]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - drivers/net/ethernet/ibm/ibmvnic.c
projects / mirror_ubuntu-jammy-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
ibmvnic: Allow extra failures before disabling
[mirror_ubuntu-jammy-kernel.git] / drivers / net / ethernet / ibm / ibmvnic.c
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /**************************************************************************/
3 /* */
4 /* IBM System i and System p Virtual NIC Device Driver */
5 /* Copyright (C) 2014 IBM Corp. */
6 /* Santiago Leon (santi_leon@yahoo.com) */
7 /* Thomas Falcon (tlfalcon@linux.vnet.ibm.com) */
8 /* John Allen (jallen@linux.vnet.ibm.com) */
9 /* */
10 /* */
11 /* This module contains the implementation of a virtual ethernet device */
12 /* for use with IBM i/p Series LPAR Linux. It utilizes the logical LAN */
13 /* option of the RS/6000 Platform Architecture to interface with virtual */
14 /* ethernet NICs that are presented to the partition by the hypervisor. */
15 /* */
16 /* Messages are passed between the VNIC driver and the VNIC server using */
17 /* Command/Response Queues (CRQs) and sub CRQs (sCRQs). CRQs are used to */
18 /* issue and receive commands that initiate communication with the server */
19 /* on driver initialization. Sub CRQs (sCRQs) are similar to CRQs, but */
20 /* are used by the driver to notify the server that a packet is */
21 /* ready for transmission or that a buffer has been added to receive a */
22 /* packet. Subsequently, sCRQs are used by the server to notify the */
23 /* driver that a packet transmission has been completed or that a packet */
24 /* has been received and placed in a waiting buffer. */
25 /* */
26 /* In lieu of a more conventional "on-the-fly" DMA mapping strategy in */
27 /* which skbs are DMA mapped and immediately unmapped when the transmit */
28 /* or receive has been completed, the VNIC driver is required to use */
29 /* "long term mapping". This entails that large, continuous DMA mapped */
30 /* buffers are allocated on driver initialization and these buffers are */
31 /* then continuously reused to pass skbs to and from the VNIC server. */
32 /* */
33 /**************************************************************************/
34
35 #include <linux/module.h>
36 #include <linux/moduleparam.h>
37 #include <linux/types.h>
38 #include <linux/errno.h>
39 #include <linux/completion.h>
40 #include <linux/ioport.h>
41 #include <linux/dma-mapping.h>
42 #include <linux/kernel.h>
43 #include <linux/netdevice.h>
44 #include <linux/etherdevice.h>
45 #include <linux/skbuff.h>
46 #include <linux/init.h>
47 #include <linux/delay.h>
48 #include <linux/mm.h>
49 #include <linux/ethtool.h>
50 #include <linux/proc_fs.h>
51 #include <linux/if_arp.h>
52 #include <linux/in.h>
53 #include <linux/ip.h>
54 #include <linux/ipv6.h>
55 #include <linux/irq.h>
56 #include <linux/kthread.h>
57 #include <linux/seq_file.h>
58 #include <linux/interrupt.h>
59 #include <net/net_namespace.h>
60 #include <asm/hvcall.h>
61 #include <linux/atomic.h>
62 #include <asm/vio.h>
63 #include <asm/iommu.h>
64 #include <linux/uaccess.h>
65 #include <asm/firmware.h>
66 #include <linux/workqueue.h>
67 #include <linux/if_vlan.h>
68 #include <linux/utsname.h>
69
70 #include "ibmvnic.h"
71
72 static const char ibmvnic_driver_name[] = "ibmvnic";
73 static const char ibmvnic_driver_string[] = "IBM System i/p Virtual NIC Driver";
74
75 MODULE_AUTHOR("Santiago Leon");
76 MODULE_DESCRIPTION("IBM System i/p Virtual NIC Driver");
77 MODULE_LICENSE("GPL");
78 MODULE_VERSION(IBMVNIC_DRIVER_VERSION);
79
80 static int ibmvnic_version = IBMVNIC_INITIAL_VERSION;
81 static void release_sub_crqs(struct ibmvnic_adapter *, bool);
82 static int ibmvnic_reset_crq(struct ibmvnic_adapter *);
83 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *);
84 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *);
85 static int ibmvnic_send_crq(struct ibmvnic_adapter *, union ibmvnic_crq *);
86 static int send_subcrq_indirect(struct ibmvnic_adapter *, u64, u64, u64);
87 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance);
88 static int enable_scrq_irq(struct ibmvnic_adapter *,
89 struct ibmvnic_sub_crq_queue *);
90 static int disable_scrq_irq(struct ibmvnic_adapter *,
91 struct ibmvnic_sub_crq_queue *);
92 static int pending_scrq(struct ibmvnic_adapter *,
93 struct ibmvnic_sub_crq_queue *);
94 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *,
95 struct ibmvnic_sub_crq_queue *);
96 static int ibmvnic_poll(struct napi_struct *napi, int data);
97 static void send_query_map(struct ibmvnic_adapter *adapter);
98 static int send_request_map(struct ibmvnic_adapter *, dma_addr_t, u32, u8);
99 static int send_request_unmap(struct ibmvnic_adapter *, u8);
100 static int send_login(struct ibmvnic_adapter *adapter);
101 static void send_query_cap(struct ibmvnic_adapter *adapter);
102 static int init_sub_crqs(struct ibmvnic_adapter *);
103 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter);
104 static int ibmvnic_reset_init(struct ibmvnic_adapter *, bool reset);
105 static void release_crq_queue(struct ibmvnic_adapter *);
106 static int __ibmvnic_set_mac(struct net_device *, u8 *);
107 static int init_crq_queue(struct ibmvnic_adapter *adapter);
108 static int send_query_phys_parms(struct ibmvnic_adapter *adapter);
109 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
110 struct ibmvnic_sub_crq_queue *tx_scrq);
111
112 struct ibmvnic_stat {
113 char name[ETH_GSTRING_LEN];
114 int offset;
115 };
116
117 #define IBMVNIC_STAT_OFF(stat) (offsetof(struct ibmvnic_adapter, stats) + \
118 offsetof(struct ibmvnic_statistics, stat))
119 #define IBMVNIC_GET_STAT(a, off) (*((u64 *)(((unsigned long)(a)) + (off))))
120
121 static const struct ibmvnic_stat ibmvnic_stats[] = {
122 {"rx_packets", IBMVNIC_STAT_OFF(rx_packets)},
123 {"rx_bytes", IBMVNIC_STAT_OFF(rx_bytes)},
124 {"tx_packets", IBMVNIC_STAT_OFF(tx_packets)},
125 {"tx_bytes", IBMVNIC_STAT_OFF(tx_bytes)},
126 {"ucast_tx_packets", IBMVNIC_STAT_OFF(ucast_tx_packets)},
127 {"ucast_rx_packets", IBMVNIC_STAT_OFF(ucast_rx_packets)},
128 {"mcast_tx_packets", IBMVNIC_STAT_OFF(mcast_tx_packets)},
129 {"mcast_rx_packets", IBMVNIC_STAT_OFF(mcast_rx_packets)},
130 {"bcast_tx_packets", IBMVNIC_STAT_OFF(bcast_tx_packets)},
131 {"bcast_rx_packets", IBMVNIC_STAT_OFF(bcast_rx_packets)},
132 {"align_errors", IBMVNIC_STAT_OFF(align_errors)},
133 {"fcs_errors", IBMVNIC_STAT_OFF(fcs_errors)},
134 {"single_collision_frames", IBMVNIC_STAT_OFF(single_collision_frames)},
135 {"multi_collision_frames", IBMVNIC_STAT_OFF(multi_collision_frames)},
136 {"sqe_test_errors", IBMVNIC_STAT_OFF(sqe_test_errors)},
137 {"deferred_tx", IBMVNIC_STAT_OFF(deferred_tx)},
138 {"late_collisions", IBMVNIC_STAT_OFF(late_collisions)},
139 {"excess_collisions", IBMVNIC_STAT_OFF(excess_collisions)},
140 {"internal_mac_tx_errors", IBMVNIC_STAT_OFF(internal_mac_tx_errors)},
141 {"carrier_sense", IBMVNIC_STAT_OFF(carrier_sense)},
142 {"too_long_frames", IBMVNIC_STAT_OFF(too_long_frames)},
143 {"internal_mac_rx_errors", IBMVNIC_STAT_OFF(internal_mac_rx_errors)},
144 };
145
146 static int send_crq_init_complete(struct ibmvnic_adapter *adapter)
147 {
148 union ibmvnic_crq crq;
149
150 memset(&crq, 0, sizeof(crq));
151 crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
152 crq.generic.cmd = IBMVNIC_CRQ_INIT_COMPLETE;
153
154 return ibmvnic_send_crq(adapter, &crq);
155 }
156
157 static int send_version_xchg(struct ibmvnic_adapter *adapter)
158 {
159 union ibmvnic_crq crq;
160
161 memset(&crq, 0, sizeof(crq));
162 crq.version_exchange.first = IBMVNIC_CRQ_CMD;
163 crq.version_exchange.cmd = VERSION_EXCHANGE;
164 crq.version_exchange.version = cpu_to_be16(ibmvnic_version);
165
166 return ibmvnic_send_crq(adapter, &crq);
167 }
168
169 static long h_reg_sub_crq(unsigned long unit_address, unsigned long token,
170 unsigned long length, unsigned long *number,
171 unsigned long *irq)
172 {
173 unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
174 long rc;
175
176 rc = plpar_hcall(H_REG_SUB_CRQ, retbuf, unit_address, token, length);
177 *number = retbuf[0];
178 *irq = retbuf[1];
179
180 return rc;
181 }
182
183 /**
184 * ibmvnic_wait_for_completion - Check device state and wait for completion
185 * @adapter: private device data
186 * @comp_done: completion structure to wait for
187 * @timeout: time to wait in milliseconds
188 *
189 * Wait for a completion signal or until the timeout limit is reached
190 * while checking that the device is still active.
