytes at a time as len can be large and we don't
2220                  * want to use huge intermediate buffers.
2221                  */
2222                 useraddr += sizeof(t);  /* advance to start of buffer */
2223                 while (t.len) {
2224                         unsigned int chunk =
2225                                 min_t(unsigned int, t.len, sizeof(buf));
2226
2227                         ret =
2228                                 t3_mc7_bd_read(mem, t.addr / 8, chunk / 8,
2229                                                 buf);
2230                         if (ret)
2231                                 return ret;
2232                         if (copy_to_user(useraddr, buf, chunk))
2233                                 return -EFAULT;
2234                         useraddr += chunk;
2235                         t.addr += chunk;
2236                         t.len -= chunk;
2237                 }
2238                 break;
2239         }
2240         case CHELSIO_SET_TRACE_FILTER:{
2241                 struct ch_trace t;
2242                 const struct trace_params *tp;
2243
2244                 if (!capable(CAP_NET_ADMIN))
2245                         return -EPERM;
2246                 if (!offload_running(adapter))
2247                         return -EAGAIN;
2248                 if (copy_from_user(&t, useraddr, sizeof(t)))
2249                         return -EFAULT;
2250
2251                 tp = (const struct trace_params *)&t.sip;
2252                 if (t.config_tx)
2253                         t3_config_trace_filter(adapter, tp, 0,
2254                                                 t.invert_match,
2255                                                 t.trace_tx);
2256                 if (t.config_rx)
2257                         t3_config_trace_filter(adapter, tp, 1,
2258                                                 t.invert_match,
2259                                                 t.trace_rx);
2260                 break;
2261         }
2262         default:
2263                 return -EOPNOTSUPP;
2264         }
2265         return 0;
2266 }
2267
2268 static int cxgb_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
2269 {
2270         struct mii_ioctl_data *data = if_mii(req);
2271         struct port_info *pi = netdev_priv(dev);
2272         struct adapter *adapter = pi->adapter;
2273         int ret, mmd;
2274
2275         switch (cmd) {
2276         case SIOCGMIIPHY:
2277                 data->phy_id = pi->phy.addr;
2278                 /* FALLTHRU */
2279         case SIOCGMIIREG:{
2280                 u32 val;
2281                 struct cphy *phy = &pi->phy;
2282
2283                 if (!phy->mdio_read)
2284                         return -EOPNOTSUPP;
2285                 if (is_10G(adapter)) {
2286                         mmd = data->phy_id >> 8;
2287                         if (!mmd)
2288                                 mmd = MDIO_DEV_PCS;
2289                         else if (mmd > MDIO_DEV_VEND2)
2290                                 return -EINVAL;
2291
2292                         ret =
2293                                 phy->mdio_read(adapter, data->phy_id & 0x1f,
2294                                                 mmd, data->reg_num, &val);
2295                 } else
2296                         ret =
2297                                 phy->mdio_read(adapter, data->phy_id & 0x1f,
2298                                                 0, data->reg_num & 0x1f,
2299                                                 &val);
2300                 if (!ret)
2301                         data->val_out = val;
2302                 break;
2303         }
2304         case SIOCSMIIREG:{
2305                 struct cphy *phy = &pi->phy;
2306
2307                 if (!capable(CAP_NET_ADMIN))
2308                         return -EPERM;
2309                 if (!phy->mdio_write)
2310                         return -EOPNOTSUPP;
2311                 if (is_10G(adapter)) {
2312                         mmd = data->phy_id >> 8;
2313                         if (!mmd)
2314                                 mmd = MDIO_DEV_PCS;
2315                         else if (mmd > MDIO_DEV_VEND2)
2316                                 return -EINVAL;
2317
2318                         ret =
2319                                 phy->mdio_write(adapter,
2320                                                 data->phy_id & 0x1f, mmd,
