Merge git://git.kernel.org/pub/scm/linux/kernel/git/sfrench/cifs-2.6
[linux-2.6] / drivers / net / cxgb3 / cxgb3_main.c
1 /*
2  * Copyright (c) 2003-2007 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 <asm/uaccess.h>
46
47 #include "common.h"
48 #include "cxgb3_ioctl.h"
49 #include "regs.h"
50 #include "cxgb3_offload.h"
51 #include "version.h"
52
53 #include "cxgb3_ctl_defs.h"
54 #include "t3_cpl.h"
55 #include "firmware_exports.h"
56
57 enum {
58         MAX_TXQ_ENTRIES = 16384,
59         MAX_CTRL_TXQ_ENTRIES = 1024,
60         MAX_RSPQ_ENTRIES = 16384,
61         MAX_RX_BUFFERS = 16384,
62         MAX_RX_JUMBO_BUFFERS = 16384,
63         MIN_TXQ_ENTRIES = 4,
64         MIN_CTRL_TXQ_ENTRIES = 4,
65         MIN_RSPQ_ENTRIES = 32,
66         MIN_FL_ENTRIES = 32
67 };
68
69 #define PORT_MASK ((1 << MAX_NPORTS) - 1)
70
71 #define DFLT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK | \
72                          NETIF_MSG_TIMER | NETIF_MSG_IFDOWN | NETIF_MSG_IFUP |\
73                          NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
74
75 #define EEPROM_MAGIC 0x38E2F10C
76
77 #define CH_DEVICE(devid, ssid, idx) \
78         { PCI_VENDOR_ID_CHELSIO, devid, PCI_ANY_ID, ssid, 0, 0, idx }
79
80 static const struct pci_device_id cxgb3_pci_tbl[] = {
81         CH_DEVICE(0x20, 1, 0),  /* PE9000 */
82         CH_DEVICE(0x21, 1, 1),  /* T302E */
83         CH_DEVICE(0x22, 1, 2),  /* T310E */
84         CH_DEVICE(0x23, 1, 3),  /* T320X */
85         CH_DEVICE(0x24, 1, 1),  /* T302X */
86         CH_DEVICE(0x25, 1, 3),  /* T320E */
87         CH_DEVICE(0x26, 1, 2),  /* T310X */
88         CH_DEVICE(0x30, 1, 2),  /* T3B10 */
89         CH_DEVICE(0x31, 1, 3),  /* T3B20 */
90         CH_DEVICE(0x32, 1, 1),  /* T3B02 */
91         {0,}
92 };
93
94 MODULE_DESCRIPTION(DRV_DESC);
95 MODULE_AUTHOR("Chelsio Communications");
96 MODULE_LICENSE("Dual BSD/GPL");
97 MODULE_VERSION(DRV_VERSION);
98 MODULE_DEVICE_TABLE(pci, cxgb3_pci_tbl);
99
100 static int dflt_msg_enable = DFLT_MSG_ENABLE;
101
102 module_param(dflt_msg_enable, int, 0644);
103 MODULE_PARM_DESC(dflt_msg_enable, "Chelsio T3 default message enable bitmap");
104
105 /*
106  * The driver uses the best interrupt scheme available on a platform in the
107  * order MSI-X, MSI, legacy pin interrupts.  This parameter determines which
108  * of these schemes the driver may consider as follows:
109  *
110  * msi = 2: choose from among all three options
111  * msi = 1: only consider MSI and pin interrupts
112  * msi = 0: force pin interrupts
113  */
114 static int msi = 2;
115
116 module_param(msi, int, 0644);
117 MODULE_PARM_DESC(msi, "whether to use MSI or MSI-X");
118
119 /*
120  * The driver enables offload as a default.
121  * To disable it, use ofld_disable = 1.
122  */
123
124 static int ofld_disable = 0;
125
126 module_param(ofld_disable, int, 0644);
127 MODULE_PARM_DESC(ofld_disable, "whether to enable offload at init time or not");
128
129 /*
130  * We have work elements that we need to cancel when an interface is taken
131  * down.  Normally the work elements would be executed by keventd but that
132  * can deadlock because of linkwatch.  If our close method takes the rtnl
133  * lock and linkwatch is ahead of our work elements in keventd, linkwatch
134  * will block keventd as it needs the rtnl lock, and we'll deadlock waiting
135  * for our work to complete.  Get our own work queue to solve this.
136  */
137 static struct workqueue_struct *cxgb3_wq;
138
139 /**
140  *      link_report - show link status and link speed/duplex
141  *      @p: the port whose settings are to be reported
142  *
143  *      Shows the link status, speed, and duplex of a port.
144  */
145 static void link_report(struct net_device *dev)
146 {
147         if (!netif_carrier_ok(dev))
148                 printk(KERN_INFO "%s: link down\n", dev->name);
149         else {
150                 const char *s = "10Mbps";
151                 const struct port_info *p = netdev_priv(dev);
152
153                 switch (p->link_config.speed) {
154                 case SPEED_10000:
155                         s = "10Gbps";
156                         break;
157                 case SPEED_1000:
158                         s = "1000Mbps";
159                         break;
160                 case SPEED_100:
161                         s = "100Mbps";
162                         break;
163                 }
164
165                 printk(KERN_INFO "%s: link up, %s, %s-duplex\n", dev->name, s,
166                        p->link_config.duplex == DUPLEX_FULL ? "full" : "half");
167         }
168 }
169
170 /**
171  *      t3_os_link_changed - handle link status changes
172  *      @adapter: the adapter associated with the link change
173  *      @port_id: the port index whose limk status has changed
174  *      @link_stat: the new status of the link
175  *      @speed: the new speed setting
176  *      @duplex: the new duplex setting
177  *      @pause: the new flow-control setting
178  *
179  *      This is the OS-dependent handler for link status changes.  The OS
180  *      neutral handler takes care of most of the processing for these events,
181  *      then calls this handler for any OS-specific processing.
182  */
183 void t3_os_link_changed(struct adapter *adapter, int port_id, int link_stat,
184                         int speed, int duplex, int pause)
185 {
186         struct net_device *dev = adapter->port[port_id];
187
188         /* Skip changes from disabled ports. */
189         if (!netif_running(dev))
190                 return;
191
192         if (link_stat != netif_carrier_ok(dev)) {
193                 if (link_stat)
194                         netif_carrier_on(dev);
195                 else
196                         netif_carrier_off(dev);
197                 link_report(dev);
198         }
199 }
200
201 static void cxgb_set_rxmode(struct net_device *dev)
202 {
203         struct t3_rx_mode rm;
204         struct port_info *pi = netdev_priv(dev);
205
206         init_rx_mode(&rm, dev, dev->mc_list);
207         t3_mac_set_rx_mode(&pi->mac, &rm);
208 }
209
210 /**
211  *      link_start - enable a port
212  *      @dev: the device to enable
213  *
214  *      Performs the MAC and PHY actions needed to enable a port.
215  */
216 static void link_start(struct net_device *dev)
217 {
218         struct t3_rx_mode rm;
219         struct port_info *pi = netdev_priv(dev);
220         struct cmac *mac = &pi->mac;
221
222         init_rx_mode(&rm, dev, dev->mc_list);
223         t3_mac_reset(mac);
224         t3_mac_set_mtu(mac, dev->mtu);
225         t3_mac_set_address(mac, 0, dev->dev_addr);
226         t3_mac_set_rx_mode(mac, &rm);
227         t3_link_start(&pi->phy, mac, &pi->link_config);
228         t3_mac_enable(mac, MAC_DIRECTION_RX | MAC_DIRECTION_TX);
229 }
230
231 static inline void cxgb_disable_msi(struct adapter *adapter)
232 {
233         if (adapter->flags & USING_MSIX) {
234                 pci_disable_msix(adapter->pdev);
235                 adapter->flags &= ~USING_MSIX;
236         } else if (adapter->flags & USING_MSI) {
237                 pci_disable_msi(adapter->pdev);
238                 adapter->flags &= ~USING_MSI;
239         }
240 }
241
242 /*
243  * Interrupt handler for asynchronous events used with MSI-X.
244  */
245 static irqreturn_t t3_async_intr_handler(int irq, void *cookie)
246 {
247         t3_slow_intr_handler(cookie);
248         return IRQ_HANDLED;
249 }
250
251 /*
252  * Name the MSI-X interrupts.
253  */
254 static void name_msix_vecs(struct adapter *adap)
255 {
256         int i, j, msi_idx = 1, n = sizeof(adap->msix_info[0].desc) - 1;
257
258         snprintf(adap->msix_info[0].desc, n, "%s", adap->name);
259         adap->msix_info[0].desc[n] = 0;
260
261         for_each_port(adap, j) {
262                 struct net_device *d = adap->port[j];
263                 const struct port_info *pi = netdev_priv(d);
264
265                 for (i = 0; i < pi->nqsets; i++, msi_idx++) {
266                         snprintf(adap->msix_info[msi_idx].desc, n,
267                                  "%s (queue %d)", d->name, i);
268                         adap->msix_info[msi_idx].desc[n] = 0;
269                 }
270         }
271 }
272
273 static int request_msix_data_irqs(struct adapter *adap)
274 {
275         int i, j, err, qidx = 0;
276
277         for_each_port(adap, i) {
278                 int nqsets = adap2pinfo(adap, i)->nqsets;
279
280                 for (j = 0; j < nqsets; ++j) {
281                         err = request_irq(adap->msix_info[qidx + 1].vec,
282                                           t3_intr_handler(adap,
283                                                           adap->sge.qs[qidx].
284                                                           rspq.polling), 0,
285                                           adap->msix_info[qidx + 1].desc,
286                                           &adap->sge.qs[qidx]);
287                         if (err) {
288                                 while (--qidx >= 0)
289                                         free_irq(adap->msix_info[qidx + 1].vec,
290                                                  &adap->sge.qs[qidx]);
291                                 return err;
292                         }
293                         qidx++;
294                 }
295         }
296         return 0;
297 }
298
299 /**
300  *      setup_rss - configure RSS
301  *      @adap: the adapter
302  *
303  *      Sets up RSS to distribute packets to multiple receive queues.  We
304  *      configure the RSS CPU lookup table to distribute to the number of HW
305  *      receive queues, and the response queue lookup table to narrow that
306  *      down to the response queues actually configured for each port.
307  *      We always configure the RSS mapping for two ports since the mapping
308  *      table has plenty of entries.
309  */
310 static void setup_rss(struct adapter *adap)
311 {
312         int i;
313         unsigned int nq0 = adap2pinfo(adap, 0)->nqsets;
314         unsigned int nq1 = adap->port[1] ? adap2pinfo(adap, 1)->nqsets : 1;
315         u8 cpus[SGE_QSETS + 1];
316         u16 rspq_map[RSS_TABLE_SIZE];
317
318         for (i = 0; i < SGE_QSETS; ++i)
319                 cpus[i] = i;
320         cpus[SGE_QSETS] = 0xff; /* terminator */
321
322         for (i = 0; i < RSS_TABLE_SIZE / 2; ++i) {
323                 rspq_map[i] = i % nq0;
324                 rspq_map[i + RSS_TABLE_SIZE / 2] = (i % nq1) + nq0;
325         }
326
327         t3_config_rss(adap, F_RQFEEDBACKENABLE | F_TNLLKPEN | F_TNLMAPEN |
328                       F_TNLPRTEN | F_TNL2TUPEN | F_TNL4TUPEN |
329                       V_RRCPLCPUSIZE(6), cpus, rspq_map);
330 }
331
332 /*
333  * If we have multiple receive queues per port serviced by NAPI we need one
334  * netdevice per queue as NAPI operates on netdevices.  We already have one
335  * netdevice, namely the one associated with the interface, so we use dummy
336  * ones for any additional queues.  Note that these netdevices exist purely
337  * so that NAPI has something to work with, they do not represent network
338  * ports and are not registered.