191 */
192 static int ibmvnic_wait_for_completion(struct ibmvnic_adapter *adapter,
193 struct completion *comp_done,
194 unsigned long timeout)
195 {
196 struct net_device *netdev;
197 unsigned long div_timeout;
198 u8 retry;
199
200 netdev = adapter->netdev;
201 retry = 5;
202 div_timeout = msecs_to_jiffies(timeout / retry);
203 while (true) {
204 if (!adapter->crq.active) {
205 netdev_err(netdev, "Device down!\n");
206 return -ENODEV;
207 }
208 if (!retry--)
209 break;
210 if (wait_for_completion_timeout(comp_done, div_timeout))
211 return 0;
212 }
213 netdev_err(netdev, "Operation timed out.\n");
214 return -ETIMEDOUT;
215 }
216
217 static int alloc_long_term_buff(struct ibmvnic_adapter *adapter,
218 struct ibmvnic_long_term_buff *ltb, int size)
219 {
220 struct device *dev = &adapter->vdev->dev;
221 int rc;
222
223 ltb->size = size;
224 ltb->buff = dma_alloc_coherent(dev, ltb->size, &ltb->addr,
225 GFP_KERNEL);
226
227 if (!ltb->buff) {
228 dev_err(dev, "Couldn't alloc long term buffer\n");
229 return -ENOMEM;
230 }
231 ltb->map_id = adapter->map_id;
232 adapter->map_id++;
233
234 mutex_lock(&adapter->fw_lock);
235 adapter->fw_done_rc = 0;
236 reinit_completion(&adapter->fw_done);
237
238 rc = send_request_map(adapter, ltb->addr, ltb->size, ltb->map_id);
239 if (rc) {
240 dev_err(dev, "send_request_map failed, rc = %d\n", rc);
241 goto out;
242 }
243
244 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
245 if (rc) {
246 dev_err(dev,
247 "Long term map request aborted or timed out,rc = %d\n",
248 rc);
249 goto out;
250 }
251
252 if (adapter->fw_done_rc) {
253 dev_err(dev, "Couldn't map long term buffer,rc = %d\n",
254 adapter->fw_done_rc);
255 rc = -1;
256 goto out;
257 }
258 rc = 0;
259 out:
260 if (rc) {
261 dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
262 ltb->buff = NULL;
263 }
264 mutex_unlock(&adapter->fw_lock);
265 return rc;
266 }
267
268 static void free_long_term_buff(struct ibmvnic_adapter *adapter,
269 struct ibmvnic_long_term_buff *ltb)
270 {
271 struct device *dev = &adapter->vdev->dev;
272
273 if (!ltb->buff)
274 return;
275
276 /* VIOS automatically unmaps the long term buffer at remote
277 * end for the following resets:
278 * FAILOVER, MOBILITY, TIMEOUT.
279 */
280 if (adapter->reset_reason != VNIC_RESET_FAILOVER &&
281 adapter->reset_reason != VNIC_RESET_MOBILITY &&
282 adapter->reset_reason != VNIC_RESET_TIMEOUT)
283 send_request_unmap(adapter, ltb->map_id);
284 dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
285 ltb->buff = NULL;
286 ltb->map_id = 0;
287 }
288
289 static int reset_long_term_buff(struct ibmvnic_adapter *adapter,
290 struct ibmvnic_long_term_buff *ltb)
291 {
292 struct device *dev = &adapter->vdev->dev;
293 int rc;
294
295 memset(ltb->buff, 0, ltb->size);
296
297 mutex_lock(&adapter->fw_lock);
298 adapter->fw_done_rc = 0;
299
300 reinit_completion(&adapter->fw_done);
301 rc = send_request_map(adapter, ltb->addr, ltb->size, ltb->map_id);
302 if (rc) {
303 mutex_unlock(&adapter->fw_lock);
304 return rc;
305 }
306
307 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
308 if (rc) {
309 dev_info(dev,
310 "Reset failed, long term map request timed out or aborted\n");
311 mutex_unlock(&adapter->fw_lock);
312 return rc;
313 }
314
315 if (adapter->fw_done_rc) {
316 dev_info(dev,
317 "Reset failed, attempting to free and reallocate buffer\n");
318 free_long_term_buff(adapter, ltb);
319 mutex_unlock(&adapter->fw_lock);
320 return alloc_long_term_buff(adapter, ltb, ltb->size);
321 }
322 mutex_unlock(&adapter->fw_lock);
323 return 0;
324 }
325
326 static void deactivate_rx_pools(struct ibmvnic_adapter *adapter)
327 {
328 int i;
329
330 for (i = 0; i < adapter->num_active_rx_pools; i++)
331 adapter->rx_pool[i].active = 0;
332 }
333
334 static void replenish_rx_pool(struct ibmvnic_adapter *adapter,
335 struct ibmvnic_rx_pool *pool)
336 {
337 int count = pool->size - atomic_read(&pool->available);
338 u64 handle = adapter->rx_scrq[pool->index]->handle;
339 struct device *dev = &adapter->vdev->dev;
340 struct ibmvnic_ind_xmit_queue *ind_bufp;
341 struct ibmvnic_sub_crq_queue *rx_scrq;
342 union sub_crq *sub_crq;
343 int buffers_added = 0;
344 unsigned long lpar_rc;
345 struct sk_buff *skb;
346 unsigned int offset;
347 dma_addr_t dma_addr;
348 unsigned char *dst;
349 int shift = 0;
350 int index;
351 int i;
352
353 if (!pool->active)
354 return;
355
356 rx_scrq = adapter->rx_scrq[pool->index];
357 ind_bufp = &rx_scrq->ind_buf;
358
359 /* netdev_skb_alloc() could have failed after we saved a few skbs
360 * in the indir_buf and we would not have sent them to VIOS yet.