2321                                                 data->reg_num,
2322                                                 data->val_in);
2323                 } else
2324                         ret =
2325                                 phy->mdio_write(adapter,
2326                                                 data->phy_id & 0x1f, 0,
2327                                                 data->reg_num & 0x1f,
2328                                                 data->val_in);
2329                 break;
2330         }
2331         case SIOCCHIOCTL:
2332                 return cxgb_extension_ioctl(dev, req->ifr_data);
2333         default:
2334                 return -EOPNOTSUPP;
2335         }
2336         return ret;
2337 }
2338
2339 static int cxgb_change_mtu(struct net_device *dev, int new_mtu)
2340 {
2341         struct port_info *pi = netdev_priv(dev);
2342         struct adapter *adapter = pi->adapter;
2343         int ret;
2344
2345         if (new_mtu < 81)       /* accommodate SACK */
2346                 return -EINVAL;
2347         if ((ret = t3_mac_set_mtu(&pi->mac, new_mtu)))
2348                 return ret;
2349         dev->mtu = new_mtu;
2350         init_port_mtus(adapter);
2351         if (adapter->params.rev == 0 && offload_running(adapter))
2352                 t3_load_mtus(adapter, adapter->params.mtus,
2353                              adapter->params.a_wnd, adapter->params.b_wnd,
2354                              adapter->port[0]->mtu);
2355         return 0;
2356 }
2357
2358 static int cxgb_set_mac_addr(struct net_device *dev, void *p)
2359 {
2360         struct port_info *pi = netdev_priv(dev);
2361         struct adapter *adapter = pi->adapter;
2362         struct sockaddr *addr = p;
2363
2364         if (!is_valid_ether_addr(addr->sa_data))
2365                 return -EINVAL;
2366
2367         memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
2368         t3_mac_set_address(&pi->mac, 0, dev->dev_addr);
2369         if (offload_running(adapter))
2370                 write_smt_entry(adapter, pi->port_id);
2371         return 0;
2372 }
2373
2374 /**
2375  * t3_synchronize_rx - wait for current Rx processing on a port to complete
2376  * @adap: the adapter
2377  * @p: the port
2378  *
2379  * Ensures that current Rx processing on any of the queues associated with
2380  * the given port completes before returning.  We do this by acquiring and
2381  * releasing the locks of the response queues associated with the port.
2382  */
2383 static void t3_synchronize_rx(struct adapter *adap, const struct port_info *p)
2384 {
2385         int i;
2386
2387         for (i = p->first_qset; i < p->first_qset + p->nqsets; i++) {
2388                 struct sge_rspq *q = &adap->sge.qs[i].rspq;
2389
2390                 spin_lock_irq(&q->lock);
2391                 spin_unlock_irq(&q->lock);
2392         }
2393 }
2394
2395 static void vlan_rx_register(struct net_device *dev, struct vlan_group *grp)
2396 {
2397         struct port_info *pi = netdev_priv(dev);
2398         struct adapter *adapter = pi->adapter;
2399
2400         pi->vlan_grp = grp;
2401         if (adapter->params.rev > 0)
2402                 t3_set_vlan_accel(adapter, 1 << pi->port_id, grp != NULL);
2403         else {
2404                 /* single control for all ports */
2405                 unsigned int i, have_vlans = 0;
2406                 for_each_port(adapter, i)
2407                     have_vlans |= adap2pinfo(adapter, i)->vlan_grp != NULL;
2408
2409                 t3_set_vlan_accel(adapter, 1, have_vlans);
2410         }
2411         t3_synchronize_rx(adapter, pi);
2412 }
2413
2414 #ifdef CONFIG_NET_POLL_CONTROLLER
2415 static void cxgb_netpoll(struct net_device *dev)
2416 {
2417         struct port_info *pi = netdev_priv(dev);
2418         struct adapter *adapter = pi->adapter;
2419         int qidx;
2420
2421         for (qidx = pi->first_qset; qidx < pi->first_qset + pi->nqsets; qidx++) {
2422                 struct sge_qset *qs = &adapter->sge.qs[qidx];
2423                 void *source;
2424
2425                 if (adapter->flags & USING_MSIX)
2426                         source = qs;
2427                 else
2428                         source = adapter;
2429
2430                 t3_intr_handler(adapter, qs->rspq.polling) (0, source);
2431         }
2432 }
2433 #endif
2434
2435 /*
2436  * Periodic accumulation of MAC statistics.