339  */
340 static int init_dummy_netdevs(struct adapter *adap)
341 {
342         int i, j, dummy_idx = 0;
343         struct net_device *nd;
344
345         for_each_port(adap, i) {
346                 struct net_device *dev = adap->port[i];
347                 const struct port_info *pi = netdev_priv(dev);
348
349                 for (j = 0; j < pi->nqsets - 1; j++) {
350                         if (!adap->dummy_netdev[dummy_idx]) {
351                                 nd = alloc_netdev(0, "", ether_setup);
352                                 if (!nd)
353                                         goto free_all;
354
355                                 nd->priv = adap;
356                                 nd->weight = 64;
357                                 set_bit(__LINK_STATE_START, &nd->state);
358                                 adap->dummy_netdev[dummy_idx] = nd;
359                         }
360                         strcpy(adap->dummy_netdev[dummy_idx]->name, dev->name);
361                         dummy_idx++;
362                 }
363         }
364         return 0;
365
366 free_all:
367         while (--dummy_idx >= 0) {
368                 free_netdev(adap->dummy_netdev[dummy_idx]);
369                 adap->dummy_netdev[dummy_idx] = NULL;
370         }
371         return -ENOMEM;
372 }
373
374 /*
375  * Wait until all NAPI handlers are descheduled.  This includes the handlers of
376  * both netdevices representing interfaces and the dummy ones for the extra
377  * queues.
378  */
379 static void quiesce_rx(struct adapter *adap)
380 {
381         int i;
382         struct net_device *dev;
383
384         for_each_port(adap, i) {
385                 dev = adap->port[i];
386                 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state))
387                         msleep(1);
388         }
389
390         for (i = 0; i < ARRAY_SIZE(adap->dummy_netdev); i++) {
391                 dev = adap->dummy_netdev[i];
392                 if (dev)
393                         while (test_bit(__LINK_STATE_RX_SCHED, &dev->state))
394                                 msleep(1);
395         }
396 }
397
398 /**
399  *      setup_sge_qsets - configure SGE Tx/Rx/response queues
400  *      @adap: the adapter
401  *
402  *      Determines how many sets of SGE queues to use and initializes them.
403  *      We support multiple queue sets per port if we have MSI-X, otherwise
404  *      just one queue set per port.
405  */
406 static int setup_sge_qsets(struct adapter *adap)
407 {
408         int i, j, err, irq_idx = 0, qset_idx = 0, dummy_dev_idx = 0;
409         unsigned int ntxq = is_offload(adap) ? SGE_TXQ_PER_SET : 1;
410
411         if (adap->params.rev > 0 && !(adap->flags & USING_MSI))
412                 irq_idx = -1;
413
414         for_each_port(adap, i) {
415                 struct net_device *dev = adap->port[i];
416                 const struct port_info *pi = netdev_priv(dev);
417
418                 for (j = 0; j < pi->nqsets; ++j, ++qset_idx) {
419                         err = t3_sge_alloc_qset(adap, qset_idx, 1,
420                                 (adap->flags & USING_MSIX) ? qset_idx + 1 :
421                                                              irq_idx,
422                                 &adap->params.sge.qset[qset_idx], ntxq,
423                                 j == 0 ? dev :
424                                          adap-> dummy_netdev[dummy_dev_idx++]);
425                         if (err) {
426                                 t3_free_sge_resources(adap);
427                                 return err;
428                         }
429                 }
430         }
431
432         return 0;
433 }
434
435 static ssize_t attr_show(struct device *d, struct device_attribute *attr,
436                          char *buf,
437                          ssize_t(*format) (struct net_device *, char *))
438 {
439         ssize_t len;
440
441         /* Synchronize with ioctls that may shut down the device */
442         rtnl_lock();
443         len = (*format) (to_net_dev(d), buf);
444         rtnl_unlock();
445         return len;
446 }
447
448 static ssize_t attr_store(struct device *d, struct device_attribute *attr,
449                           const char *buf, size_t len,
450                           ssize_t(*set) (struct net_device *, unsigned int),
451                           unsigned int min_val, unsigned int max_val)
452 {
453         char *endp;
454         ssize_t ret;
455         unsigned int val;
456
457         if (!capable(CAP_NET_ADMIN))
458                 return -EPERM;
459
460         val = simple_strtoul(buf, &endp, 0);
461         if (endp == buf || val < min_val || val > max_val)
462                 return -EINVAL;
463
464         rtnl_lock();
465         ret = (*set) (to_net_dev(d), val);
466         if (!ret)
467                 ret = len;
468         rtnl_unlock();
469         return ret;
470 }
471
472 #define CXGB3_SHOW(name, val_expr) \
473 static ssize_t format_##name(struct net_device *dev, char *buf) \
474 { \
475         struct adapter *adap = dev->priv; \
476         return sprintf(buf, "%u\n", val_expr); \
477 } \
478 static ssize_t show_##name(struct device *d, struct device_attribute *attr, \
479                            char *buf) \
480 { \
481         return attr_show(d, attr, buf, format_##name); \
482 }
483
484 static ssize_t set_nfilters(struct net_device *dev, unsigned int val)
485 {
486         struct adapter *adap = dev->priv;
487
488         if (adap->flags & FULL_INIT_DONE)
489                 return -EBUSY;
490         if (val && adap->params.rev == 0)
491                 return -EINVAL;
492         if (val > t3_mc5_size(&adap->mc5) - adap->params.mc5.nservers)
493                 return -EINVAL;
494         adap->params.mc5.nfilters = val;
495         return 0;
496 }
497
498 static ssize_t store_nfilters(struct device *d, struct device_attribute *attr,
499                               const char *buf, size_t len)
500 {
501         return attr_store(d, attr, buf, len, set_nfilters, 0, ~0);
502 }
503
504 static ssize_t set_nservers(struct net_device *dev, unsigned int val)
505 {
506         struct adapter *adap = dev->priv;
507
508         if (adap->flags & FULL_INIT_DONE)
509                 return -EBUSY;
510         if (val > t3_mc5_size(&adap->mc5) - adap->params.mc5.nfilters)
511                 return -EINVAL;
512         adap->params.mc5.nservers = val;
513         return 0;
514 }
515
516 static ssize_t store_nservers(struct device *d, struct device_attribute *attr,
517                               const char *buf, size_t len)
518 {
519         return attr_store(d, attr, buf, len, set_nservers, 0, ~0);
520 }
521
522 #define CXGB3_ATTR_R(name, val_expr) \
523 CXGB3_SHOW(name, val_expr) \
524 static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
525
526 #define CXGB3_ATTR_RW(name, val_expr, store_method) \
527 CXGB3_SHOW(name, val_expr) \
528 static DEVICE_ATTR(name, S_IRUGO | S_IWUSR, show_##name, store_method)
529
530 CXGB3_ATTR_R(cam_size, t3_mc5_size(&adap->mc5));
531 CXGB3_ATTR_RW(nfilters, adap->params.mc5.nfilters, store_nfilters);
532 CXGB3_ATTR_RW(nservers, adap->params.mc5.nservers, store_nservers);
533
534 static struct attribute *cxgb3_attrs[] = {
535         &dev_attr_cam_size.attr,
536         &dev_attr_nfilters.attr,
537         &dev_attr_nservers.attr,
538         NULL
539 };
540
541 static struct attribute_group cxgb3_attr_group = {.attrs = cxgb3_attrs };
542
543 static ssize_t tm_attr_show(struct device *d, struct device_attribute *attr,
544                             char *buf, int sched)
545 {
546         ssize_t len;
547         unsigned int v, addr, bpt, cpt;
548         struct adapter *adap = to_net_dev(d)->priv;
549
550         addr = A_TP_TX_MOD_Q1_Q0_RATE_LIMIT - sched / 2;
551         rtnl_lock();
552         t3_write_reg(adap, A_TP_TM_PIO_ADDR, addr);
553         v = t3_read_reg(adap, A_TP_TM_PIO_DATA);
554         if (sched & 1)
555                 v >>= 16;
556         bpt = (v >> 8) & 0xff;
557         cpt = v & 0xff;
558         if (!cpt)
559                 len = sprintf(buf, "disabled\n");
560         else {
561                 v = (adap->params.vpd.cclk * 1000) / cpt;
562                 len = sprintf(buf, "%u Kbps\n", (v * bpt) / 125);
563         }
564         rtnl_unlock();
565         return len;
566 }
567
568 static ssize_t tm_attr_store(struct device *d, struct device_attribute *attr,
569                              const char *buf, size_t len, int sched)
570 {
571         char *endp;
572         ssize_t ret;
573         unsigned int val;
574         struct adapter *adap = to_net_dev(d)->priv;
575
576         if (!capable(CAP_NET_ADMIN))
577                 return -EPERM;
578
579         val = simple_strtoul(buf, &endp, 0);
580         if (endp == buf || val > 10000000)
581                 return -EINVAL;
582
583         rtnl_lock();
584         ret = t3_config_sched(adap, val, sched);
585         if (!ret)
586                 ret = len;
587         rtnl_unlock();
588         return ret;
589 }
590
591 #define TM_ATTR(name, sched) \
592 static ssize_t show_##name(struct device *d, struct device_attribute *attr, \
593                            char *buf) \
594 { \
595         return tm_attr_show(d, attr, buf, sched); \
596 } \
597 static ssize_t store_##name(struct device *d, struct device_attribute *attr, \
598                             const char *buf, size_t len) \
599 { \
600         return tm_attr_store(d, attr, buf, len, sched); \
601 } \
602 static DEVICE_ATTR(name, S_IRUGO | S_IWUSR, show_##name, store_##name)
603
604 TM_ATTR(sched0, 0);
605 TM_ATTR(sched1, 1);
606 TM_ATTR(sched2, 2);
607 TM_ATTR(sched3, 3);
608 TM_ATTR(sched4, 4);
609 TM_ATTR(sched5, 5);
610 TM_ATTR(sched6, 6);
611 TM_ATTR(sched7, 7);
612
613 static struct attribute *offload_attrs[] = {
614         &dev_attr_sched0.attr,
615         &dev_attr_sched1.attr,
616         &dev_attr_sched2.attr,
617         &dev_attr_sched3.attr,
618         &dev_attr_sched4.attr,
619         &dev_attr_sched5.attr,
620         &dev_attr_sched6.attr,
621         &dev_attr_sched7.attr,
622         NULL
623 };
624
625 static struct attribute_group offload_attr_group = {.attrs = offload_attrs };
626
627 /*
628  * Sends an sk_buff to an offload queue driver
629  * after dealing with any active network taps.
630  */
631 static inline int offload_tx(struct t3cdev *tdev, struct sk_buff *skb)
632 {
633         int ret;
634
635         local_bh_disable();
636         ret = t3_offload_tx(tdev, skb);
637         local_bh_enable();
638         return ret;
639 }
640
641 static int write_smt_entry(struct adapter *adapter, int idx)
642 {
643         struct cpl_smt_write_req *req;
644         struct sk_buff *skb = alloc_skb(sizeof(*req), GFP_KERNEL);
645
646         if (!skb)
647                 return -ENOMEM;
648
649         req = (struct cpl_smt_write_req *)__skb_put(skb, sizeof(*req));
650         req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
651         OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SMT_WRITE_REQ, idx));
652         req->mtu_idx = NMTUS - 1;       /* should be 0 but there's a T3 bug */
653         req->iff = idx;
654         memset(req->src_mac1, 0, sizeof(req->src_mac1));
655         memcpy(req->src_mac0, adapter->port[idx]->dev_addr, ETH_ALEN);
656         skb->priority = 1;
657         offload_tx(&adapter->tdev, skb);
658         return 0;
659 }
660
661 static int init_smt(struct adapter *adapter)
662 {
663         int i;
664
665         for_each_port(adapter, i)
666             write_smt_entry(adapter, i);
667         return 0;
668 }
669
670 static void init_port_mtus(struct adapter *adapter)
671 {
672         unsigned int mtus = adapter->port[0]->mtu;
673
674         if (adapter->port[1])
675                 mtus |= adapter->port[1]->mtu << 16;
676         t3_write_reg(adapter, A_TP_MTU_PORT_TABLE, mtus);
677 }
678
679 static void send_pktsched_cmd(struct adapter *adap, int sched, int qidx, int lo,
680                               int hi, int port)
681 {
682         struct sk_buff *skb;
683         struct mngt_pktsched_wr *req;
684
685         skb = alloc_skb(sizeof(*req), GFP_KERNEL | __GFP_NOFAIL);
686         req = (struct mngt_pktsched_wr *)skb_put(skb, sizeof(*req));
687         req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_MNGT));
688         req->mngt_opcode = FW_MNGTOPCODE_PKTSCHED_SET;
689         req->sched = sched;
690         req->idx = qidx;
691         req->min = lo;
692         req->max = hi;
693         req->binding = port;
694         t3_mgmt_tx(adap, skb);
695 }
696
697 static void bind_qsets(struct adapter *adap)
698 {
699         int i, j;
700
701         for_each_port(adap, i) {
702                 const struct port_info *pi = adap2pinfo(adap, i);
703
704                 for (j = 0; j < pi->nqsets; ++j)
705                         send_pktsched_cmd(adap, 1, pi->first_qset + j, -1,
706                                           -1, i);
707         }
708 }
709
710 /**
711  *      cxgb_up - enable the adapter
712  *      @adapter: adapter being enabled
713  *
714  *      Called when the first port is enabled, this function performs the
715  *      actions necessary to make an adapter operational, such as completing
716  *      the initialization of HW modules, and enabling interrupts.