361 * To account for them, start the loop at ind_bufp->index rather
362 * than 0. If we pushed all the skbs to VIOS, ind_bufp->index will
363 * be 0.
364 */
365 for (i = ind_bufp->index; i < count; ++i) {
366 skb = netdev_alloc_skb(adapter->netdev, pool->buff_size);
367 if (!skb) {
368 dev_err(dev, "Couldn't replenish rx buff\n");
369 adapter->replenish_no_mem++;
370 break;
371 }
372
373 index = pool->free_map[pool->next_free];
374
375 if (pool->rx_buff[index].skb)
376 dev_err(dev, "Inconsistent free_map!\n");
377
378 /* Copy the skb to the long term mapped DMA buffer */
379 offset = index * pool->buff_size;
380 dst = pool->long_term_buff.buff + offset;
381 memset(dst, 0, pool->buff_size);
382 dma_addr = pool->long_term_buff.addr + offset;
383 pool->rx_buff[index].data = dst;
384
385 pool->free_map[pool->next_free] = IBMVNIC_INVALID_MAP;
386 pool->rx_buff[index].dma = dma_addr;
387 pool->rx_buff[index].skb = skb;
388 pool->rx_buff[index].pool_index = pool->index;
389 pool->rx_buff[index].size = pool->buff_size;
390
391 sub_crq = &ind_bufp->indir_arr[ind_bufp->index++];
392 memset(sub_crq, 0, sizeof(*sub_crq));
393 sub_crq->rx_add.first = IBMVNIC_CRQ_CMD;
394 sub_crq->rx_add.correlator =
395 cpu_to_be64((u64)&pool->rx_buff[index]);
396 sub_crq->rx_add.ioba = cpu_to_be32(dma_addr);
397 sub_crq->rx_add.map_id = pool->long_term_buff.map_id;
398
399 /* The length field of the sCRQ is defined to be 24 bits so the
400 * buffer size needs to be left shifted by a byte before it is
401 * converted to big endian to prevent the last byte from being
402 * truncated.
403 */
404 #ifdef __LITTLE_ENDIAN__
405 shift = 8;
406 #endif
407 sub_crq->rx_add.len = cpu_to_be32(pool->buff_size << shift);
408 pool->next_free = (pool->next_free + 1) % pool->size;
409 if (ind_bufp->index == IBMVNIC_MAX_IND_DESCS ||
410 i == count - 1) {
411 lpar_rc =
412 send_subcrq_indirect(adapter, handle,
413 (u64)ind_bufp->indir_dma,
414 (u64)ind_bufp->index);
415 if (lpar_rc != H_SUCCESS)
416 goto failure;
417 buffers_added += ind_bufp->index;
418 adapter->replenish_add_buff_success += ind_bufp->index;
419 ind_bufp->index = 0;
420 }
421 }
422 atomic_add(buffers_added, &pool->available);
423 return;
424
425 failure:
426 if (lpar_rc != H_PARAMETER && lpar_rc != H_CLOSED)
427 dev_err_ratelimited(dev, "rx: replenish packet buffer failed\n");
428 for (i = ind_bufp->index - 1; i >= 0; --i) {
429 struct ibmvnic_rx_buff *rx_buff;
430
431 pool->next_free = pool->next_free == 0 ?
432 pool->size - 1 : pool->next_free - 1;
433 sub_crq = &ind_bufp->indir_arr[i];
434 rx_buff = (struct ibmvnic_rx_buff *)
435 be64_to_cpu(sub_crq->rx_add.correlator);
436 index = (int)(rx_buff - pool->rx_buff);
437 pool->free_map[pool->next_free] = index;
438 dev_kfree_skb_any(pool->rx_buff[index].skb);
439 pool->rx_buff[index].skb = NULL;
440 }
441 adapter->replenish_add_buff_failure += ind_bufp->index;
442 atomic_add(buffers_added, &pool->available);
443 ind_bufp->index = 0;
444 if (lpar_rc == H_CLOSED || adapter->failover_pending) {
445 /* Disable buffer pool replenishment and report carrier off if
446 * queue is closed or pending failover.
447 * Firmware guarantees that a signal will be sent to the
448 * driver, triggering a reset.
449 */
450 deactivate_rx_pools(adapter);
451 netif_carrier_off(adapter->netdev);
452 }
453 }
454
455 static void replenish_pools(struct ibmvnic_adapter *adapter)
456 {
457 int i;
458
459 adapter->replenish_task_cycles++;
460 for (i = 0; i < adapter->num_active_rx_pools; i++) {
461 if (adapter->rx_pool[i].active)
462 replenish_rx_pool(adapter, &adapter->rx_pool[i]);
463 }
464
465 netdev_dbg(adapter->netdev, "Replenished %d pools\n", i);
466 }
467
468 static void release_stats_buffers(struct ibmvnic_adapter *adapter)
469 {
470 kfree(adapter->tx_stats_buffers);
471 kfree(adapter->rx_stats_buffers);
472 adapter->tx_stats_buffers = NULL;
473 adapter->rx_stats_buffers = NULL;
474 }
475
476 static int init_stats_buffers(struct ibmvnic_adapter *adapter)
477 {
478 adapter->tx_stats_buffers =
479 kcalloc(IBMVNIC_MAX_QUEUES,
480 sizeof(struct ibmvnic_tx_queue_stats),
481 GFP_KERNEL);
482 if (!adapter->tx_stats_buffers)
483 return -ENOMEM;
484
485 adapter->rx_stats_buffers =
486 kcalloc(IBMVNIC_MAX_QUEUES,
487 sizeof(struct ibmvnic_rx_queue_stats),
488 GFP_KERNEL);
489 if (!adapter->rx_stats_buffers)
490 return -ENOMEM;
491
492 return 0;
493 }
494
495 static void release_stats_token(struct ibmvnic_adapter *adapter)
496 {
497 struct device *dev = &adapter->vdev->dev;
498
499 if (!adapter->stats_token)
500 return;
501
502 dma_unmap_single(dev, adapter->stats_token,
503 sizeof(struct ibmvnic_statistics),
504 DMA_FROM_DEVICE);
505 adapter->stats_token = 0;
506 }
507
508 static int init_stats_token(struct ibmvnic_adapter *adapter)
509 {
510 struct device *dev = &adapter->vdev->dev;
511 dma_addr_t stok;
512
513 stok = dma_map_single(dev, &adapter->stats,
514 sizeof(struct ibmvnic_statistics),
515 DMA_FROM_DEVICE);
516 if (dma_mapping_error(dev, stok)) {
517 dev_err(dev, "Couldn't map stats buffer\n");
518 return -1;
519 }
520
521 adapter->stats_token = stok;
522 netdev_dbg(adapter->netdev, "Stats token initialized (%llx)\n", stok);
523 return 0;
524 }
525
526 static int reset_rx_pools(struct ibmvnic_adapter *adapter)
527 {
528 struct ibmvnic_rx_pool *rx_pool;
529 u64 buff_size;
530 int rx_scrqs;
531 int i, j, rc;
532
533 if (!adapter->rx_pool)
534 return -1;
535
536 buff_size = adapter->cur_rx_buf_sz;
537 rx_scrqs = adapter->num_active_rx_pools;
538 for (i = 0; i < rx_scrqs; i++) {
539 rx_pool = &adapter->rx_pool[i];
540
541 netdev_dbg(adapter->netdev, "Re-setting rx_pool[%d]\n", i);
542
543 if (rx_pool->buff_size != buff_size) {
544 free_long_term_buff(adapter, &rx_pool->long_term_buff);
545 rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
546 rc = alloc_long_term_buff(adapter,
547 &rx_pool->long_term_buff,
548 rx_pool->size *
549 rx_pool->buff_size);
550 } else {
551 rc = reset_long_term_buff(adapter,
552 &rx_pool->long_term_buff);
553 }
554
555 if (rc)
556 return rc;
557
558 for (j = 0; j < rx_pool->size; j++)
559 rx_pool->free_map[j] = j;
560