2437  */
2438 static void mac_stats_update(struct adapter *adapter)
2439 {
2440         int i;
2441
2442         for_each_port(adapter, i) {
2443                 struct net_device *dev = adapter->port[i];
2444                 struct port_info *p = netdev_priv(dev);
2445
2446                 if (netif_running(dev)) {
2447                         spin_lock(&adapter->stats_lock);
2448                         t3_mac_update_stats(&p->mac);
2449                         spin_unlock(&adapter->stats_lock);
2450                 }
2451         }
2452 }
2453
2454 static void check_link_status(struct adapter *adapter)
2455 {
2456         int i;
2457
2458         for_each_port(adapter, i) {
2459                 struct net_device *dev = adapter->port[i];
2460                 struct port_info *p = netdev_priv(dev);
2461
2462                 if (!(p->phy.caps & SUPPORTED_IRQ) && netif_running(dev))
2463                         t3_link_changed(adapter, i);
2464         }
2465 }
2466
2467 static void check_t3b2_mac(struct adapter *adapter)
2468 {
2469         int i;
2470
2471         if (!rtnl_trylock())    /* synchronize with ifdown */
2472                 return;
2473
2474         for_each_port(adapter, i) {
2475                 struct net_device *dev = adapter->port[i];
2476                 struct port_info *p = netdev_priv(dev);
2477                 int status;
2478
2479                 if (!netif_running(dev))
2480                         continue;
2481
2482                 status = 0;
2483                 if (netif_running(dev) && netif_carrier_ok(dev))
2484                         status = t3b2_mac_watchdog_task(&p->mac);
2485                 if (status == 1)
2486                         p->mac.stats.num_toggled++;
2487                 else if (status == 2) {
2488                         struct cmac *mac = &p->mac;
2489
2490                         t3_mac_set_mtu(mac, dev->mtu);
2491                         t3_mac_set_address(mac, 0, dev->dev_addr);
2492                         cxgb_set_rxmode(dev);
2493                         t3_link_start(&p->phy, mac, &p->link_config);
2494                         t3_mac_enable(mac, MAC_DIRECTION_RX | MAC_DIRECTION_TX);
2495                         t3_port_intr_enable(adapter, p->port_id);
2496                         p->mac.stats.num_resets++;
2497                 }
2498         }
2499         rtnl_unlock();
2500 }
2501
2502
2503 static void t3_adap_check_task(struct work_struct *work)
2504 {
2505         struct adapter *adapter = container_of(work, struct adapter,
2506                                                adap_check_task.work);
2507         const struct adapter_params *p = &adapter->params;
2508
2509         adapter->check_task_cnt++;
2510
2511         /* Check link status for PHYs without interrupts */
2512         if (p->linkpoll_period)
2513                 check_link_status(adapter);
2514
2515         /* Accumulate MAC stats if needed */
2516         if (!p->linkpoll_period ||
2517             (adapter->check_task_cnt * p->linkpoll_period) / 10 >=
2518             p->stats_update_period) {
2519                 mac_stats_update(adapter);
2520                 adapter->check_task_cnt = 0;
2521         }
2522
2523         if (p->rev == T3_REV_B2)
2524                 check_t3b2_mac(adapter);
2525
2526         /* Schedule the next check update if any port is active. */
2527         spin_lock_irq(&adapter->work_lock);
2528         if (adapter->open_device_map & PORT_MASK)
2529                 schedule_chk_task(adapter);
2530         spin_unlock_irq(&adapter->work_lock);
2531 }
2532
2533 /*
2534  * Processes external (PHY) interrupts in process context.
2535  */
2536 static void ext_intr_task(struct work_struct *work)
2537 {
2538         struct adapter *adapter = container_of(work, struct adapter,
2539                                                ext_intr_handler_task);
2540
2541         t3_phy_intr_handler(adapter);
2542
2543         /* Now reenable external interrupts */
2544         spin_lock_irq(&adapter->work_lock);
2545         if (adapter->slow_intr_mask) {
2546                 adapter->slow_intr_mask |= F_T3DBG;
2547                 t3_write_reg(adapter, A_PL_INT_CAUSE0, F_T3DBG);
2548                 t3_write_reg(adapter, A_PL_INT_ENABLE0,
2549                              adapter->slow_intr_mask);
2550         }
2551         spin_unlock_irq(&adapter->work_lock);
2552 }
2553
2554 /*
2555  * Interrupt-context handler for external (PHY) interrupts.
2556  */
2557 void t3_os_ext_intr_handler(struct adapter *adapter)
2558 {
2559         /*
2560          * Schedule a task to handle external interrupts as they may be slow
2561          * and we use a mutex to protect MDIO registers.  We disable PHY
2562          * interrupts in the meantime and let the task reenable them when
2563          * it's done.