717  *
718  *      Must be called with the rtnl lock held.
719  */
720 static int cxgb_up(struct adapter *adap)
721 {
722         int err = 0;
723
724         if (!(adap->flags & FULL_INIT_DONE)) {
725                 err = t3_check_fw_version(adap);
726                 if (err)
727                         goto out;
728
729                 err = init_dummy_netdevs(adap);
730                 if (err)
731                         goto out;
732
733                 err = t3_init_hw(adap, 0);
734                 if (err)
735                         goto out;
736
737                 err = setup_sge_qsets(adap);
738                 if (err)
739                         goto out;
740
741                 setup_rss(adap);
742                 adap->flags |= FULL_INIT_DONE;
743         }
744
745         t3_intr_clear(adap);
746
747         if (adap->flags & USING_MSIX) {
748                 name_msix_vecs(adap);
749                 err = request_irq(adap->msix_info[0].vec,
750                                   t3_async_intr_handler, 0,
751                                   adap->msix_info[0].desc, adap);
752                 if (err)
753                         goto irq_err;
754
755                 if (request_msix_data_irqs(adap)) {
756                         free_irq(adap->msix_info[0].vec, adap);
757                         goto irq_err;
758                 }
759         } else if ((err = request_irq(adap->pdev->irq,
760                                       t3_intr_handler(adap,
761                                                       adap->sge.qs[0].rspq.
762                                                       polling),
763                                       (adap->flags & USING_MSI) ?
764                                        0 : IRQF_SHARED,
765                                       adap->name, adap)))
766                 goto irq_err;
767
768         t3_sge_start(adap);
769         t3_intr_enable(adap);
770
771         if ((adap->flags & (USING_MSIX | QUEUES_BOUND)) == USING_MSIX)
772                 bind_qsets(adap);
773         adap->flags |= QUEUES_BOUND;
774
775 out:
776         return err;
777 irq_err:
778         CH_ERR(adap, "request_irq failed, err %d\n", err);
779         goto out;
780 }
781
782 /*
783  * Release resources when all the ports and offloading have been stopped.
784  */
785 static void cxgb_down(struct adapter *adapter)
786 {
787         t3_sge_stop(adapter);
788         spin_lock_irq(&adapter->work_lock);     /* sync with PHY intr task */
789         t3_intr_disable(adapter);
790         spin_unlock_irq(&adapter->work_lock);
791
792         if (adapter->flags & USING_MSIX) {
793                 int i, n = 0;
794
795                 free_irq(adapter->msix_info[0].vec, adapter);
796                 for_each_port(adapter, i)
797                     n += adap2pinfo(adapter, i)->nqsets;
798
799                 for (i = 0; i < n; ++i)
800                         free_irq(adapter->msix_info[i + 1].vec,
801                                  &adapter->sge.qs[i]);
802         } else
803                 free_irq(adapter->pdev->irq, adapter);
804
805         flush_workqueue(cxgb3_wq);      /* wait for external IRQ handler */
806         quiesce_rx(adapter);
807 }
808
809 static void schedule_chk_task(struct adapter *adap)
810 {
811         unsigned int timeo;
812
813         timeo = adap->params.linkpoll_period ?
814             (HZ * adap->params.linkpoll_period) / 10 :
815             adap->params.stats_update_period * HZ;
816         if (timeo)
817                 queue_delayed_work(cxgb3_wq, &adap->adap_check_task, timeo);
818 }
819
820 static int offload_open(struct net_device *dev)
821 {
822         struct adapter *adapter = dev->priv;
823         struct t3cdev *tdev = T3CDEV(dev);
824         int adap_up = adapter->open_device_map & PORT_MASK;
825         int err = 0;
826
827         if (test_and_set_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map))
828                 return 0;
829
830         if (!adap_up && (err = cxgb_up(adapter)) < 0)
831                 return err;
832
833         t3_tp_set_offload_mode(adapter, 1);
834         tdev->lldev = adapter->port[0];
835         err = cxgb3_offload_activate(adapter);
836         if (err)
837                 goto out;
838
839         init_port_mtus(adapter);
840         t3_load_mtus(adapter, adapter->params.mtus, adapter->params.a_wnd,
841                      adapter->params.b_wnd,
842                      adapter->params.rev == 0 ?
843                      adapter->port[0]->mtu : 0xffff);
844         init_smt(adapter);
845
846         /* Never mind if the next step fails */
847         sysfs_create_group(&tdev->lldev->dev.kobj, &offload_attr_group);
848
849         /* Call back all registered clients */
850         cxgb3_add_clients(tdev);
851
852 out:
853         /* restore them in case the offload module has changed them */
854         if (err) {
855                 t3_tp_set_offload_mode(adapter, 0);
856                 clear_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map);
857                 cxgb3_set_dummy_ops(tdev);
858         }
859         return err;
860 }
861
862 static int offload_close(struct t3cdev *tdev)
863 {
864         struct adapter *adapter = tdev2adap(tdev);
865
866         if (!test_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map))
867                 return 0;
868
869         /* Call back all registered clients */
870         cxgb3_remove_clients(tdev);
871
872         sysfs_remove_group(&tdev->lldev->dev.kobj, &offload_attr_group);
873
874         tdev->lldev = NULL;
875         cxgb3_set_dummy_ops(tdev);
876         t3_tp_set_offload_mode(adapter, 0);
877         clear_bit(OFFLOAD_DEVMAP_BIT, &adapter->open_device_map);
878
879         if (!adapter->open_device_map)
880                 cxgb_down(adapter);
881
882         cxgb3_offload_deactivate(adapter);
883         return 0;
884 }
885
886 static int cxgb_open(struct net_device *dev)
887 {
888         int err;
889         struct adapter *adapter = dev->priv;
890         struct port_info *pi = netdev_priv(dev);
891         int other_ports = adapter->open_device_map & PORT_MASK;
892
893         if (!adapter->open_device_map && (err = cxgb_up(adapter)) < 0)
894                 return err;
895
896         set_bit(pi->port_id, &adapter->open_device_map);
897         if (!ofld_disable) {
898                 err = offload_open(dev);
899                 if (err)
900                         printk(KERN_WARNING
901                                "Could not initialize offload capabilities\n");
902         }
903
904         link_start(dev);
905         t3_port_intr_enable(adapter, pi->port_id);
906         netif_start_queue(dev);
907         if (!other_ports)
908                 schedule_chk_task(adapter);
909
910         return 0;
911 }
912
913 static int cxgb_close(struct net_device *dev)
914 {
915         struct adapter *adapter = dev->priv;
916         struct port_info *p = netdev_priv(dev);
917
918         t3_port_intr_disable(adapter, p->port_id);
919         netif_stop_queue(dev);
920         p->phy.ops->power_down(&p->phy, 1);
921         netif_carrier_off(dev);
922         t3_mac_disable(&p->mac, MAC_DIRECTION_TX | MAC_DIRECTION_RX);
923
924         spin_lock(&adapter->work_lock); /* sync with update task */
925         clear_bit(p->port_id, &adapter->open_device_map);
926         spin_unlock(&adapter->work_lock);
927
928         if (!(adapter->open_device_map & PORT_MASK))
929                 cancel_rearming_delayed_workqueue(cxgb3_wq,
930                                                   &adapter->adap_check_task);
931
932         if (!adapter->open_device_map)
933                 cxgb_down(adapter);
934
935         return 0;
936 }
937
938 static struct net_device_stats *cxgb_get_stats(struct net_device *dev)
939 {
940         struct adapter *adapter = dev->priv;
941         struct port_info *p = netdev_priv(dev);
942         struct net_device_stats *ns = &p->netstats;
943         const struct mac_stats *pstats;
944
945         spin_lock(&adapter->stats_lock);
946         pstats = t3_mac_update_stats(&p->mac);
947         spin_unlock(&adapter->stats_lock);
948
949         ns->tx_bytes = pstats->tx_octets;
950         ns->tx_packets = pstats->tx_frames;
951         ns->rx_bytes = pstats->rx_octets;
952         ns->rx_packets = pstats->rx_frames;
953         ns->multicast = pstats->rx_mcast_frames;
954
955         ns->tx_errors = pstats->tx_underrun;
956         ns->rx_errors = pstats->rx_symbol_errs + pstats->rx_fcs_errs +
957             pstats->rx_too_long + pstats->rx_jabber + pstats->rx_short +
958             pstats->rx_fifo_ovfl;
959
960         /* detailed rx_errors */
961         ns->rx_length_errors = pstats->rx_jabber + pstats->rx_too_long;
962         ns->rx_over_errors = 0;
963         ns->rx_crc_errors = pstats->rx_fcs_errs;
964         ns->rx_frame_errors = pstats->rx_symbol_errs;
965         ns->rx_fifo_errors = pstats->rx_fifo_ovfl;
966         ns->rx_missed_errors = pstats->rx_cong_drops;
967
968         /* detailed tx_errors */
969         ns->tx_aborted_errors = 0;
970         ns->tx_carrier_errors = 0;
971         ns->tx_fifo_errors = pstats->tx_underrun;
972         ns->tx_heartbeat_errors = 0;
973         ns->tx_window_errors = 0;
974         return ns;
975 }
976
977 static u32 get_msglevel(struct net_device *dev)
978 {
979         struct adapter *adapter = dev->priv;
980
981         return adapter->msg_enable;
982 }
983
984 static void set_msglevel(struct net_device *dev, u32 val)
985 {
986         struct adapter *adapter = dev->priv;
987
988         adapter->msg_enable = val;
989 }
990
991 static char stats_strings[][ETH_GSTRING_LEN] = {
992         "TxOctetsOK         ",
993         "TxFramesOK         ",
994         "TxMulticastFramesOK",
995         "TxBroadcastFramesOK",
996         "TxPauseFrames      ",
997         "TxUnderrun         ",
998         "TxExtUnderrun      ",
999
1000         "TxFrames64         ",
1001         "TxFrames65To127    ",
1002         "TxFrames128To255   ",
1003         "TxFrames256To511   ",
1004         "TxFrames512To1023  ",
1005         "TxFrames1024To1518 ",
1006         "TxFrames1519ToMax  ",
1007
1008         "RxOctetsOK         ",
1009         "RxFramesOK         ",
1010         "RxMulticastFramesOK",
1011         "RxBroadcastFramesOK",
1012         "RxPauseFrames      ",
1013         "RxFCSErrors        ",
1014         "RxSymbolErrors     ",
1015         "RxShortErrors      ",
1016         "RxJabberErrors     ",
1017         "RxLengthErrors     ",
1018         "RxFIFOoverflow     ",
1019
1020         "RxFrames64         ",
1021         "RxFrames65To127    ",
1022         "RxFrames128To255   ",
1023         "RxFrames256To511   ",
1024         "RxFrames512To1023  ",
1025         "RxFrames1024To1518 ",
1026         "RxFrames1519ToMax  ",
1027
1028         "PhyFIFOErrors      ",
1029         "TSO                ",
1030         "VLANextractions    ",
1031         "VLANinsertions     ",
1032         "TxCsumOffload      ",
1033         "RxCsumGood         ",
1034         "RxDrops            "
1035 };
1036
1037 static int get_stats_count(struct net_device *dev)
1038 {
1039         return ARRAY_SIZE(stats_strings);
1040 }
1041
1042 #define T3_REGMAP_SIZE (3 * 1024)
1043
1044 static int get_regs_len(struct net_device *dev)
1045 {
1046         return T3_REGMAP_SIZE;
1047 }
1048
1049 static int get_eeprom_len(struct net_device *dev)