561 memset(rx_pool->rx_buff, 0,
562 rx_pool->size * sizeof(struct ibmvnic_rx_buff));
563
564 atomic_set(&rx_pool->available, 0);
565 rx_pool->next_alloc = 0;
566 rx_pool->next_free = 0;
567 rx_pool->active = 1;
568 }
569
570 return 0;
571 }
572
573 static void release_rx_pools(struct ibmvnic_adapter *adapter)
574 {
575 struct ibmvnic_rx_pool *rx_pool;
576 int i, j;
577
578 if (!adapter->rx_pool)
579 return;
580
581 for (i = 0; i < adapter->num_active_rx_pools; i++) {
582 rx_pool = &adapter->rx_pool[i];
583
584 netdev_dbg(adapter->netdev, "Releasing rx_pool[%d]\n", i);
585
586 kfree(rx_pool->free_map);
587 free_long_term_buff(adapter, &rx_pool->long_term_buff);
588
589 if (!rx_pool->rx_buff)
590 continue;
591
592 for (j = 0; j < rx_pool->size; j++) {
593 if (rx_pool->rx_buff[j].skb) {
594 dev_kfree_skb_any(rx_pool->rx_buff[j].skb);
595 rx_pool->rx_buff[j].skb = NULL;
596 }
597 }
598
599 kfree(rx_pool->rx_buff);
600 }
601
602 kfree(adapter->rx_pool);
603 adapter->rx_pool = NULL;
604 adapter->num_active_rx_pools = 0;
605 }
606
607 static int init_rx_pools(struct net_device *netdev)
608 {
609 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
610 struct device *dev = &adapter->vdev->dev;
611 struct ibmvnic_rx_pool *rx_pool;
612 int rxadd_subcrqs;
613 u64 buff_size;
614 int i, j;
615
616 rxadd_subcrqs = adapter->num_active_rx_scrqs;
617 buff_size = adapter->cur_rx_buf_sz;
618
619 adapter->rx_pool = kcalloc(rxadd_subcrqs,
620 sizeof(struct ibmvnic_rx_pool),
621 GFP_KERNEL);
622 if (!adapter->rx_pool) {
623 dev_err(dev, "Failed to allocate rx pools\n");
624 return -1;
625 }
626
627 adapter->num_active_rx_pools = rxadd_subcrqs;
628
629 for (i = 0; i < rxadd_subcrqs; i++) {
630 rx_pool = &adapter->rx_pool[i];
631
632 netdev_dbg(adapter->netdev,
633 "Initializing rx_pool[%d], %lld buffs, %lld bytes each\n",
634 i, adapter->req_rx_add_entries_per_subcrq,
635 buff_size);
636
637 rx_pool->size = adapter->req_rx_add_entries_per_subcrq;
638 rx_pool->index = i;
639 rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
640 rx_pool->active = 1;
641
642 rx_pool->free_map = kcalloc(rx_pool->size, sizeof(int),
643 GFP_KERNEL);
644 if (!rx_pool->free_map) {
645 release_rx_pools(adapter);
646 return -1;
647 }
648
649 rx_pool->rx_buff = kcalloc(rx_pool->size,
650 sizeof(struct ibmvnic_rx_buff),
651 GFP_KERNEL);
652 if (!rx_pool->rx_buff) {
653 dev_err(dev, "Couldn't alloc rx buffers\n");
654 release_rx_pools(adapter);
655 return -1;
656 }
657
658 if (alloc_long_term_buff(adapter, &rx_pool->long_term_buff,
659 rx_pool->size * rx_pool->buff_size)) {
660 release_rx_pools(adapter);
661 return -1;
662 }
663
664 for (j = 0; j < rx_pool->size; ++j)
665 rx_pool->free_map[j] = j;
666
667 atomic_set(&rx_pool->available, 0);
668 rx_pool->next_alloc = 0;
669 rx_pool->next_free = 0;
670 }
671
672 return 0;
673 }
674
675 static int reset_one_tx_pool(struct ibmvnic_adapter *adapter,
676 struct ibmvnic_tx_pool *tx_pool)
677 {
678 int rc, i;
679
680 rc = reset_long_term_buff(adapter, &tx_pool->long_term_buff);
681 if (rc)
682 return rc;
683
684 memset(tx_pool->tx_buff, 0,
685 tx_pool->num_buffers *
686 sizeof(struct ibmvnic_tx_buff));
687
688 for (i = 0; i < tx_pool->num_buffers; i++)
689 tx_pool->free_map[i] = i;
690
691 tx_pool->consumer_index = 0;
692 tx_pool->producer_index = 0;
693
694 return 0;
695 }
696
697 static int reset_tx_pools(struct ibmvnic_adapter *adapter)
698 {
699 int tx_scrqs;
700 int i, rc;
701
702 if (!adapter->tx_pool)
703 return -1;
704
705 tx_scrqs = adapter->num_active_tx_pools;
706 for (i = 0; i < tx_scrqs; i++) {
707 ibmvnic_tx_scrq_clean_buffer(adapter, adapter->tx_scrq[i]);
708 rc = reset_one_tx_pool(adapter, &adapter->tso_pool[i]);
709 if (rc)
710 return rc;
711 rc = reset_one_tx_pool(adapter, &adapter->tx_pool[i]);
712 if (rc)
713 return rc;
714 }
715
716 return 0;
717 }
718
719 static void release_vpd_data(struct ibmvnic_adapter *adapter)
720 {
721 if (!adapter->vpd)
722 return;
723
724 kfree(adapter->vpd->buff);
725 kfree(adapter->vpd);
726
727 adapter->vpd = NULL;
728 }
729
730 static void release_one_tx_pool(struct ibmvnic_adapter *adapter,
731 struct ibmvnic_tx_pool *tx_pool)
732 {
733 kfree(tx_pool->tx_buff);
734 kfree(tx_pool->free_map);
735 free_long_term_buff(adapter, &tx_pool->long_term_buff);
736 }
737
738 static void release_tx_pools(struct ibmvnic_adapter *adapter)
739 {
740 int i;
741
742 if (!adapter->tx_pool)
743 return;
744
745 for (i = 0; i < adapter->num_active_tx_pools; i++) {
746 release_one_tx_pool(adapter, &adapter->tx_pool[i]);
747 release_one_tx_pool(adapter, &adapter->tso_pool[i]);
748 }
749
750 kfree(adapter->tx_pool);
751 adapter->tx_pool = NULL;
752 kfree(adapter->tso_pool);
753 adapter->tso_pool = NULL;
754 adapter->num_active_tx_pools = 0;
755 }
756
757 static int init_one_tx_pool(struct net_device *netdev,
758 struct ibmvnic_tx_pool *tx_pool,
759 int num_entries, int buf_size)
760 {
761 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
762 int i;
763
764 tx_pool->tx_buff = kcalloc(num_entries,
765 sizeof(struct ibmvnic_tx_buff),
766 GFP_KERNEL);
767 if (!tx_pool->tx_buff)
768 return -1;
769
770 if (alloc_long_term_buff(adapter, &tx_pool->long_term_buff,
771 num_entries * buf_size))
772 return -1;
773
774 tx_pool->free_map = kcalloc(num_entries, sizeof(int), GFP_KERNEL);
775 if (!tx_pool->free_map)
776 return -1;
777
778 for (i = 0; i < num_entries; i++)
779 tx_pool->free_map[i] = i;
780
781 tx_pool->consumer_index = 0;
782 tx_pool->producer_index = 0;
783 tx_pool->num_buffers = num_entries;
784 tx_pool->buf_size = buf_size;
785
786 return 0;
787 }
788
789 static int init_tx_pools(struct net_device *netdev)
790 {
791 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
792 int tx_subcrqs;
793 u64 buff_size;
794 int i, rc;
795
796 tx_subcrqs = adapter->num_active_tx_scrqs;
797 adapter->tx_pool = kcalloc(tx_subcrqs,
798 sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
799 if (!adapter->tx_pool)
800 return -1;
801
802 adapter->tso_pool = kcalloc(tx_subcrqs,
803 sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
804 if (!adapter->tso_pool) {
805 kfree(adapter->tx_pool);