2564          */
2565         spin_lock(&adapter->work_lock);
2566         if (adapter->slow_intr_mask) {
2567                 adapter->slow_intr_mask &= ~F_T3DBG;
2568                 t3_write_reg(adapter, A_PL_INT_ENABLE0,
2569                              adapter->slow_intr_mask);
2570                 queue_work(cxgb3_wq, &adapter->ext_intr_handler_task);
2571         }
2572         spin_unlock(&adapter->work_lock);
2573 }
2574
2575 static int t3_adapter_error(struct adapter *adapter, int reset)
2576 {
2577         int i, ret = 0;
2578
2579         /* Stop all ports */
2580         for_each_port(adapter, i) {
2581                 struct net_device *netdev = adapter->port[i];
2582
2583                 if (netif_running(netdev))
2584                         cxgb_close(netdev);
2585         }
2586
2587         if (is_offload(adapter) &&
2588             test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map))
2589                 offload_close(&adapter->tdev);
2590
2591         /* Stop SGE timers */
2592         t3_stop_sge_timers(adapter);
2593
2594         adapter->flags &= ~FULL_INIT_DONE;
2595
2596         if (reset)
2597                 ret = t3_reset_adapter(adapter);
2598
2599         pci_disable_device(adapter->pdev);
2600
2601         return ret;
2602 }
2603
2604 static int t3_reenable_adapter(struct adapter *adapter)
2605 {
2606         if (pci_enable_device(adapter->pdev)) {
2607                 dev_err(&adapter->pdev->dev,
2608                         "Cannot re-enable PCI device after reset.\n");
2609                 goto err;
2610         }
2611         pci_set_master(adapter->pdev);
2612         pci_restore_state(adapter->pdev);
2613
2614         /* Free sge resources */
2615         t3_free_sge_resources(adapter);
2616
2617         if (t3_replay_prep_adapter(adapter))
2618                 goto err;
2619
2620         return 0;
2621 err:
2622         return -1;
2623 }
2624
2625 static void t3_resume_ports(struct adapter *adapter)
2626 {
2627         int i;
2628
2629         /* Restart the ports */
2630         for_each_port(adapter, i) {
2631                 struct net_device *netdev = adapter->port[i];
2632
2633                 if (netif_running(netdev)) {
2634                         if (cxgb_open(netdev)) {
2635                                 dev_err(&adapter->pdev->dev,
2636                                         "can't bring device back up"
2637                                         " after reset\n");
2638                                 continue;
2639                         }
2640                 }
2641         }
2642 }
2643
2644 /*
2645  * processes a fatal error.
2646  * Bring the ports down, reset the chip, bring the ports back up.
2647  */
2648 static void fatal_error_task(struct work_struct *work)
2649 {
2650         struct adapter *adapter = container_of(work, struct adapter,
2651                                                fatal_error_handler_task);
2652         int err = 0;
2653
2654         rtnl_lock();
2655         err = t3_adapter_error(adapter, 1);
2656         if (!err)
2657                 err = t3_reenable_adapter(adapter);
2658         if (!err)
2659                 t3_resume_ports(adapter);
2660
2661         CH_ALERT(adapter, "adapter reset %s\n", err ? "failed" : "succeeded");
2662         rtnl_unlock();
2663 }
2664
2665 void t3_fatal_err(struct adapter *adapter)
2666 {
2667         unsigned int fw_status[4];
2668
2669         if (adapter->flags & FULL_INIT_DONE) {
2670                 t3_sge_stop(adapter);
2671                 t3_write_reg(adapter, A_XGM_TX_CTRL, 0);
2672                 t3_write_reg(adapter, A_XGM_RX_CTRL, 0);
2673                 t3_write_reg(adapter, XGM_REG(A_XGM_TX_CTRL, 1), 0);
2674                 t3_write_reg(adapter, XGM_REG(A_XGM_RX_CTRL, 1), 0);
2675
2676                 spin_lock(&adapter->work_lock);
2677                 t3_intr_disable(adapter);
2678                 queue_work(cxgb3_wq, &adapter->fatal_error_handler_task);
2679                 spin_unlock(&adapter->work_lock);
2680         }
2681         CH_ALERT(adapter, "encountered fatal error, operation suspended\n");
2682         if (!t3_cim_ctl_blk_read(adapter, 0xa0, 4, fw_status))
2683                 CH_ALERT(adapter, "FW status: 0x%x, 0x%x, 0x%x, 0x%x\n",
2684                          fw_status[0], fw_status[1],
2685                          fw_status[2], fw_status[3]);
2686
2687 }
2688
2689 /**
2690  * t3_io_error_detected - called when PCI error is detected
2691  * @pdev: Pointer to PCI device
2692  * @state: The current pci connection state
2693  *
2694  * This function is called after a PCI bus error affecting
2695  * this device has been detected.