1050 {
1051         return EEPROMSIZE;
1052 }
1053
1054 static void get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info)
1055 {
1056         u32 fw_vers = 0;
1057         struct adapter *adapter = dev->priv;
1058
1059         t3_get_fw_version(adapter, &fw_vers);
1060
1061         strcpy(info->driver, DRV_NAME);
1062         strcpy(info->version, DRV_VERSION);
1063         strcpy(info->bus_info, pci_name(adapter->pdev));
1064         if (!fw_vers)
1065                 strcpy(info->fw_version, "N/A");
1066         else {
1067                 snprintf(info->fw_version, sizeof(info->fw_version),
1068                          "%s %u.%u.%u",
1069                          G_FW_VERSION_TYPE(fw_vers) ? "T" : "N",
1070                          G_FW_VERSION_MAJOR(fw_vers),
1071                          G_FW_VERSION_MINOR(fw_vers),
1072                          G_FW_VERSION_MICRO(fw_vers));
1073         }
1074 }
1075
1076 static void get_strings(struct net_device *dev, u32 stringset, u8 * data)
1077 {
1078         if (stringset == ETH_SS_STATS)
1079                 memcpy(data, stats_strings, sizeof(stats_strings));
1080 }
1081
1082 static unsigned long collect_sge_port_stats(struct adapter *adapter,
1083                                             struct port_info *p, int idx)
1084 {
1085         int i;
1086         unsigned long tot = 0;
1087
1088         for (i = 0; i < p->nqsets; ++i)
1089                 tot += adapter->sge.qs[i + p->first_qset].port_stats[idx];
1090         return tot;
1091 }
1092
1093 static void get_stats(struct net_device *dev, struct ethtool_stats *stats,
1094                       u64 *data)
1095 {
1096         struct adapter *adapter = dev->priv;
1097         struct port_info *pi = netdev_priv(dev);
1098         const struct mac_stats *s;
1099
1100         spin_lock(&adapter->stats_lock);
1101         s = t3_mac_update_stats(&pi->mac);
1102         spin_unlock(&adapter->stats_lock);
1103
1104         *data++ = s->tx_octets;
1105         *data++ = s->tx_frames;
1106         *data++ = s->tx_mcast_frames;
1107         *data++ = s->tx_bcast_frames;
1108         *data++ = s->tx_pause;
1109         *data++ = s->tx_underrun;
1110         *data++ = s->tx_fifo_urun;
1111
1112         *data++ = s->tx_frames_64;
1113         *data++ = s->tx_frames_65_127;
1114         *data++ = s->tx_frames_128_255;
1115         *data++ = s->tx_frames_256_511;
1116         *data++ = s->tx_frames_512_1023;
1117         *data++ = s->tx_frames_1024_1518;
1118         *data++ = s->tx_frames_1519_max;
1119
1120         *data++ = s->rx_octets;
1121         *data++ = s->rx_frames;
1122         *data++ = s->rx_mcast_frames;
1123         *data++ = s->rx_bcast_frames;
1124         *data++ = s->rx_pause;
1125         *data++ = s->rx_fcs_errs;
1126         *data++ = s->rx_symbol_errs;
1127         *data++ = s->rx_short;
1128         *data++ = s->rx_jabber;
1129         *data++ = s->rx_too_long;
1130         *data++ = s->rx_fifo_ovfl;
1131
1132         *data++ = s->rx_frames_64;
1133         *data++ = s->rx_frames_65_127;
1134         *data++ = s->rx_frames_128_255;
1135         *data++ = s->rx_frames_256_511;
1136         *data++ = s->rx_frames_512_1023;
1137         *data++ = s->rx_frames_1024_1518;
1138         *data++ = s->rx_frames_1519_max;
1139
1140         *data++ = pi->phy.fifo_errors;
1141
1142         *data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_TSO);
1143         *data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_VLANEX);
1144         *data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_VLANINS);
1145         *data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_TX_CSUM);
1146         *data++ = collect_sge_port_stats(adapter, pi, SGE_PSTAT_RX_CSUM_GOOD);
1147         *data++ = s->rx_cong_drops;
1148 }
1149
1150 static inline void reg_block_dump(struct adapter *ap, void *buf,
1151                                   unsigned int start, unsigned int end)
1152 {
1153         u32 *p = buf + start;
1154
1155         for (; start <= end; start += sizeof(u32))
1156                 *p++ = t3_read_reg(ap, start);
1157 }
1158
1159 static void get_regs(struct net_device *dev, struct ethtool_regs *regs,
1160                      void *buf)
1161 {
1162         struct adapter *ap = dev->priv;
1163
1164         /*
1165          * Version scheme:
1166          * bits 0..9: chip version
1167          * bits 10..15: chip revision
1168          * bit 31: set for PCIe cards
1169          */
1170         regs->version = 3 | (ap->params.rev << 10) | (is_pcie(ap) << 31);
1171
1172         /*
1173          * We skip the MAC statistics registers because they are clear-on-read.
1174          * Also reading multi-register stats would need to synchronize with the
1175          * periodic mac stats accumulation.  Hard to justify the complexity.
1176          */
1177         memset(buf, 0, T3_REGMAP_SIZE);
1178         reg_block_dump(ap, buf, 0, A_SG_RSPQ_CREDIT_RETURN);
1179         reg_block_dump(ap, buf, A_SG_HI_DRB_HI_THRSH, A_ULPRX_PBL_ULIMIT);
1180         reg_block_dump(ap, buf, A_ULPTX_CONFIG, A_MPS_INT_CAUSE);
1181         reg_block_dump(ap, buf, A_CPL_SWITCH_CNTRL, A_CPL_MAP_TBL_DATA);
1182         reg_block_dump(ap, buf, A_SMB_GLOBAL_TIME_CFG, A_XGM_SERDES_STAT3);
1183         reg_block_dump(ap, buf, A_XGM_SERDES_STATUS0,
1184                        XGM_REG(A_XGM_SERDES_STAT3, 1));
1185         reg_block_dump(ap, buf, XGM_REG(A_XGM_SERDES_STATUS0, 1),
1186                        XGM_REG(A_XGM_RX_SPI4_SOP_EOP_CNT, 1));
1187 }
1188
1189 static int restart_autoneg(struct net_device *dev)
1190 {
1191         struct port_info *p = netdev_priv(dev);
1192
1193         if (!netif_running(dev))
1194                 return -EAGAIN;
1195         if (p->link_config.autoneg != AUTONEG_ENABLE)
1196                 return -EINVAL;
1197         p->phy.ops->autoneg_restart(&p->phy);
1198         return 0;
1199 }
1200
1201 static int cxgb3_phys_id(struct net_device *dev, u32 data)
1202 {
1203         int i;
1204         struct adapter *adapter = dev->priv;
1205
1206         if (data == 0)
1207                 data = 2;
1208
1209         for (i = 0; i < data * 2; i++) {
1210                 t3_set_reg_field(adapter, A_T3DBG_GPIO_EN, F_GPIO0_OUT_VAL,
1211                                  (i & 1) ? F_GPIO0_OUT_VAL : 0);
1212                 if (msleep_interruptible(500))
1213                         break;
1214         }
1215         t3_set_reg_field(adapter, A_T3DBG_GPIO_EN, F_GPIO0_OUT_VAL,
1216                          F_GPIO0_OUT_VAL);
1217         return 0;
1218 }
1219
1220 static int get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1221 {
1222         struct port_info *p = netdev_priv(dev);
1223
1224         cmd->supported = p->link_config.supported;
1225         cmd->advertising = p->link_config.advertising;
1226
1227         if (netif_carrier_ok(dev)) {
1228                 cmd->speed = p->link_config.speed;
1229                 cmd->duplex = p->link_config.duplex;
1230         } else {
1231                 cmd->speed = -1;
1232                 cmd->duplex = -1;
1233         }
1234
1235         cmd->port = (cmd->supported & SUPPORTED_TP) ? PORT_TP : PORT_FIBRE;
1236         cmd->phy_address = p->phy.addr;
1237         cmd->transceiver = XCVR_EXTERNAL;
1238         cmd->autoneg = p->link_config.autoneg;
1239         cmd->maxtxpkt = 0;
1240         cmd->maxrxpkt = 0;
1241         return 0;
1242 }
1243
1244 static int speed_duplex_to_caps(int speed, int duplex)
1245 {
1246         int cap = 0;
1247
1248         switch (speed) {
1249         case SPEED_10:
1250                 if (duplex == DUPLEX_FULL)
1251                         cap = SUPPORTED_10baseT_Full;
1252                 else
1253                         cap = SUPPORTED_10baseT_Half;
1254                 break;
1255         case SPEED_100:
1256                 if (duplex == DUPLEX_FULL)
1257                         cap = SUPPORTED_100baseT_Full;
1258                 else
1259                         cap = SUPPORTED_100baseT_Half;
1260                 break;
1261         case SPEED_1000:
1262                 if (duplex == DUPLEX_FULL)
1263                         cap = SUPPORTED_1000baseT_Full;
1264                 else
1265                         cap = SUPPORTED_1000baseT_Half;
1266                 break;
1267         case SPEED_10000:
1268                 if (duplex == DUPLEX_FULL)
1269                         cap = SUPPORTED_10000baseT_Full;
1270         }
1271         return cap;
1272 }
1273
1274 #define ADVERTISED_MASK (ADVERTISED_10baseT_Half | ADVERTISED_10baseT_Full | \
1275                       ADVERTISED_100baseT_Half | ADVERTISED_100baseT_Full | \
1276                       ADVERTISED_1000baseT_Half | ADVERTISED_1000baseT_Full | \
1277                       ADVERTISED_10000baseT_Full)
1278
1279 static int set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1280 {
1281         struct port_info *p = netdev_priv(dev);
1282         struct link_config *lc = &p->link_config;
1283
1284         if (!(lc->supported & SUPPORTED_Autoneg))
1285                 return -EOPNOTSUPP;     /* can't change speed/duplex */
1286
1287         if (cmd->autoneg == AUTONEG_DISABLE) {
1288                 int cap = speed_duplex_to_caps(cmd->speed, cmd->duplex);
1289
1290                 if (!(lc->supported & cap) || cmd->speed == SPEED_1000)
1291                         return -EINVAL;
1292                 lc->requested_speed = cmd->speed;
1293                 lc->requested_duplex = cmd->duplex;
1294                 lc->advertising = 0;
1295         } else {
1296                 cmd->advertising &= ADVERTISED_MASK;
1297                 cmd->advertising &= lc->supported;
1298                 if (!cmd->advertising)
1299                         return -EINVAL;
1300                 lc->requested_speed = SPEED_INVALID;
1301                 lc->requested_duplex = DUPLEX_INVALID;
1302                 lc->advertising = cmd->advertising | ADVERTISED_Autoneg;
1303         }
1304         lc->autoneg = cmd->autoneg;
1305         if (netif_running(dev))
1306                 t3_link_start(&p->phy, &p->mac, lc);
1307         return 0;
1308 }
1309
1310 static void get_pauseparam(struct net_device *dev,
1311                            struct ethtool_pauseparam *epause)
1312 {
1313         struct port_info *p = netdev_priv(dev);
1314
1315         epause->autoneg = (p->link_config.requested_fc & PAUSE_AUTONEG) != 0;
1316         epause->rx_pause = (p->link_config.fc & PAUSE_RX) != 0;
1317         epause->tx_pause = (p->link_config.fc & PAUSE_TX) != 0;
1318 }
1319
1320 static int set_pauseparam(struct net_device *dev,
1321                           struct ethtool_pauseparam *epause)
1322 {
1323         struct port_info *p = netdev_priv(dev);
1324         struct link_config *lc = &p->link_config;
1325
1326         if (epause->autoneg == AUTONEG_DISABLE)
1327                 lc->requested_fc = 0;
1328         else if (lc->supported & SUPPORTED_Autoneg)
1329                 lc->requested_fc = PAUSE_AUTONEG;
1330         else
1331                 return -EINVAL;
1332
1333         if (epause->rx_pause)
1334                 lc->requested_fc |= PAUSE_RX;
1335         if (epause->tx_pause)
1336                 lc->requested_fc |= PAUSE_TX;
1337         if (lc->autoneg == AUTONEG_ENABLE) {
1338                 if (netif_running(dev))
1339                         t3_link_start(&p->phy, &p->mac, lc);
1340         } else {