806 adapter->tx_pool = NULL;
807 return -1;
808 }
809
810 adapter->num_active_tx_pools = tx_subcrqs;
811
812 for (i = 0; i < tx_subcrqs; i++) {
813 buff_size = adapter->req_mtu + VLAN_HLEN;
814 buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
815 rc = init_one_tx_pool(netdev, &adapter->tx_pool[i],
816 adapter->req_tx_entries_per_subcrq,
817 buff_size);
818 if (rc) {
819 release_tx_pools(adapter);
820 return rc;
821 }
822
823 rc = init_one_tx_pool(netdev, &adapter->tso_pool[i],
824 IBMVNIC_TSO_BUFS,
825 IBMVNIC_TSO_BUF_SZ);
826 if (rc) {
827 release_tx_pools(adapter);
828 return rc;
829 }
830 }
831
832 return 0;
833 }
834
835 static void ibmvnic_napi_enable(struct ibmvnic_adapter *adapter)
836 {
837 int i;
838
839 if (adapter->napi_enabled)
840 return;
841
842 for (i = 0; i < adapter->req_rx_queues; i++)
843 napi_enable(&adapter->napi[i]);
844
845 adapter->napi_enabled = true;
846 }
847
848 static void ibmvnic_napi_disable(struct ibmvnic_adapter *adapter)
849 {
850 int i;
851
852 if (!adapter->napi_enabled)
853 return;
854
855 for (i = 0; i < adapter->req_rx_queues; i++) {
856 netdev_dbg(adapter->netdev, "Disabling napi[%d]\n", i);
857 napi_disable(&adapter->napi[i]);
858 }
859
860 adapter->napi_enabled = false;
861 }
862
863 static int init_napi(struct ibmvnic_adapter *adapter)
864 {
865 int i;
866
867 adapter->napi = kcalloc(adapter->req_rx_queues,
868 sizeof(struct napi_struct), GFP_KERNEL);
869 if (!adapter->napi)
870 return -ENOMEM;
871
872 for (i = 0; i < adapter->req_rx_queues; i++) {
873 netdev_dbg(adapter->netdev, "Adding napi[%d]\n", i);
874 netif_napi_add(adapter->netdev, &adapter->napi[i],
875 ibmvnic_poll, NAPI_POLL_WEIGHT);
876 }
877
878 adapter->num_active_rx_napi = adapter->req_rx_queues;
879 return 0;
880 }
881
882 static void release_napi(struct ibmvnic_adapter *adapter)
883 {
884 int i;
885
886 if (!adapter->napi)
887 return;
888
889 for (i = 0; i < adapter->num_active_rx_napi; i++) {
890 netdev_dbg(adapter->netdev, "Releasing napi[%d]\n", i);
891 netif_napi_del(&adapter->napi[i]);
892 }
893
894 kfree(adapter->napi);
895 adapter->napi = NULL;
896 adapter->num_active_rx_napi = 0;
897 adapter->napi_enabled = false;
898 }
899
900 static const char *adapter_state_to_string(enum vnic_state state)
901 {
902 switch (state) {
903 case VNIC_PROBING:
904 return "PROBING";
905 case VNIC_PROBED:
906 return "PROBED";
907 case VNIC_OPENING:
908 return "OPENING";
909 case VNIC_OPEN:
910 return "OPEN";
911 case VNIC_CLOSING:
912 return "CLOSING";
913 case VNIC_CLOSED:
914 return "CLOSED";
915 case VNIC_REMOVING:
916 return "REMOVING";
917 case VNIC_REMOVED:
918 return "REMOVED";
919 case VNIC_DOWN:
920 return "DOWN";
921 }
922 return "UNKNOWN";
923 }
924
925 static int ibmvnic_login(struct net_device *netdev)
926 {
927 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
928 unsigned long timeout = msecs_to_jiffies(20000);
929 int retry_count = 0;
930 int retries = 10;
931 bool retry;
932 int rc;
933
934 do {
935 retry = false;
936 if (retry_count > retries) {
937 netdev_warn(netdev, "Login attempts exceeded\n");
938 return -1;
939 }
940
941 adapter->init_done_rc = 0;
942 reinit_completion(&adapter->init_done);
943 rc = send_login(adapter);
944 if (rc)
945 return rc;
946
947 if (!wait_for_completion_timeout(&adapter->init_done,
948 timeout)) {
949 netdev_warn(netdev, "Login timed out, retrying...\n");
950 retry = true;
951 adapter->init_done_rc = 0;
952 retry_count++;
953 continue;
954 }
955
956 if (adapter->init_done_rc == ABORTED) {
957 netdev_warn(netdev, "Login aborted, retrying...\n");
958 retry = true;
959 adapter->init_done_rc = 0;
960 retry_count++;
961 /* FW or device may be busy, so
962 * wait a bit before retrying login
963 */
964 msleep(500);
965 } else if (adapter->init_done_rc == PARTIALSUCCESS) {
966 retry_count++;
967 release_sub_crqs(adapter, 1);
968
969 retry = true;
970 netdev_dbg(netdev,
971 "Received partial success, retrying...\n");
972 adapter->init_done_rc = 0;
973 reinit_completion(&adapter->init_done);
974 send_query_cap(adapter);
975 if (!wait_for_completion_timeout(&adapter->init_done,
976 timeout)) {
977 netdev_warn(netdev,
978 "Capabilities query timed out\n");
979 return -1;
980 }
981
982 rc = init_sub_crqs(adapter);
983 if (rc) {
984 netdev_warn(netdev,
985 "SCRQ initialization failed\n");
986 return -1;
987 }
988
989 rc = init_sub_crq_irqs(adapter);
990 if (rc) {
991 netdev_warn(netdev,
992 "SCRQ irq initialization failed\n");
993 return -1;
994 }
995 } else if (adapter->init_done_rc) {
996 netdev_warn(netdev, "Adapter login failed\n");
997 return -1;
998 }
999 } while (retry);
1000
1001 __ibmvnic_set_mac(netdev, adapter->mac_addr);
1002
1003 netdev_dbg(netdev, "[S:%s] Login succeeded\n", adapter_state_to_string(adapter->state));
1004 return 0;
1005 }
1006
1007 static void release_login_buffer(struct ibmvnic_adapter *adapter)
1008 {
1009 kfree(adapter->login_buf);
1010 adapter->login_buf = NULL;
1011 }
1012
1013 static void release_login_rsp_buffer(struct ibmvnic_adapter *adapter)
1014 {
1015 kfree(adapter->login_rsp_buf);
1016 adapter->login_rsp_buf = NULL;
1017 }
1018
1019 static void release_resources(struct ibmvnic_adapter *adapter)
1020 {
1021 release_vpd_data(adapter);
1022
1023 release_tx_pools(adapter);
1024 release_rx_pools(adapter);
1025
1026 release_napi(adapter);
1027 release_login_buffer(adapter);
1028 release_login_rsp_buffer(adapter);
1029 }
1030
1031 static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state)
1032 {
1033 struct net_device *netdev = adapter->netdev;
1034 unsigned long timeout = msecs_to_jiffies(20000);
1035 union ibmvnic_crq crq;
1036 bool resend;
1037 int rc;
1038
1039 netdev_dbg(netdev, "setting link state %d\n", link_state);
1040
1041 memset(&crq, 0, sizeof(crq));
1042 crq.logical_link_state.first = IBMVNIC_CRQ_CMD;
1043 crq.logical_link_state.cmd = LOGICAL_LINK_STATE;
1044 crq.logical_link_state.link_state = link_state;
1045
1046 do {
1047 resend = false;
1048
1049 reinit_completion(&adapter->init_done);
1050 rc = ibmvnic_send_crq(adapter, &crq);
1051 if (rc) {
1052 netdev_err(netdev, "Failed to set link state\n");