2696  */
2697 static pci_ers_result_t t3_io_error_detected(struct pci_dev *pdev,
2698                                              pci_channel_state_t state)
2699 {
2700         struct adapter *adapter = pci_get_drvdata(pdev);
2701         int ret;
2702
2703         ret = t3_adapter_error(adapter, 0);
2704
2705         /* Request a slot reset. */
2706         return PCI_ERS_RESULT_NEED_RESET;
2707 }
2708
2709 /**
2710  * t3_io_slot_reset - called after the pci bus has been reset.
2711  * @pdev: Pointer to PCI device
2712  *
2713  * Restart the card from scratch, as if from a cold-boot.
2714  */
2715 static pci_ers_result_t t3_io_slot_reset(struct pci_dev *pdev)
2716 {
2717         struct adapter *adapter = pci_get_drvdata(pdev);
2718
2719         if (!t3_reenable_adapter(adapter))
2720                 return PCI_ERS_RESULT_RECOVERED;
2721
2722         return PCI_ERS_RESULT_DISCONNECT;
2723 }
2724
2725 /**
2726  * t3_io_resume - called when traffic can start flowing again.
2727  * @pdev: Pointer to PCI device
2728  *
2729  * This callback is called when the error recovery driver tells us that
2730  * its OK to resume normal operation.
2731  */
2732 static void t3_io_resume(struct pci_dev *pdev)
2733 {
2734         struct adapter *adapter = pci_get_drvdata(pdev);
2735
2736         t3_resume_ports(adapter);
2737 }
2738
2739 static struct pci_error_handlers t3_err_handler = {
2740         .error_detected = t3_io_error_detected,
2741         .slot_reset = t3_io_slot_reset,
2742         .resume = t3_io_resume,
2743 };
2744
2745 /*
2746  * Set the number of qsets based on the number of CPUs and the number of ports,
2747  * not to exceed the number of available qsets, assuming there are enough qsets
2748  * per port in HW.
2749  */
2750 static void set_nqsets(struct adapter *adap)
2751 {
2752         int i, j = 0;
2753         int num_cpus = num_online_cpus();
2754         int hwports = adap->params.nports;
2755         int nqsets = SGE_QSETS;
2756
2757         if (adap->params.rev > 0 && adap->flags & USING_MSIX) {
2758                 if (hwports == 2 &&
2759                     (hwports * nqsets > SGE_QSETS ||
2760                      num_cpus >= nqsets / hwports))
2761                         nqsets /= hwports;
2762                 if (nqsets > num_cpus)
2763                         nqsets = num_cpus;
2764                 if (nqsets < 1 || hwports == 4)
2765                         nqsets = 1;
2766         } else
2767                 nqsets = 1;
2768
2769         for_each_port(adap, i) {
2770                 struct port_info *pi = adap2pinfo(adap, i);
2771
2772                 pi->first_qset = j;
2773                 pi->nqsets = nqsets;
2774                 j = pi->first_qset + nqsets;
2775
2776                 dev_info(&adap->pdev->dev,
2777                          "Port %d using %d queue sets.\n", i, nqsets);
2778         }
2779 }
2780
2781 static int __devinit cxgb_enable_msix(struct adapter *adap)
2782 {
2783         struct msix_entry entries[SGE_QSETS + 1];
2784         int i, err;
2785
2786         for (i = 0; i < ARRAY_SIZE(entries); ++i)
2787                 entries[i].entry = i;
2788
2789         err = pci_enable_msix(adap->pdev, entries, ARRAY_SIZE(entries));
2790         if (!err) {
2791                 for (i = 0; i < ARRAY_SIZE(entries); ++i)
2792                         adap->msix_info[i].vec = entries[i].vector;
2793         } else if (err > 0)
2794                 dev_info(&adap->pdev->dev,
2795                        "only %d MSI-X vectors left, not using MSI-X\n", err);
2796         return err;
2797 }
2798
2799 static void __devinit print_port_info(struct adapter *adap,
2800                                       const struct adapter_info *ai)
2801 {
2802         static const char *pci_variant[] = {
2803                 "PCI", "PCI-X", "PCI-X ECC", "PCI-X 266", "PCI Express"
2804         };
2805