1341                 lc->fc = lc->requested_fc & (PAUSE_RX | PAUSE_TX);
1342                 if (netif_running(dev))
1343                         t3_mac_set_speed_duplex_fc(&p->mac, -1, -1, lc->fc);
1344         }
1345         return 0;
1346 }
1347
1348 static u32 get_rx_csum(struct net_device *dev)
1349 {
1350         struct port_info *p = netdev_priv(dev);
1351
1352         return p->rx_csum_offload;
1353 }
1354
1355 static int set_rx_csum(struct net_device *dev, u32 data)
1356 {
1357         struct port_info *p = netdev_priv(dev);
1358
1359         p->rx_csum_offload = data;
1360         return 0;
1361 }
1362
1363 static void get_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
1364 {
1365         struct adapter *adapter = dev->priv;
1366
1367         e->rx_max_pending = MAX_RX_BUFFERS;
1368         e->rx_mini_max_pending = 0;
1369         e->rx_jumbo_max_pending = MAX_RX_JUMBO_BUFFERS;
1370         e->tx_max_pending = MAX_TXQ_ENTRIES;
1371
1372         e->rx_pending = adapter->params.sge.qset[0].fl_size;
1373         e->rx_mini_pending = adapter->params.sge.qset[0].rspq_size;
1374         e->rx_jumbo_pending = adapter->params.sge.qset[0].jumbo_size;
1375         e->tx_pending = adapter->params.sge.qset[0].txq_size[0];
1376 }
1377
1378 static int set_sge_param(struct net_device *dev, struct ethtool_ringparam *e)
1379 {
1380         int i;
1381         struct adapter *adapter = dev->priv;
1382
1383         if (e->rx_pending > MAX_RX_BUFFERS ||
1384             e->rx_jumbo_pending > MAX_RX_JUMBO_BUFFERS ||
1385             e->tx_pending > MAX_TXQ_ENTRIES ||
1386             e->rx_mini_pending > MAX_RSPQ_ENTRIES ||
1387             e->rx_mini_pending < MIN_RSPQ_ENTRIES ||
1388             e->rx_pending < MIN_FL_ENTRIES ||
1389             e->rx_jumbo_pending < MIN_FL_ENTRIES ||
1390             e->tx_pending < adapter->params.nports * MIN_TXQ_ENTRIES)
1391                 return -EINVAL;
1392
1393         if (adapter->flags & FULL_INIT_DONE)
1394                 return -EBUSY;
1395
1396         for (i = 0; i < SGE_QSETS; ++i) {
1397                 struct qset_params *q = &adapter->params.sge.qset[i];
1398
1399                 q->rspq_size = e->rx_mini_pending;
1400                 q->fl_size = e->rx_pending;
1401                 q->jumbo_size = e->rx_jumbo_pending;
1402                 q->txq_size[0] = e->tx_pending;
1403                 q->txq_size[1] = e->tx_pending;
1404                 q->txq_size[2] = e->tx_pending;
1405         }
1406         return 0;
1407 }
1408
1409 static int set_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
1410 {
1411         struct adapter *adapter = dev->priv;
1412         struct qset_params *qsp = &adapter->params.sge.qset[0];
1413         struct sge_qset *qs = &adapter->sge.qs[0];
1414
1415         if (c->rx_coalesce_usecs * 10 > M_NEWTIMER)
1416                 return -EINVAL;
1417
1418         qsp->coalesce_usecs = c->rx_coalesce_usecs;
1419         t3_update_qset_coalesce(qs, qsp);
1420         return 0;
1421 }
1422
1423 static int get_coalesce(struct net_device *dev, struct ethtool_coalesce *c)
1424 {
1425         struct adapter *adapter = dev->priv;
1426         struct qset_params *q = adapter->params.sge.qset;
1427
1428         c->rx_coalesce_usecs = q->coalesce_usecs;
1429         return 0;
1430 }
1431
1432 static int get_eeprom(struct net_device *dev, struct ethtool_eeprom *e,
1433                       u8 * data)
1434 {
1435         int i, err = 0;
1436         struct adapter *adapter = dev->priv;
1437
1438         u8 *buf = kmalloc(EEPROMSIZE, GFP_KERNEL);
1439         if (!buf)
1440                 return -ENOMEM;
1441
1442         e->magic = EEPROM_MAGIC;
1443         for (i = e->offset & ~3; !err && i < e->offset + e->len; i += 4)
1444                 err = t3_seeprom_read(adapter, i, (u32 *) & buf[i]);
1445
1446         if (!err)
1447                 memcpy(data, buf + e->offset, e->len);
1448         kfree(buf);
1449         return err;
1450 }
1451
1452 static int set_eeprom(struct net_device *dev, struct ethtool_eeprom *eeprom,
1453                       u8 * data)
1454 {
1455         u8 *buf;
1456         int err = 0;
1457         u32 aligned_offset, aligned_len, *p;
1458         struct adapter *adapter = dev->priv;
1459
1460         if (eeprom->magic != EEPROM_MAGIC)
1461                 return -EINVAL;
1462
1463         aligned_offset = eeprom->offset & ~3;
1464         aligned_len = (eeprom->len + (eeprom->offset & 3) + 3) & ~3;
1465
1466         if (aligned_offset != eeprom->offset || aligned_len != eeprom->len) {
1467                 buf = kmalloc(aligned_len, GFP_KERNEL);
1468                 if (!buf)
1469                         return -ENOMEM;
1470                 err = t3_seeprom_read(adapter, aligned_offset, (u32 *) buf);
1471                 if (!err && aligned_len > 4)
1472                         err = t3_seeprom_read(adapter,
1473                                               aligned_offset + aligned_len - 4,
1474                                               (u32 *) & buf[aligned_len - 4]);
1475                 if (err)
1476                         goto out;
1477                 memcpy(buf + (eeprom->offset & 3), data, eeprom->len);
1478         } else
1479                 buf = data;
1480
1481         err = t3_seeprom_wp(adapter, 0);
1482         if (err)
1483                 goto out;
1484
1485         for (p = (u32 *) buf; !err && aligned_len; aligned_len -= 4, p++) {
1486                 err = t3_seeprom_write(adapter, aligned_offset, *p);
1487                 aligned_offset += 4;
1488         }
1489
1490         if (!err)
1491                 err = t3_seeprom_wp(adapter, 1);
1492 out:
1493         if (buf != data)
1494                 kfree(buf);
1495         return err;
1496 }
1497
1498 static void get_wol(struct net_device *dev, struct ethtool_wolinfo *wol)
1499 {
1500         wol->supported = 0;
1501         wol->wolopts = 0;
1502         memset(&wol->sopass, 0, sizeof(wol->sopass));
1503 }
1504
1505 static const struct ethtool_ops cxgb_ethtool_ops = {
1506         .get_settings = get_settings,
1507         .set_settings = set_settings,
1508         .get_drvinfo = get_drvinfo,
1509         .get_msglevel = get_msglevel,
1510         .set_msglevel = set_msglevel,
1511         .get_ringparam = get_sge_param,
1512         .set_ringparam = set_sge_param,
1513         .get_coalesce = get_coalesce,
1514         .set_coalesce = set_coalesce,
1515         .get_eeprom_len = get_eeprom_len,
1516         .get_eeprom = get_eeprom,
1517         .set_eeprom = set_eeprom,
1518         .get_pauseparam = get_pauseparam,
1519         .set_pauseparam = set_pauseparam,
1520         .get_rx_csum = get_rx_csum,
1521         .set_rx_csum = set_rx_csum,
1522         .get_tx_csum = ethtool_op_get_tx_csum,
1523         .set_tx_csum = ethtool_op_set_tx_csum,
1524         .get_sg = ethtool_op_get_sg,
1525         .set_sg = ethtool_op_set_sg,
1526         .get_link = ethtool_op_get_link,
1527         .get_strings = get_strings,
1528         .phys_id = cxgb3_phys_id,
1529         .nway_reset = restart_autoneg,
1530         .get_stats_count = get_stats_count,
1531         .get_ethtool_stats = get_stats,
1532         .get_regs_len = get_regs_len,
1533         .get_regs = get_regs,
1534         .get_wol = get_wol,
1535         .get_tso = ethtool_op_get_tso,
1536         .set_tso = ethtool_op_set_tso,
1537         .get_perm_addr = ethtool_op_get_perm_addr
1538 };
1539
1540 static int in_range(int val, int lo, int hi)
1541 {
1542         return val < 0 || (val <= hi && val >= lo);
1543 }
1544
1545 static int cxgb_extension_ioctl(struct net_device *dev, void __user *useraddr)
1546 {
1547         int ret;
1548         u32 cmd;
1549         struct adapter *adapter = dev->priv;
1550
1551         if (copy_from_user(&cmd, useraddr, sizeof(cmd)))
1552                 return -EFAULT;
1553
1554         switch (cmd) {
1555         case CHELSIO_SET_QSET_PARAMS:{
1556                 int i;
1557                 struct qset_params *q;
1558                 struct ch_qset_params t;
1559
1560                 if (!capable(CAP_NET_ADMIN))
1561                         return -EPERM;
1562                 if (copy_from_user(&t, useraddr, sizeof(t)))
1563                         return -EFAULT;
1564                 if (t.qset_idx >= SGE_QSETS)
1565                         return -EINVAL;
1566                 if (!in_range(t.intr_lat, 0, M_NEWTIMER) ||
1567                         !in_range(t.cong_thres, 0, 255) ||
1568                         !in_range(t.txq_size[0], MIN_TXQ_ENTRIES,
1569                                 MAX_TXQ_ENTRIES) ||
1570                         !in_range(t.txq_size[1], MIN_TXQ_ENTRIES,
1571                                 MAX_TXQ_ENTRIES) ||
1572                         !in_range(t.txq_size[2], MIN_CTRL_TXQ_ENTRIES,
1573                                 MAX_CTRL_TXQ_ENTRIES) ||
1574                         !in_range(t.fl_size[0], MIN_FL_ENTRIES,
1575                                 MAX_RX_BUFFERS)
1576                         || !in_range(t.fl_size[1], MIN_FL_ENTRIES,
1577                                         MAX_RX_JUMBO_BUFFERS)
1578                         || !in_range(t.rspq_size, MIN_RSPQ_ENTRIES,
1579                                         MAX_RSPQ_ENTRIES))
1580                         return -EINVAL;
1581                 if ((adapter->flags & FULL_INIT_DONE) &&
1582                         (t.rspq_size >= 0 || t.fl_size[0] >= 0 ||
1583                         t.fl_size[1] >= 0 || t.txq_size[0] >= 0 ||
1584                         t.txq_size[1] >= 0 || t.txq_size[2] >= 0 ||
1585                         t.polling >= 0 || t.cong_thres >= 0))
1586                         return -EBUSY;
1587
1588                 q = &adapter->params.sge.qset[t.qset_idx];
1589
1590                 if (t.rspq_size >= 0)
1591                         q->rspq_size = t.rspq_size;
1592                 if (t.fl_size[0] >= 0)
1593                         q->fl_size = t.fl_size[0];
1594                 if (t.fl_size[1] >= 0)
1595                         q->jumbo_size = t.fl_size[1];
1596                 if (t.txq_size[0] >= 0)
1597                         q->txq_size[0] = t.txq_size[0];
1598                 if (t.txq_size[1] >= 0)
1599                         q->txq_size[1] = t.txq_size[1];
1600                 if (t.txq_size[2] >= 0)
1601                         q->txq_size[2] = t.txq_size[2];
1602                 if (t.cong_thres >= 0)
1603                         q->cong_thres = t.cong_thres;
1604                 if (t.intr_lat >= 0) {
1605                         struct sge_qset *qs =
1606                                 &adapter->sge.qs[t.qset_idx];
1607
1608                         q->coalesce_usecs = t.intr_lat;
1609                         t3_update_qset_coalesce(qs, q);
1610                 }
1611                 if (t.polling >= 0) {
1612                         if (adapter->flags & USING_MSIX)
1613                                 q->polling = t.polling;
1614                         else {
1615                                 /* No polling with INTx for T3A */
1616                                 if (adapter->params.rev == 0 &&
1617                                         !(adapter->flags & USING_MSI))
1618                                         t.polling = 0;
1619
1620                                 for (i = 0; i < SGE_QSETS; i++) {
1621                                         q = &adapter->params.sge.