1053 return rc;
1054 }
1055
1056 if (!wait_for_completion_timeout(&adapter->init_done,
1057 timeout)) {
1058 netdev_err(netdev, "timeout setting link state\n");
1059 return -1;
1060 }
1061
1062 if (adapter->init_done_rc == PARTIALSUCCESS) {
1063 /* Partuial success, delay and re-send */
1064 mdelay(1000);
1065 resend = true;
1066 } else if (adapter->init_done_rc) {
1067 netdev_warn(netdev, "Unable to set link state, rc=%d\n",
1068 adapter->init_done_rc);
1069 return adapter->init_done_rc;
1070 }
1071 } while (resend);
1072
1073 return 0;
1074 }
1075
1076 static int set_real_num_queues(struct net_device *netdev)
1077 {
1078 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1079 int rc;
1080
1081 netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n",
1082 adapter->req_tx_queues, adapter->req_rx_queues);
1083
1084 rc = netif_set_real_num_tx_queues(netdev, adapter->req_tx_queues);
1085 if (rc) {
1086 netdev_err(netdev, "failed to set the number of tx queues\n");
1087 return rc;
1088 }
1089
1090 rc = netif_set_real_num_rx_queues(netdev, adapter->req_rx_queues);
1091 if (rc)
1092 netdev_err(netdev, "failed to set the number of rx queues\n");
1093
1094 return rc;
1095 }
1096
1097 static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter)
1098 {
1099 struct device *dev = &adapter->vdev->dev;
1100 union ibmvnic_crq crq;
1101 int len = 0;
1102 int rc;
1103
1104 if (adapter->vpd->buff)
1105 len = adapter->vpd->len;
1106
1107 mutex_lock(&adapter->fw_lock);
1108 adapter->fw_done_rc = 0;
1109 reinit_completion(&adapter->fw_done);
1110
1111 crq.get_vpd_size.first = IBMVNIC_CRQ_CMD;
1112 crq.get_vpd_size.cmd = GET_VPD_SIZE;
1113 rc = ibmvnic_send_crq(adapter, &crq);
1114 if (rc) {
1115 mutex_unlock(&adapter->fw_lock);
1116 return rc;
1117 }
1118
1119 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1120 if (rc) {
1121 dev_err(dev, "Could not retrieve VPD size, rc = %d\n", rc);
1122 mutex_unlock(&adapter->fw_lock);
1123 return rc;
1124 }
1125 mutex_unlock(&adapter->fw_lock);
1126
1127 if (!adapter->vpd->len)
1128 return -ENODATA;
1129
1130 if (!adapter->vpd->buff)
1131 adapter->vpd->buff = kzalloc(adapter->vpd->len, GFP_KERNEL);
1132 else if (adapter->vpd->len != len)
1133 adapter->vpd->buff =
1134 krealloc(adapter->vpd->buff,
1135 adapter->vpd->len, GFP_KERNEL);
1136
1137 if (!adapter->vpd->buff) {
1138 dev_err(dev, "Could allocate VPD buffer\n");
1139 return -ENOMEM;
1140 }
1141
1142 adapter->vpd->dma_addr =
1143 dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len,
1144 DMA_FROM_DEVICE);
1145 if (dma_mapping_error(dev, adapter->vpd->dma_addr)) {
1146 dev_err(dev, "Could not map VPD buffer\n");
1147 kfree(adapter->vpd->buff);
1148 adapter->vpd->buff = NULL;
1149 return -ENOMEM;
1150 }
1151
1152 mutex_lock(&adapter->fw_lock);
1153 adapter->fw_done_rc = 0;
1154 reinit_completion(&adapter->fw_done);
1155
1156 crq.get_vpd.first = IBMVNIC_CRQ_CMD;
1157 crq.get_vpd.cmd = GET_VPD;
1158 crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr);
1159 crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len);
1160 rc = ibmvnic_send_crq(adapter, &crq);
1161 if (rc) {
1162 kfree(adapter->vpd->buff);
1163 adapter->vpd->buff = NULL;
1164 mutex_unlock(&adapter->fw_lock);
1165 return rc;
1166 }
1167
1168 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1169 if (rc) {
1170 dev_err(dev, "Unable to retrieve VPD, rc = %d\n", rc);
1171 kfree(adapter->vpd->buff);
1172 adapter->vpd->buff = NULL;
1173 mutex_unlock(&adapter->fw_lock);
1174 return rc;
1175 }
1176
1177 mutex_unlock(&adapter->fw_lock);
1178 return 0;
1179 }
1180
1181 static int init_resources(struct ibmvnic_adapter *adapter)
1182 {
1183 struct net_device *netdev = adapter->netdev;
1184 int rc;
1185
1186 rc = set_real_num_queues(netdev);
1187 if (rc)
1188 return rc;
1189
1190 adapter->vpd = kzalloc(sizeof(*adapter->vpd), GFP_KERNEL);
1191 if (!adapter->vpd)
1192 return -ENOMEM;
1193
1194 /* Vital Product Data (VPD) */
1195 rc = ibmvnic_get_vpd(adapter);
1196 if (rc) {
1197 netdev_err(netdev, "failed to initialize Vital Product Data (VPD)\n");
1198 return rc;
1199 }
1200
1201 adapter->map_id = 1;
1202
1203 rc = init_napi(adapter);
1204 if (rc)
1205 return rc;
1206
1207 send_query_map(adapter);
1208
1209 rc = init_rx_pools(netdev);
1210 if (rc)
1211 return rc;
1212
1213 rc = init_tx_pools(netdev);
1214 return rc;
1215 }
1216
1217 static int __ibmvnic_open(struct net_device *netdev)
1218 {
1219 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1220 enum vnic_state prev_state = adapter->state;
1221 int i, rc;
1222
1223 adapter->state = VNIC_OPENING;
1224 replenish_pools(adapter);
1225 ibmvnic_napi_enable(adapter);
1226
1227 /* We're ready to receive frames, enable the sub-crq interrupts and
1228 * set the logical link state to up
1229 */
1230 for (i = 0; i < adapter->req_rx_queues; i++) {
1231 netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i);
1232 if (prev_state == VNIC_CLOSED)
1233 enable_irq(adapter->rx_scrq[i]->irq);
1234 enable_scrq_irq(adapter, adapter->rx_scrq[i]);
1235 }
1236
1237 for (i = 0; i < adapter->req_tx_queues; i++) {
1238 netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i);
1239 if (prev_state == VNIC_CLOSED)
1240 enable_irq(adapter->tx_scrq[i]->irq);
1241 enable_scrq_irq(adapter, adapter->tx_scrq[i]);
1242 netdev_tx_reset_queue(netdev_get_tx_queue(netdev, i));
1243 }
1244
1245 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP);
1246 if (rc) {
1247 ibmvnic_napi_disable(adapter);
1248 release_resources(adapter);
1249 return rc;
1250 }
1251
1252 netif_tx_start_all_queues(netdev);
1253
1254 if (prev_state == VNIC_CLOSED) {
1255 for (i = 0; i < adapter->req_rx_queues; i++)
1256 napi_schedule(&adapter->napi[i]);
1257 }
1258
1259 adapter->state = VNIC_OPEN;
1260 return rc;
1261 }
1262
1263 static int ibmvnic_open(struct net_device *netdev)
1264 {
1265 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1266 int rc;
1267
1268 ASSERT_RTNL();
1269
1270 /* If device failover is pending or we are about to reset, just set
1271 * device state and return. Device operation will be handled by reset
1272 * routine.