2806         int i;
2807         char buf[80];
2808
2809         if (is_pcie(adap))
2810                 snprintf(buf, sizeof(buf), "%s x%d",
2811                          pci_variant[adap->params.pci.variant],
2812                          adap->params.pci.width);
2813         else
2814                 snprintf(buf, sizeof(buf), "%s %dMHz/%d-bit",
2815                          pci_variant[adap->params.pci.variant],
2816                          adap->params.pci.speed, adap->params.pci.width);
2817
2818         for_each_port(adap, i) {
2819                 struct net_device *dev = adap->port[i];
2820                 const struct port_info *pi = netdev_priv(dev);
2821
2822                 if (!test_bit(i, &adap->registered_device_map))
2823                         continue;
2824                 printk(KERN_INFO "%s: %s %s %sNIC (rev %d) %s%s\n",
2825                        dev->name, ai->desc, pi->phy.desc,
2826                        is_offload(adap) ? "R" : "", adap->params.rev, buf,
2827                        (adap->flags & USING_MSIX) ? " MSI-X" :
2828                        (adap->flags & USING_MSI) ? " MSI" : "");
2829                 if (adap->name == dev->name && adap->params.vpd.mclk)
2830                         printk(KERN_INFO
2831                                "%s: %uMB CM, %uMB PMTX, %uMB PMRX, S/N: %s\n",
2832                                adap->name, t3_mc7_size(&adap->cm) >> 20,
2833                                t3_mc7_size(&adap->pmtx) >> 20,
2834                                t3_mc7_size(&adap->pmrx) >> 20,
2835                                adap->params.vpd.sn);
2836         }
2837 }
2838
2839 static const struct net_device_ops cxgb_netdev_ops = {
2840         .ndo_open               = cxgb_open,
2841         .ndo_stop               = cxgb_close,
2842         .ndo_start_xmit         = t3_eth_xmit,
2843         .ndo_get_stats          = cxgb_get_stats,
2844         .ndo_validate_addr      = eth_validate_addr,
2845         .ndo_set_multicast_list = cxgb_set_rxmode,
2846         .ndo_do_ioctl           = cxgb_ioctl,
2847         .ndo_change_mtu         = cxgb_change_mtu,
2848         .ndo_set_mac_address    = cxgb_set_mac_addr,
2849         .ndo_vlan_rx_register   = vlan_rx_register,
2850 #ifdef CONFIG_NET_POLL_CONTROLLER
2851         .ndo_poll_controller    = cxgb_netpoll,
2852 #endif
2853 };
2854
2855 static int __devinit init_one(struct pci_dev *pdev,
2856                               const struct pci_device_id *ent)
2857 {
2858         static int version_printed;
2859
2860         int i, err, pci_using_dac = 0;
2861         unsigned long mmio_start, mmio_len;
2862         const struct adapter_info *ai;
2863         struct adapter *adapter = NULL;
2864         struct port_info *pi;
2865
2866         if (!version_printed) {
2867                 printk(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
2868                 ++version_printed;
2869         }
2870
2871         if (!cxgb3_wq) {
2872                 cxgb3_wq = create_singlethread_workqueue(DRV_NAME);
2873                 if (!cxgb3_wq) {
2874                         printk(KERN_ERR DRV_NAME
2875                                ": cannot initialize work queue\n");
2876                         return -ENOMEM;
2877                 }
2878         }
2879
2880         err = pci_request_regions(pdev, DRV_NAME);
2881         if (err) {
2882                 /* Just info, some other driver may have claimed the device. */
2883                 dev_info(&pdev->dev, "cannot obtain PCI resources\n");
2884                 return err;
2885         }
2886
2887         err = pci_enable_device(pdev);
2888         if (err) {
2889                 dev_err(&pdev->dev, "cannot enable PCI device\n");
2890                 goto out_release_regions;
2891         }
2892
2893         if (!pci_set_dma_mask(pdev, DMA_64BIT_MASK)) {
2894                 pci_using_dac = 1;