1622                                                 qset[i];
1623                                         q->polling = t.polling;
1624                                 }
1625                         }
1626                 }
1627                 break;
1628         }
1629         case CHELSIO_GET_QSET_PARAMS:{
1630                 struct qset_params *q;
1631                 struct ch_qset_params t;
1632
1633                 if (copy_from_user(&t, useraddr, sizeof(t)))
1634                         return -EFAULT;
1635                 if (t.qset_idx >= SGE_QSETS)
1636                         return -EINVAL;
1637
1638                 q = &adapter->params.sge.qset[t.qset_idx];
1639                 t.rspq_size = q->rspq_size;
1640                 t.txq_size[0] = q->txq_size[0];
1641                 t.txq_size[1] = q->txq_size[1];
1642                 t.txq_size[2] = q->txq_size[2];
1643                 t.fl_size[0] = q->fl_size;
1644                 t.fl_size[1] = q->jumbo_size;
1645                 t.polling = q->polling;
1646                 t.intr_lat = q->coalesce_usecs;
1647                 t.cong_thres = q->cong_thres;
1648
1649                 if (copy_to_user(useraddr, &t, sizeof(t)))
1650                         return -EFAULT;
1651                 break;
1652         }
1653         case CHELSIO_SET_QSET_NUM:{
1654                 struct ch_reg edata;
1655                 struct port_info *pi = netdev_priv(dev);
1656                 unsigned int i, first_qset = 0, other_qsets = 0;
1657
1658                 if (!capable(CAP_NET_ADMIN))
1659                         return -EPERM;
1660                 if (adapter->flags & FULL_INIT_DONE)
1661                         return -EBUSY;
1662                 if (copy_from_user(&edata, useraddr, sizeof(edata)))
1663                         return -EFAULT;
1664                 if (edata.val < 1 ||
1665                         (edata.val > 1 && !(adapter->flags & USING_MSIX)))
1666                         return -EINVAL;
1667
1668                 for_each_port(adapter, i)
1669                         if (adapter->port[i] && adapter->port[i] != dev)
1670                                 other_qsets += adap2pinfo(adapter, i)->nqsets;
1671
1672                 if (edata.val + other_qsets > SGE_QSETS)
1673                         return -EINVAL;
1674
1675                 pi->nqsets = edata.val;
1676
1677                 for_each_port(adapter, i)
1678                         if (adapter->port[i]) {
1679                                 pi = adap2pinfo(adapter, i);
1680                                 pi->first_qset = first_qset;
1681                                 first_qset += pi->nqsets;
1682                         }
1683                 break;
1684         }
1685         case CHELSIO_GET_QSET_NUM:{
1686                 struct ch_reg edata;
1687                 struct port_info *pi = netdev_priv(dev);
1688
1689                 edata.cmd = CHELSIO_GET_QSET_NUM;
1690                 edata.val = pi->nqsets;
1691                 if (copy_to_user(useraddr, &edata, sizeof(edata)))
1692                         return -EFAULT;
1693                 break;
1694         }
1695         case CHELSIO_LOAD_FW:{
1696                 u8 *fw_data;
1697                 struct ch_mem_range t;
1698
1699                 if (!capable(CAP_NET_ADMIN))
1700                         return -EPERM;
1701                 if (copy_from_user(&t, useraddr, sizeof(t)))
1702                         return -EFAULT;
1703
1704                 fw_data = kmalloc(t.len, GFP_KERNEL);
1705                 if (!fw_data)
1706                         return -ENOMEM;
1707
1708                 if (copy_from_user
1709                         (fw_data, useraddr + sizeof(t), t.len)) {
1710                         kfree(fw_data);
1711                         return -EFAULT;
1712                 }
1713
1714                 ret = t3_load_fw(adapter, fw_data, t.len);
1715                 kfree(fw_data);
1716                 if (ret)
1717                         return ret;
1718                 break;
1719         }
1720         case CHELSIO_SETMTUTAB:{
1721                 struct ch_mtus m;
1722                 int i;
1723
1724                 if (!is_offload(adapter))
1725                         return -EOPNOTSUPP;
1726                 if (!capable(CAP_NET_ADMIN))
1727                         return -EPERM;
1728                 if (offload_running(adapter))
1729                         return -EBUSY;
1730                 if (copy_from_user(&m, useraddr, sizeof(m)))
1731                         return -EFAULT;
1732                 if (m.nmtus != NMTUS)
1733                         return -EINVAL;
1734                 if (m.mtus[0] < 81)     /* accommodate SACK */
1735                         return -EINVAL;
1736
1737                 /* MTUs must be in ascending order */
1738                 for (i = 1; i < NMTUS; ++i)
1739                         if (m.mtus[i] < m.mtus[i - 1])
1740                                 return -EINVAL;
1741
1742                 memcpy(adapter->params.mtus, m.mtus,
1743                         sizeof(adapter->params.mtus));
1744                 break;
1745         }
1746         case CHELSIO_GET_PM:{
1747                 struct tp_params *p = &adapter->params.tp;
1748                 struct ch_pm m = {.cmd = CHELSIO_GET_PM };
1749
1750                 if (!is_offload(adapter))
1751                         return -EOPNOTSUPP;
1752                 m.tx_pg_sz = p->tx_pg_size;
1753                 m.tx_num_pg = p->tx_num_pgs;
1754                 m.rx_pg_sz = p->rx_pg_size;
1755                 m.rx_num_pg = p->rx_num_pgs;
1756                 m.pm_total = p->pmtx_size + p->chan_rx_size * p->nchan;
1757                 if (copy_to_user(useraddr, &m, sizeof(m)))
1758                         return -EFAULT;
1759                 break;
1760         }
1761         case CHELSIO_SET_PM:{
1762                 struct ch_pm m;
1763                 struct tp_params *p = &adapter->params.tp;
1764
1765                 if (!is_offload(adapter))
1766                         return -EOPNOTSUPP;
1767                 if (!capable(CAP_NET_ADMIN))
1768                         return -EPERM;
1769                 if (adapter->flags & FULL_INIT_DONE)
1770                         return -EBUSY;
1771                 if (copy_from_user(&m, useraddr, sizeof(m)))
1772                         return -EFAULT;
1773                 if (!m.rx_pg_sz || (m.rx_pg_sz & (m.rx_pg_sz - 1)) ||
1774                         !m.tx_pg_sz || (m.tx_pg_sz & (m.tx_pg_sz - 1)))
1775                         return -EINVAL; /* not power of 2 */
1776                 if (!(m.rx_pg_sz & 0x14000))
1777                         return -EINVAL; /* not 16KB or 64KB */
1778                 if (!(m.tx_pg_sz & 0x1554000))
1779                         return -EINVAL;
1780                 if (m.tx_num_pg == -1)
1781                         m.tx_num_pg = p->tx_num_pgs;
1782                 if (m.rx_num_pg == -1)
1783                         m.rx_num_pg = p->rx_num_pgs;
1784                 if (m.tx_num_pg % 24 || m.rx_num_pg % 24)
1785                         return -EINVAL;
1786                 if (m.rx_num_pg * m.rx_pg_sz > p->chan_rx_size ||
1787                         m.tx_num_pg * m.tx_pg_sz > p->chan_tx_size)
1788                         return -EINVAL;
1789                 p->rx_pg_size = m.rx_pg_sz;
1790                 p->tx_pg_size = m.tx_pg_sz;
1791                 p->rx_num_pgs = m.rx_num_pg;
1792                 p->tx_num_pgs = m.tx_num_pg;
1793                 break;
1794         }
1795         case CHELSIO_GET_MEM:{
1796                 struct ch_mem_range t;
1797                 struct mc7 *mem;
1798                 u64 buf[32];
1799
1800                 if (!is_offload(adapter))
1801                         return -EOPNOTSUPP;
1802                 if (!(adapter->flags & FULL_INIT_DONE))
1803                         return -EIO;    /* need the memory controllers */
1804                 if (copy_from_user(&t, useraddr, sizeof(t)))
1805                         return -EFAULT;
1806                 if ((t.addr & 7) || (t.len & 7))
1807                         return -EINVAL;
1808                 if (t.mem_id == MEM_CM)
1809                         mem = &adapter->cm;
1810                 else if (t.mem_id == MEM_PMRX)
1811                         mem = &adapter->pmrx;
1812                 else if (t.mem_id == MEM_PMTX)
1813                         mem = &adapter->pmtx;
1814                 else
1815                         return -EINVAL;
1816
1817                 /*
1818                  * Version scheme:
1819                  * bits 0..9: chip version
1820                  * bits 10..15: chip revision
1821                  */
1822                 t.version = 3 | (adapter->params.rev << 10);
1823                 if (copy_to_user(useraddr, &t, sizeof(t)))
1824                         return -EFAULT;
1825
1826                 /*
1827                  * Read 256 bytes at a time as len can be large and we don't
1828                  * want to use huge intermediate buffers.