1273 *
1274 * It should be safe to overwrite the adapter->state here. Since
1275 * we hold the rtnl, either the reset has not actually started or
1276 * the rtnl got dropped during the set_link_state() in do_reset().
1277 * In the former case, no one else is changing the state (again we
1278 * have the rtnl) and in the latter case, do_reset() will detect and
1279 * honor our setting below.
1280 */
1281 if (adapter->failover_pending || (test_bit(0, &adapter->resetting))) {
1282 netdev_dbg(netdev, "[S:%s FOP:%d] Resetting, deferring open\n",
1283 adapter_state_to_string(adapter->state),
1284 adapter->failover_pending);
1285 adapter->state = VNIC_OPEN;
1286 rc = 0;
1287 goto out;
1288 }
1289
1290 if (adapter->state != VNIC_CLOSED) {
1291 rc = ibmvnic_login(netdev);
1292 if (rc)
1293 goto out;
1294
1295 rc = init_resources(adapter);
1296 if (rc) {
1297 netdev_err(netdev, "failed to initialize resources\n");
1298 release_resources(adapter);
1299 goto out;
1300 }
1301 }
1302
1303 rc = __ibmvnic_open(netdev);
1304
1305 out:
1306 /* If open failed and there is a pending failover or in-progress reset,
1307 * set device state and return. Device operation will be handled by
1308 * reset routine. See also comments above regarding rtnl.
1309 */
1310 if (rc &&
1311 (adapter->failover_pending || (test_bit(0, &adapter->resetting)))) {
1312 adapter->state = VNIC_OPEN;
1313 rc = 0;
1314 }
1315 return rc;
1316 }
1317
1318 static void clean_rx_pools(struct ibmvnic_adapter *adapter)
1319 {
1320 struct ibmvnic_rx_pool *rx_pool;
1321 struct ibmvnic_rx_buff *rx_buff;
1322 u64 rx_entries;
1323 int rx_scrqs;
1324 int i, j;
1325
1326 if (!adapter->rx_pool)
1327 return;
1328
1329 rx_scrqs = adapter->num_active_rx_pools;
1330 rx_entries = adapter->req_rx_add_entries_per_subcrq;
1331
1332 /* Free any remaining skbs in the rx buffer pools */
1333 for (i = 0; i < rx_scrqs; i++) {
1334 rx_pool = &adapter->rx_pool[i];
1335 if (!rx_pool || !rx_pool->rx_buff)
1336 continue;
1337
1338 netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i);
1339 for (j = 0; j < rx_entries; j++) {
1340 rx_buff = &rx_pool->rx_buff[j];
1341 if (rx_buff && rx_buff->skb) {
1342 dev_kfree_skb_any(rx_buff->skb);
1343 rx_buff->skb = NULL;
1344 }
1345 }
1346 }
1347 }
1348
1349 static void clean_one_tx_pool(struct ibmvnic_adapter *adapter,
1350 struct ibmvnic_tx_pool *tx_pool)
1351 {
1352 struct ibmvnic_tx_buff *tx_buff;
1353 u64 tx_entries;
1354 int i;
1355
1356 if (!tx_pool || !tx_pool->tx_buff)
1357 return;
1358
1359 tx_entries = tx_pool->num_buffers;
1360
1361 for (i = 0; i < tx_entries; i++) {
1362 tx_buff = &tx_pool->tx_buff[i];
1363 if (tx_buff && tx_buff->skb) {
1364 dev_kfree_skb_any(tx_buff->skb);
1365 tx_buff->skb = NULL;
1366 }
1367 }
1368 }
1369
1370 static void clean_tx_pools(struct ibmvnic_adapter *adapter)
1371 {
1372 int tx_scrqs;
1373 int i;
1374
1375 if (!adapter->tx_pool || !adapter->tso_pool)
1376 return;
1377
1378 tx_scrqs = adapter->num_active_tx_pools;
1379
1380 /* Free any remaining skbs in the tx buffer pools */
1381 for (i = 0; i < tx_scrqs; i++) {
1382 netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i);
1383 clean_one_tx_pool(adapter, &adapter->tx_pool[i]);
1384 clean_one_tx_pool(adapter, &adapter->tso_pool[i]);
1385 }
1386 }
1387
1388 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter)
1389 {
1390 struct net_device *netdev = adapter->netdev;
1391 int i;
1392
1393 if (adapter->tx_scrq) {
1394 for (i = 0; i < adapter->req_tx_queues; i++)
1395 if (adapter->tx_scrq[i]->irq) {
1396 netdev_dbg(netdev,
1397 "Disabling tx_scrq[%d] irq\n", i);
1398 disable_scrq_irq(adapter, adapter->tx_scrq[i]);
1399 disable_irq(adapter->tx_scrq[i]->irq);
1400 }
1401 }
1402
1403 if (adapter->rx_scrq) {
1404 for (i = 0; i < adapter->req_rx_queues; i++) {
1405 if (adapter->rx_scrq[i]->irq) {
1406 netdev_dbg(netdev,
1407 "Disabling rx_scrq[%d] irq\n", i);
1408 disable_scrq_irq(adapter, adapter->rx_scrq[i]);
1409 disable_irq(adapter->rx_scrq[i]->irq);
1410 }
1411 }
1412 }
1413 }
1414
1415 static void ibmvnic_cleanup(struct net_device *netdev)
1416 {
1417 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1418
1419 /* ensure that transmissions are stopped if called by do_reset */
1420 if (test_bit(0, &adapter->resetting))
1421 netif_tx_disable(netdev);
1422 else
1423 netif_tx_stop_all_queues(netdev);
1424
1425 ibmvnic_napi_disable(adapter);
1426 ibmvnic_disable_irqs(adapter);
1427
1428 clean_rx_pools(adapter);
1429 clean_tx_pools(adapter);
1430 }
1431
1432 static int __ibmvnic_close(struct net_device *netdev)
1433 {
1434 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1435 int rc = 0;
1436
1437 adapter->state = VNIC_CLOSING;
1438 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
1439 adapter->state = VNIC_CLOSED;
1440 return rc;
1441 }
1442
1443 static int ibmvnic_close(struct net_device *netdev)
1444 {
1445 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1446 int rc;
1447
1448 netdev_dbg(netdev, "[S:%s FOP:%d FRR:%d] Closing\n",
1449 adapter_state_to_string(adapter->state),
1450 adapter->failover_pending,
1451 adapter->force_reset_recovery);
1452
1453 /* If device failover is pending, just set device state and return.
1454 * Device operation will be handled by reset routine.
1455 */
1456 if (adapter->failover_pending) {
1457 adapter->state = VNIC_CLOSED;
1458 return 0;
1459 }
1460
1461 rc = __ibmvnic_close(netdev);
1462 ibmvnic_cleanup(netdev);
1463
1464 return rc;
1465 }
1466
1467 /**
1468 * build_hdr_data - creates L2/L3/L4 header data buffer
1469 * @hdr_field: bitfield determining needed headers
1470 * @skb: socket buffer
1471 * @hdr_len: array of header lengths
1472 * @hdr_data: buffer to write the header to
1473 *
1474 * Reads hdr_field to determine which headers are needed by firmware.
1475 * Builds a buffer containing these headers. Saves individual header
1476 * lengths and total buffer length to be used to build descriptors.