2895                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
2896                 if (err) {
2897                         dev_err(&pdev->dev, "unable to obtain 64-bit DMA for "
2898                                "coherent allocations\n");
2899                         goto out_disable_device;
2900                 }
2901         } else if ((err = pci_set_dma_mask(pdev, DMA_32BIT_MASK)) != 0) {
2902                 dev_err(&pdev->dev, "no usable DMA configuration\n");
2903                 goto out_disable_device;
2904         }
2905
2906         pci_set_master(pdev);
2907         pci_save_state(pdev);
2908
2909         mmio_start = pci_resource_start(pdev, 0);
2910         mmio_len = pci_resource_len(pdev, 0);
2911         ai = t3_get_adapter_info(ent->driver_data);
2912
2913         adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
2914         if (!adapter) {
2915                 err = -ENOMEM;
2916                 goto out_disable_device;
2917         }
2918
2919         adapter->regs = ioremap_nocache(mmio_start, mmio_len);
2920         if (!adapter->regs) {
2921                 dev_err(&pdev->dev, "cannot map device registers\n");
2922                 err = -ENOMEM;
2923                 goto out_free_adapter;
2924         }
2925
2926         adapter->pdev = pdev;
2927         adapter->name = pci_name(pdev);
2928         adapter->msg_enable = dflt_msg_enable;
2929         adapter->mmio_len = mmio_len;
2930
2931         mutex_init(&adapter->mdio_lock);
2932         spin_lock_init(&adapter->work_lock);
2933         spin_lock_init(&adapter->stats_lock);
2934
2935         INIT_LIST_HEAD(&adapter->adapter_list);
2936         INIT_WORK(&adapter->ext_intr_handler_task, ext_intr_task);
2937         INIT_WORK(&adapter->fatal_error_handler_task, fatal_error_task);
2938         INIT_DELAYED_WORK(&adapter->adap_check_task, t3_adap_check_task);
2939
2940         for (i = 0; i < ai->nports; ++i) {
2941                 struct net_device *netdev;
2942
2943                 netdev = alloc_etherdev_mq(sizeof(struct port_info), SGE_QSETS);
2944                 if (!netdev) {
2945                         err = -ENOMEM;
2946                         goto out_free_dev;
2947                 }
2948
2949                 SET_NETDEV_DEV(netdev, &pdev->dev);
2950
2951                 adapter->port[i] = netdev;
2952                 pi = netdev_priv(netdev);
2953                 pi->adapter = adapter;
2954                 pi->rx_offload = T3_RX_CSUM | T3_LRO;
2955                 pi->port_id = i;
2956                 netif_carrier_off(netdev);
2957                 netif_tx_stop_all_queues(netdev);
2958                 netdev->irq = pdev->irq;
2959                 netdev->mem_start = mmio_start;
2960                 netdev->mem_end = mmio_start + mmio_len - 1;
2961                 netdev->features |= NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_TSO;
2962                 netdev->features |= NETIF_F_LLTX;
2963                 netdev->features |= NETIF_F_LRO;
2964                 if (pci_using_dac)
2965                         netdev->features |= NETIF_F_HIGHDMA;
2966
2967                 netdev->features |= NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX;
2968                 netdev->netdev_ops = &cxgb_netdev_ops;
2969                 SET_ETHTOOL_OPS(netdev, &cxgb_ethtool_ops);
2970         }
2971
2972         pci_set_drvdata(pdev, adapter);
2973         if (t3_prep_adapter(adapter, ai, 1) < 0) {
2974                 err = -ENODEV;
2975                 goto out_free_dev;
2976         }
2977
2978         /*
2979          * The card is now ready to go.  If any errors occur during device
2980          * registration we do not fail the whole card but rather proceed only
2981          * with the ports we manage to register successfully.  However we must
2982          * register at least one net device.