1829                  */
1830                 useraddr += sizeof(t);  /* advance to start of buffer */
1831                 while (t.len) {
1832                         unsigned int chunk =
1833                                 min_t(unsigned int, t.len, sizeof(buf));
1834
1835                         ret =
1836                                 t3_mc7_bd_read(mem, t.addr / 8, chunk / 8,
1837                                                 buf);
1838                         if (ret)
1839                                 return ret;
1840                         if (copy_to_user(useraddr, buf, chunk))
1841                                 return -EFAULT;
1842                         useraddr += chunk;
1843                         t.addr += chunk;
1844                         t.len -= chunk;
1845                 }
1846                 break;
1847         }
1848         case CHELSIO_SET_TRACE_FILTER:{
1849                 struct ch_trace t;
1850                 const struct trace_params *tp;
1851
1852                 if (!capable(CAP_NET_ADMIN))
1853                         return -EPERM;
1854                 if (!offload_running(adapter))
1855                         return -EAGAIN;
1856                 if (copy_from_user(&t, useraddr, sizeof(t)))
1857                         return -EFAULT;
1858
1859                 tp = (const struct trace_params *)&t.sip;
1860                 if (t.config_tx)
1861                         t3_config_trace_filter(adapter, tp, 0,
1862                                                 t.invert_match,
1863                                                 t.trace_tx);
1864                 if (t.config_rx)
1865                         t3_config_trace_filter(adapter, tp, 1,
1866                                                 t.invert_match,
1867                                                 t.trace_rx);
1868                 break;
1869         }
1870         default:
1871                 return -EOPNOTSUPP;
1872         }
1873         return 0;
1874 }
1875
1876 static int cxgb_ioctl(struct net_device *dev, struct ifreq *req, int cmd)
1877 {
1878         int ret, mmd;
1879         struct adapter *adapter = dev->priv;
1880         struct port_info *pi = netdev_priv(dev);
1881         struct mii_ioctl_data *data = if_mii(req);
1882
1883         switch (cmd) {
1884         case SIOCGMIIPHY:
1885                 data->phy_id = pi->phy.addr;
1886                 /* FALLTHRU */
1887         case SIOCGMIIREG:{
1888                 u32 val;
1889                 struct cphy *phy = &pi->phy;
1890
1891                 if (!phy->mdio_read)
1892                         return -EOPNOTSUPP;
1893                 if (is_10G(adapter)) {
1894                         mmd = data->phy_id >> 8;
1895                         if (!mmd)
1896                                 mmd = MDIO_DEV_PCS;
1897                         else if (mmd > MDIO_DEV_XGXS)
1898                                 return -EINVAL;
1899
1900                         ret =
1901                                 phy->mdio_read(adapter, data->phy_id & 0x1f,
1902                                                 mmd, data->reg_num, &val);
1903                 } else
1904                         ret =
1905                                 phy->mdio_read(adapter, data->phy_id & 0x1f,
1906                                                 0, data->reg_num & 0x1f,
1907                                                 &val);
1908                 if (!ret)
1909                         data->val_out = val;
1910                 break;
1911         }
1912         case SIOCSMIIREG:{
1913                 struct cphy *phy = &pi->phy;
1914
1915                 if (!capable(CAP_NET_ADMIN))
1916                         return -EPERM;
1917                 if (!phy->mdio_write)
1918                         return -EOPNOTSUPP;
1919                 if (is_10G(adapter)) {
1920                         mmd = data->phy_id >> 8;
1921                         if (!mmd)
1922                                 mmd = MDIO_DEV_PCS;
1923                         else if (mmd > MDIO_DEV_XGXS)
1924                                 return -EINVAL;
1925
1926                         ret =
1927                                 phy->mdio_write(adapter,
1928                                                 data->phy_id & 0x1f, mmd,
1929                                                 data->reg_num,
1930                                                 data->val_in);
1931                 } else
1932                         ret =
1933                                 phy->mdio_write(adapter,
1934                                                 data->phy_id & 0x1f, 0,
1935                                                 data->reg_num & 0x1f,
1936                                                 data->val_in);
1937                 break;
1938         }
1939         case SIOCCHIOCTL:
1940                 return cxgb_extension_ioctl(dev, req->ifr_data);
1941         default:
1942                 return -EOPNOTSUPP;
1943         }
1944         return ret;
1945 }
1946
1947 static int cxgb_change_mtu(struct net_device *dev, int new_mtu)
1948 {
1949         int ret;
1950         struct adapter *adapter = dev->priv;
1951         struct port_info *pi = netdev_priv(dev);
1952
1953         if (new_mtu < 81)       /* accommodate SACK */
1954                 return -EINVAL;
1955         if ((ret = t3_mac_set_mtu(&pi->mac, new_mtu)))
1956                 return ret;
1957         dev->mtu = new_mtu;
1958         init_port_mtus(adapter);
1959         if (adapter->params.rev == 0 && offload_running(adapter))
1960                 t3_load_mtus(adapter, adapter->params.mtus,
1961                              adapter->params.a_wnd, adapter->params.b_wnd,
1962                              adapter->port[0]->mtu);
1963         return 0;
1964 }
1965
1966 static int cxgb_set_mac_addr(struct net_device *dev, void *p)
1967 {
1968         struct adapter *adapter = dev->priv;
1969         struct port_info *pi = netdev_priv(dev);
1970         struct sockaddr *addr = p;
1971
1972         if (!is_valid_ether_addr(addr->sa_data))
1973                 return -EINVAL;
1974
1975         memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
1976         t3_mac_set_address(&pi->mac, 0, dev->dev_addr);
1977         if (offload_running(adapter))
1978                 write_smt_entry(adapter, pi->port_id);
1979         return 0;
1980 }
1981
1982 /**
1983  * t3_synchronize_rx - wait for current Rx processing on a port to complete
1984  * @adap: the adapter
1985  * @p: the port
1986  *
1987  * Ensures that current Rx processing on any of the queues associated with
1988  * the given port completes before returning.  We do this by acquiring and
1989  * releasing the locks of the response queues associated with the port.
1990  */
1991 static void t3_synchronize_rx(struct adapter *adap, const struct port_info *p)
1992 {
1993         int i;
1994
1995         for (i = 0; i < p->nqsets; i++) {
1996                 struct sge_rspq *q = &adap->sge.qs[i + p->first_qset].rspq;
1997
1998                 spin_lock_irq(&q->lock);
1999                 spin_unlock_irq(&q->lock);
2000         }
2001 }
2002
2003 static void vlan_rx_register(struct net_device *dev, struct vlan_group *grp)
2004 {
2005         struct adapter *adapter = dev->priv;
2006         struct port_info *pi = netdev_priv(dev);
2007
2008         pi->vlan_grp = grp;
2009         if (adapter->params.rev > 0)
2010                 t3_set_vlan_accel(adapter, 1 << pi->port_id, grp != NULL);
2011         else {
2012                 /* single control for all ports */
2013                 unsigned int i, have_vlans = 0;
2014                 for_each_port(adapter, i)
2015                     have_vlans |= adap2pinfo(adapter, i)->vlan_grp != NULL;
2016
2017                 t3_set_vlan_accel(adapter, 1, have_vlans);
2018         }
2019         t3_synchronize_rx(adapter, pi);
2020 }
2021
2022 static void vlan_rx_kill_vid(struct net_device *dev, unsigned short vid)
2023 {
2024         /* nothing */
2025 }
2026
2027 #ifdef CONFIG_NET_POLL_CONTROLLER
2028 static void cxgb_netpoll(struct net_device *dev)
2029 {
2030         struct adapter *adapter = dev->priv;
2031         struct sge_qset *qs = dev2qset(dev);
2032
2033         t3_intr_handler(adapter, qs->rspq.polling) (adapter->pdev->irq,
2034                                                     adapter);
2035 }
2036 #endif
2037
2038 /*
2039  * Periodic accumulation of MAC statistics.
2040  */
2041 static void mac_stats_update(struct adapter *adapter)
2042 {
2043         int i;
2044
2045         for_each_port(adapter, i) {
2046                 struct net_device *dev = adapter->port[i];
2047                 struct port_info *p = netdev_priv(dev);
2048
2049                 if (netif_running(dev)) {
2050                         spin_lock(&adapter->stats_lock);
2051                         t3_mac_update_stats(&p->mac);
2052                         spin_unlock(&adapter->stats_lock);
2053                 }
2054         }
2055 }
2056
2057 static void check_link_status(struct adapter *adapter)
2058 {
2059         int i;
2060
2061         for_each_port(adapter, i) {
2062                 struct net_device *dev = adapter->port[i];
2063                 struct port_info *p = netdev_priv(dev);
2064
2065                 if (!(p->port_type->caps & SUPPORTED_IRQ) && netif_running(dev))
2066                         t3_link_changed(adapter, i);
2067         }
2068 }
2069
2070 static void t3_adap_check_task(struct work_struct *work)
2071 {
2072         struct adapter *adapter = container_of(work, struct adapter,
2073                                                adap_check_task.work);
2074         const struct adapter_params *p = &adapter->params;
2075
2076         adapter->check_task_cnt++;
2077
2078         /* Check link status for PHYs without interrupts */
2079         if (p->linkpoll_period)
2080                 check_link_status(adapter);
2081
2082         /* Accumulate MAC stats if needed */
2083         if (!p->linkpoll_period ||
2084             (adapter->check_task_cnt * p->linkpoll_period) / 10 >=
2085             p->stats_update_period) {
2086                 mac_stats_update(adapter);
2087                 adapter->check_task_cnt = 0;
2088         }
2089
2090         /* Schedule the next check update if any port is active. */
2091         spin_lock(&adapter->work_lock);
2092         if (adapter->open_device_map & PORT_MASK)
2093                 schedule_chk_task(adapter);
2094         spin_unlock(&adapter->work_lock);
2095 }
2096
2097 /*
2098  * Processes external (PHY) interrupts in process context.
2099  */
2100 static void ext_intr_task(struct work_struct *work)
2101 {
2102         struct adapter *adapter = container_of(work, struct adapter,
2103                                                ext_intr_handler_task);
2104
2105         t3_phy_intr_handler(adapter);
2106
2107         /* Now reenable external interrupts */
2108         spin_lock_irq(&adapter->work_lock);
2109         if (adapter->slow_intr_mask) {
2110                 adapter->slow_intr_mask |= F_T3DBG;
2111                 t3_write_reg(adapter, A_PL_INT_CAUSE0, F_T3DBG);
2112                 t3_write_reg(adapter, A_PL_INT_ENABLE0,
2113                              adapter->slow_intr_mask);
2114         }
2115         spin_unlock_irq(&adapter->work_lock);
2116 }
2117
2118 /*
2119  * Interrupt-context handler for external (PHY) interrupts.
2120  */
2121 void t3_os_ext_intr_handler(struct adapter *adapter)
2122 {
2123         /*
2124          * Schedule a task to handle external interrupts as they may be slow
2125          * and we use a mutex to protect MDIO registers.  We disable PHY
2126          * interrupts in the meantime and let the task reenable them when
2127          * it's done.
2128          */
2129         spin_lock(&adapter->work_lock);
2130         if (adapter->slow_intr_mask) {
2131                 adapter->slow_intr_mask &= ~F_T3DBG;
2132                 t3_write_reg(adapter, A_PL_INT_ENABLE0,
2133                              adapter->slow_intr_mask);
2134                 queue_work(cxgb3_wq, &adapter->ext_intr_handler_task);
2135         }
2136         spin_unlock(&adapter->work_lock);
2137 }
2138
2139 void t3_fatal_err(struct adapter *adapter)
2140 {
2141         unsigned int fw_status[4];
2142
2143         if (adapter->flags & FULL_INIT_DONE) {
2144                 t3_sge_stop(adapter);
2145                 t3_intr_disable(adapter);
2146         }
2147         CH_ALERT(adapter, "encountered fatal error, operation suspended\n");
2148         if (!t3_cim_ctl_blk_read(adapter, 0xa0, 4, fw_status))
2149                 CH_ALERT(adapter, "FW status: 0x%x, 0x%x, 0x%x, 0x%x\n",
2150                          fw_status[0], fw_status[1],
2151                          fw_status[2], fw_status[3]);
2152
2153 }
2154
2155 static int __devinit cxgb_enable_msix(struct adapter *adap)
2156 {
2157         struct msix_entry entries[SGE_QSETS + 1];
2158         int i, err;
2159
2160         for (i = 0; i < ARRAY_SIZE(entries); ++i)
2161                 entries[i].entry = i;
2162
2163         err = pci_enable_msix(adap->pdev, entries, ARRAY_SIZE(entries));
2164         if (!err) {
2165                 for (i = 0; i < ARRAY_SIZE(entries); ++i)
2166                         adap->msix_info[i].vec = entries[i].vector;
2167         } else if (err > 0)
2168                 dev_info(&adap->pdev->dev,
2169                        "only %d MSI-X vectors left, not using MSI-X\n", err);
2170         return err;
2171 }
2172
2173 static void __devinit print_port_info(struct adapter *adap,
2174                                       const struct adapter_info *ai)
2175 {
2176         static const char *pci_variant[] = {
2177                 "PCI", "PCI-X", "PCI-X ECC", "PCI-X 266", "PCI Express"
2178         };
2179
2180         int i;
2181         char buf[80];
2182
2183         if (is_pcie(adap))
2184                 snprintf(buf, sizeof(buf), "%s x%d",
2185                          pci_variant[adap->params.pci.variant],
2186                          adap->params.pci.width);
2187         else
2188                 snprintf(buf, sizeof(buf), "%s %dMHz/%d-bit",
2189                          pci_variant[adap->params.pci.variant],
2190                          adap->params.pci.speed, adap->params.pci.width);
2191
2192         for_each_port(adap, i) {
2193                 struct net_device *dev = adap->port[i];
2194                 const struct port_info *pi = netdev_priv(dev);
2195
2196                 if (!test_bit(i, &adap->registered_device_map))
2197                         continue;
2198                 printk(KERN_INFO "%s: %s %s RNIC (rev %d) %s%s\n",
2199                        dev->name, ai->desc, pi->port_type->desc,
2200                        adap->params.rev, buf,
2201                        (adap->flags & USING_MSIX) ? " MSI-X" :
2202                        (adap->flags & USING_MSI) ? " MSI" : "");
2203                 if (adap->name == dev->name && adap->params.vpd.mclk)
2204                         printk(KERN_INFO "%s: %uMB CM, %uMB PMTX, %uMB PMRX\n",
2205                                adap->name, t3_mc7_size(&adap->cm) >> 20,
2206                                t3_mc7_size(&adap->pmtx) >> 20,
2207                                t3_mc7_size(&adap->pmrx) >> 20);
2208         }
2209 }
2210
2211 static int __devinit init_one(struct pci_dev *pdev,
2212                               const struct pci_device_id *ent)
2213 {
2214         static int version_printed;
2215
2216         int i, err, pci_using_dac = 0;
2217         unsigned long mmio_start, mmio_len;
2218         const struct adapter_info *ai;
2219         struct adapter *adapter = NULL;
2220         struct port_info *pi;
2221
2222         if (!version_printed) {
2223                 printk(KERN_INFO "%s - version %s\n", DRV_DESC, DRV_VERSION);
2224                 ++version_printed;
2225         }
2226
2227         if (!cxgb3_wq) {
2228                 cxgb3_wq = create_singlethread_workqueue(DRV_NAME);
2229                 if (!cxgb3_wq) {
2230                         printk(KERN_ERR DRV_NAME
2231                                ": cannot initialize work queue\n");
2232                         return -ENOMEM;
2233                 }
2234         }
2235
2236         err = pci_request_regions(pdev, DRV_NAME);
2237         if (err) {
2238                 /* Just info, some other driver may have claimed the device. */
2239                 dev_info(&pdev->dev, "cannot obtain PCI resources\n");
2240                 return err;
2241         }
2242
2243         err = pci_enable_device(pdev);
2244         if (err) {
2245                 dev_err(&pdev->dev, "cannot enable PCI device\n");
2246                 goto out_release_regions;
2247         }
2248
2249         if (!pci_set_dma_mask(pdev, DMA_64BIT_MASK)) {
2250                 pci_using_dac = 1;
2251                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
2252                 if (err) {
2253                         dev_err(&pdev->dev, "unable to obtain 64-bit DMA for "
2254                                "coherent allocations\n");
2255                         goto out_disable_device;
2256                 }
2257         } else if ((err = pci_set_dma_mask(pdev, DMA_32BIT_MASK)) != 0) {
2258                 dev_err(&pdev->dev, "no usable DMA configuration\n");
2259                 goto out_disable_device;
2260         }
2261
2262         pci_set_master(pdev);
2263
2264         mmio_start = pci_resource_start(pdev, 0);
2265         mmio_len = pci_resource_len(pdev, 0);
2266         ai = t3_get_adapter_info(ent->driver_data);
2267
2268         adapter = kzalloc(sizeof(*adapter), GFP_KERNEL);
2269         if (!adapter) {
2270                 err = -ENOMEM;
2271                 goto out_disable_device;
2272         }
2273
2274         adapter->regs = ioremap_nocache(mmio_start, mmio_len);
2275         if (!adapter->regs) {
2276                 dev_err(&pdev->dev, "cannot map device registers\n");
2277                 err = -ENOMEM;
2278                 goto out_free_adapter;
2279         }
2280
2281         adapter->pdev = pdev;
2282         adapter->name = pci_name(pdev);
2283         adapter->msg_enable = dflt_msg_enable;
2284         adapter->mmio_len = mmio_len;
2285
2286         mutex_init(&adapter->mdio_lock);
2287         spin_lock_init(&adapter->work_lock);
2288         spin_lock_init(&adapter->stats_lock);
2289
2290         INIT_LIST_HEAD(&adapter->adapter_list);
2291         INIT_WORK(&adapter->ext_intr_handler_task, ext_intr_task);
2292         INIT_DELAYED_WORK(&adapter->adap_check_task, t3_adap_check_task);
2293
2294         for (i = 0; i < ai->nports; ++i) {
2295                 struct net_device *netdev;
2296
2297                 netdev = alloc_etherdev(sizeof(struct port_info));
2298                 if (!netdev) {
2299                         err = -ENOMEM;
2300                         goto out_free_dev;
2301                 }
2302
2303                 SET_MODULE_OWNER(netdev);
2304                 SET_NETDEV_DEV(netdev, &pdev->dev);
2305
2306                 adapter->port[i] = netdev;
2307                 pi = netdev_priv(netdev);
2308                 pi->rx_csum_offload = 1;
2309                 pi->nqsets = 1;
2310                 pi->first_qset = i;
2311                 pi->activity = 0;
2312                 pi->port_id = i;
2313                 netif_carrier_off(netdev);
2314                 netdev->irq = pdev->irq;
2315                 netdev->mem_start = mmio_start;
2316                 netdev->mem_end = mmio_start + mmio_len - 1;
2317                 netdev->priv = adapter;
2318                 netdev->features |= NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_TSO;
2319                 netdev->features |= NETIF_F_LLTX;
2320                 if (pci_using_dac)
2321                         netdev->features |= NETIF_F_HIGHDMA;
2322
2323                 netdev->features |= NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX;
2324                 netdev->vlan_rx_register = vlan_rx_register;
2325                 netdev->vlan_rx_kill_vid = vlan_rx_kill_vid;
2326
2327                 netdev->open = cxgb_open;
2328                 netdev->stop = cxgb_close;
2329                 netdev->hard_start_xmit = t3_eth_xmit;
2330                 netdev->get_stats = cxgb_get_stats;
2331                 netdev->set_multicast_list = cxgb_set_rxmode;
2332                 netdev->do_ioctl = cxgb_ioctl;
2333                 netdev->change_mtu = cxgb_change_mtu;
2334                 netdev->set_mac_address = cxgb_set_mac_addr;
2335 #ifdef CONFIG_NET_POLL_CONTROLLER
2336                 netdev->poll_controller = cxgb_netpoll;
2337 #endif
2338                 netdev->weight = 64;
2339
2340                 SET_ETHTOOL_OPS(netdev, &cxgb_ethtool_ops);
2341         }
2342
2343         pci_set_drvdata(pdev, adapter->port[0]);
2344         if (t3_prep_adapter(adapter, ai, 1) < 0) {
2345                 err = -ENODEV;
2346                 goto out_free_dev;
2347         }
2348
2349         /*
2350          * The card is now ready to go.  If any errors occur during device
2351          * registration we do not fail the whole card but rather proceed only
2352          * with the ports we manage to register successfully.  However we must
2353          * register at least one net device.
2354          */
2355         for_each_port(adapter, i) {
2356                 err = register_netdev(adapter->port[i]);
2357                 if (err)
2358                         dev_warn(&pdev->dev,
2359                                  "cannot register net device %s, skipping\n",
2360                                  adapter->port[i]->name);
2361                 else {
2362                         /*
2363                          * Change the name we use for messages to the name of
2364                          * the first successfully registered interface.
2365                          */
2366                         if (!adapter->registered_device_map)
2367                                 adapter->name = adapter->port[i]->name;
2368
2369                         __set_bit(i, &adapter->registered_device_map);
2370                 }
2371         }
2372         if (!adapter->registered_device_map) {
2373                 dev_err(&pdev->dev, "could not register any net devices\n");
2374                 goto out_free_dev;
2375         }
2376
2377         /* Driver's ready. Reflect it on LEDs */
2378         t3_led_ready(adapter);
2379
2380         if (is_offload(adapter)) {
2381                 __set_bit(OFFLOAD_DEVMAP_BIT, &adapter->registered_device_map);
2382                 cxgb3_adapter_ofld(adapter);
2383         }
2384
2385         /* See what interrupts we'll be using */
2386         if (msi > 1 && cxgb_enable_msix(adapter) == 0)
2387                 adapter->flags |= USING_MSIX;
2388         else if (msi > 0 && pci_enable_msi(pdev) == 0)
2389                 adapter->flags |= USING_MSI;
2390
2391         err = sysfs_create_group(&adapter->port[0]->dev.kobj,
2392                                  &cxgb3_attr_group);
2393
2394         print_port_info(adapter, ai);
2395         return 0;
2396
2397 out_free_dev:
2398         iounmap(adapter->regs);
2399         for (i = ai->nports - 1; i >= 0; --i)
2400                 if (adapter->port[i])
2401                         free_netdev(adapter->port[i]);
2402
2403 out_free_adapter:
2404         kfree(adapter);
2405
2406 out_disable_device:
2407         pci_disable_device(pdev);
2408 out_release_regions:
2409         pci_release_regions(pdev);
2410         pci_set_drvdata(pdev, NULL);
2411         return err;
2412 }
2413
2414 static void __devexit remove_one(struct pci_dev *pdev)
2415 {
2416         struct net_device *dev = pci_get_drvdata(pdev);
2417
2418         if (dev) {
2419                 int i;
2420                 struct adapter *adapter = dev->priv;
2421
2422                 t3_sge_stop(adapter);
2423                 sysfs_remove_group(&adapter->port[0]->dev.kobj,
2424                                    &cxgb3_attr_group);
2425
2426                 for_each_port(adapter, i)
2427                     if (test_bit(i, &adapter->registered_device_map))
2428                         unregister_netdev(adapter->port[i]);
2429
2430                 if (is_offload(adapter)) {
2431                         cxgb3_adapter_unofld(adapter);
2432                         if (test_bit(OFFLOAD_DEVMAP_BIT,
2433                                      &adapter->open_device_map))
2434                                 offload_close(&adapter->tdev);
2435                 }
2436
2437                 t3_free_sge_resources(adapter);
2438                 cxgb_disable_msi(adapter);
2439
2440                 for (i = 0; i < ARRAY_SIZE(adapter->dummy_netdev); i++)
2441                         if (adapter->dummy_netdev[i]) {
2442                                 free_netdev(adapter->dummy_netdev[i]);
2443                                 adapter->dummy_netdev[i] = NULL;
2444                         }
2445
2446                 for_each_port(adapter, i)
2447                         if (adapter->port[i])
2448                                 free_netdev(adapter->port[i]);
2449
2450                 iounmap(adapter->regs);
2451                 kfree(adapter);
2452                 pci_release_regions(pdev);
2453                 pci_disable_device(pdev);
2454                 pci_set_drvdata(pdev, NULL);
2455         }
2456 }
2457
2458 static struct pci_driver driver = {
2459         .name = DRV_NAME,
2460         .id_table = cxgb3_pci_tbl,
2461         .probe = init_one,
2462         .remove = __devexit_p(remove_one),
2463 };
2464
2465 static int __init cxgb3_init_module(void)
2466 {
2467         int ret;
2468
2469         cxgb3_offload_init();
2470
2471         ret = pci_register_driver(&driver);
2472         return ret;
2473 }
2474
2475 static void __exit cxgb3_cleanup_module(void)
2476 {
2477         pci_unregister_driver(&driver);
2478         if (cxgb3_wq)
2479                 destroy_workqueue(cxgb3_wq);
2480 }
2481
2482 module_init(cxgb3_init_module);
2483 module_exit(cxgb3_cleanup_module);