2983          */
2984         for_each_port(adapter, i) {
2985                 err = register_netdev(adapter->port[i]);
2986                 if (err)
2987                         dev_warn(&pdev->dev,
2988                                  "cannot register net device %s, skipping\n",
2989                                  adapter->port[i]->name);
2990                 else {
2991                         /*
2992                          * Change the name we use for messages to the name of
2993                          * the first successfully registered interface.
2994                          */
2995                         if (!adapter->registered_device_map)
2996                                 adapter->name = adapter->port[i]->name;
2997
2998                         __set_bit(i, &adapter->registered_device_map);
2999                 }
3000         }
3001         if (!adapter->registered_device_map) {
3002                 dev_err(&pdev->dev, "could not register any net devices\n");
3003                 goto out_free_dev;
3004         }
3005
3006         /* Driver's ready. Reflect it on LEDs */
3007         t3_led_ready(adapter);
3008
3009         if (is_offload(adapter)) {
3010                 __set_bit(OFFLOAD_DEVMAP_BIT, &adapter->registered_device_map);
3011                 cxgb3_adapter_ofld(adapter);
3012         }
3013
3014         /* See what interrupts we'll be using */
3015         if (msi > 1 && cxgb_enable_msix(adapter) == 0)
3016                 adapter->flags |= USING_MSIX;
3017         else if (msi > 0 && pci_enable_msi(pdev) == 0)
3018                 adapter->flags |= USING_MSI;
3019
3020         set_nqsets(adapter);
3021
3022         err = sysfs_create_group(&adapter->port[0]->dev.kobj,
3023                                  &cxgb3_attr_group);
3024
3025         print_port_info(adapter, ai);
3026         return 0;
3027
3028 out_free_dev:
3029         iounmap(adapter->regs);
3030         for (i = ai->nports - 1; i >= 0; --i)
3031                 if (adapter->port[i])
3032                         free_netdev(adapter->port[i]);
3033
3034 out_free_adapter:
3035         kfree(adapter);
3036
3037 out_disable_device:
3038         pci_disable_device(pdev);
3039 out_release_regions:
3040         pci_release_regions(pdev);
3041         pci_set_drvdata(pdev, NULL);
3042         return err;
3043 }
3044
3045 static void __devexit remove_one(struct pci_dev *pdev)
3046 {
3047         struct adapter *adapter = pci_get_drvdata(pdev);
3048
3049         if (adapter) {
3050                 int i;
3051
3052                 t3_sge_stop(adapter);
3053                 sysfs_remove_group(&adapter->port[0]->dev.kobj,
3054                                    &cxgb3_attr_group);
3055
3056                 if (is_offload(adapter)) {
3057                         cxgb3_adapter_unofld(adapter);
3058                         if (test_bit(OFFLOAD_DEVMAP_BIT,
3059                                      &adapter->open_device_map))
3060                                 offload_close(&adapter->tdev);
3061                 }
3062
3063                 for_each_port(adapter, i)
3064                     if (test_bit(i, &adapter->registered_device_map))
3065                         unregister_netdev(adapter->port[i]);
3066
3067                 t3_stop_sge_timers(adapter);
3068                 t3_free_sge_resources(adapter);
3069                 cxgb_disable_msi(adapter);
3070
3071                 for_each_port(adapter, i)
3072                         if (adapter->port[i])
3073                                 free_netdev(adapter->port[i]);
3074
3075                 iounmap(adapter->regs);
3076                 kfree(adapter);
3077                 pci_release_regions(pdev);
3078                 pci_disable_device(pdev);
3079                 pci_set_drvdata(pdev, NULL);
3080         }
3081 }
3082
3083 static struct pci_driver driver = {
3084         .name = DRV_NAME,
3085         .id_table = cxgb3_pci_tbl,
3086         .probe = init_one,
3087         .remove = __devexit_p(remove_one),
3088         .err_handler = &t3_err_handler,
3089 };
3090
3091 static int __init cxgb3_init_module(void)
3092 {
3093         int ret;
3094
3095         cxgb3_offload_init();
3096
3097         ret = pci_register_driver(&driver);
3098         return ret;
3099 }
3100
3101 static void __exit cxgb3_cleanup_module(void)
3102 {
3103         pci_unregister_driver(&driver);
3104         if (cxgb3_wq)
3105                 destroy_workqueue(cxgb3_wq);
3106 }
3107
3108 module_init(cxgb3_init_module);
3109 module_exit(cxgb3_cleanup_module);