1477 */
1478 static int build_hdr_data(u8 hdr_field, struct sk_buff *skb,
1479 int *hdr_len, u8 *hdr_data)
1480 {
1481 int len = 0;
1482 u8 *hdr;
1483
1484 if (skb_vlan_tagged(skb) && !skb_vlan_tag_present(skb))
1485 hdr_len[0] = sizeof(struct vlan_ethhdr);
1486 else
1487 hdr_len[0] = sizeof(struct ethhdr);
1488
1489 if (skb->protocol == htons(ETH_P_IP)) {
1490 hdr_len[1] = ip_hdr(skb)->ihl * 4;
1491 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1492 hdr_len[2] = tcp_hdrlen(skb);
1493 else if (ip_hdr(skb)->protocol == IPPROTO_UDP)
1494 hdr_len[2] = sizeof(struct udphdr);
1495 } else if (skb->protocol == htons(ETH_P_IPV6)) {
1496 hdr_len[1] = sizeof(struct ipv6hdr);
1497 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1498 hdr_len[2] = tcp_hdrlen(skb);
1499 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
1500 hdr_len[2] = sizeof(struct udphdr);
1501 } else if (skb->protocol == htons(ETH_P_ARP)) {
1502 hdr_len[1] = arp_hdr_len(skb->dev);
1503 hdr_len[2] = 0;
1504 }
1505
1506 memset(hdr_data, 0, 120);
1507 if ((hdr_field >> 6) & 1) {
1508 hdr = skb_mac_header(skb);
1509 memcpy(hdr_data, hdr, hdr_len[0]);
1510 len += hdr_len[0];
1511 }
1512
1513 if ((hdr_field >> 5) & 1) {
1514 hdr = skb_network_header(skb);
1515 memcpy(hdr_data + len, hdr, hdr_len[1]);
1516 len += hdr_len[1];
1517 }
1518
1519 if ((hdr_field >> 4) & 1) {
1520 hdr = skb_transport_header(skb);
1521 memcpy(hdr_data + len, hdr, hdr_len[2]);
1522 len += hdr_len[2];
1523 }
1524 return len;
1525 }
1526
1527 /**
1528 * create_hdr_descs - create header and header extension descriptors
1529 * @hdr_field: bitfield determining needed headers
1530 * @hdr_data: buffer containing header data
1531 * @len: length of data buffer
1532 * @hdr_len: array of individual header lengths
1533 * @scrq_arr: descriptor array
1534 *
1535 * Creates header and, if needed, header extension descriptors and
1536 * places them in a descriptor array, scrq_arr
1537 */
1538
1539 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len,
1540 union sub_crq *scrq_arr)
1541 {
1542 union sub_crq hdr_desc;
1543 int tmp_len = len;
1544 int num_descs = 0;
1545 u8 *data, *cur;
1546 int tmp;
1547
1548 while (tmp_len > 0) {
1549 cur = hdr_data + len - tmp_len;
1550
1551 memset(&hdr_desc, 0, sizeof(hdr_desc));
1552 if (cur != hdr_data) {
1553 data = hdr_desc.hdr_ext.data;
1554 tmp = tmp_len > 29 ? 29 : tmp_len;
1555 hdr_desc.hdr_ext.first = IBMVNIC_CRQ_CMD;
1556 hdr_desc.hdr_ext.type = IBMVNIC_HDR_EXT_DESC;
1557 hdr_desc.hdr_ext.len = tmp;
1558 } else {
1559 data = hdr_desc.hdr.data;
1560 tmp = tmp_len > 24 ? 24 : tmp_len;
1561 hdr_desc.hdr.first = IBMVNIC_CRQ_CMD;
1562 hdr_desc.hdr.type = IBMVNIC_HDR_DESC;
1563 hdr_desc.hdr.len = tmp;
1564 hdr_desc.hdr.l2_len = (u8)hdr_len[0];
1565 hdr_desc.hdr.l3_len = cpu_to_be16((u16)hdr_len[1]);
1566 hdr_desc.hdr.l4_len = (u8)hdr_len[2];
1567 hdr_desc.hdr.flag = hdr_field << 1;
1568 }
1569 memcpy(data, cur, tmp);
1570 tmp_len -= tmp;
1571 *scrq_arr = hdr_desc;
1572 scrq_arr++;
1573 num_descs++;
1574 }
1575
1576 return num_descs;
1577 }
1578
1579 /**
1580 * build_hdr_descs_arr - build a header descriptor array
1581 * @skb: tx socket buffer
1582 * @indir_arr: indirect array
1583 * @num_entries: number of descriptors to be sent
1584 * @hdr_field: bit field determining which headers will be sent
1585 *
1586 * This function will build a TX descriptor array with applicable
1587 * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect.
1588 */
1589
1590 static void build_hdr_descs_arr(struct sk_buff *skb,
1591 union sub_crq *indir_arr,
1592 int *num_entries, u8 hdr_field)
1593 {
1594 int hdr_len[3] = {0, 0, 0};
1595 u8 hdr_data[140] = {0};
1596 int tot_len;
1597
1598 tot_len = build_hdr_data(hdr_field, skb, hdr_len,
1599 hdr_data);
1600 *num_entries += create_hdr_descs(hdr_field, hdr_data, tot_len, hdr_len,
1601 indir_arr + 1);
1602 }
1603
1604 static int ibmvnic_xmit_workarounds(struct sk_buff *skb,
1605 struct net_device *netdev)
1606 {
1607 /* For some backing devices, mishandling of small packets
1608 * can result in a loss of connection or TX stall. Device
1609 * architects recommend that no packet should be smaller
1610 * than the minimum MTU value provided to the driver, so
1611 * pad any packets to that length
1612 */
1613 if (skb->len < netdev->min_mtu)
1614 return skb_put_padto(skb, netdev->min_mtu);
1615
1616 return 0;
1617 }
1618
1619 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
1620 struct ibmvnic_sub_crq_queue *tx_scrq)
1621 {
1622 struct ibmvnic_ind_xmit_queue *ind_bufp;
1623 struct ibmvnic_tx_buff *tx_buff;
1624 struct ibmvnic_tx_pool *tx_pool;
1625 union sub_crq tx_scrq_entry;
1626 int queue_num;
1627 int entries;
1628 int index;
1629 int i;
1630
1631 ind_bufp = &tx_scrq->ind_buf;
1632 entries = (u64)ind_bufp->index;
1633 queue_num = tx_scrq->pool_index;
1634
1635 for (i = entries - 1; i >= 0; --i) {
1636 tx_scrq_entry = ind_bufp->indir_arr[i];
1637 if (tx_scrq_entry.v1.type != IBMVNIC_TX_DESC)
1638 continue;
1639 index = be32_to_cpu(tx_scrq_entry.v1.correlator);
1640 if (index & IBMVNIC_TSO_POOL_MASK) {
1641 tx_pool = &adapter->tso_pool[queue_num];
1642 index &= ~IBMVNIC_TSO_POOL_MASK;
1643 } else {
1644 tx_pool = &adapter->tx_pool[queue_num];
1645 }
1646 tx_pool->free_map[tx_pool->consumer_index] = index;
1647 tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
1648 tx_pool->num_buffers - 1 :
1649 tx_pool->consumer_index - 1;
1650 tx_buff = &tx_pool->tx_buff[index];
1651 adapter->netdev->stats.tx_packets--;
1652 adapter->netdev->stats.tx_bytes -= tx_buff->skb->len;
1653 adapter->tx_stats_buffers[queue_num].packets--;
1654 adapter->tx_stats_buffers[queue_num].bytes -=
1655 tx_buff->skb->len;
1656 dev_kfree_skb_any(tx_buff->skb);
1657 tx_buff->skb = NULL;
1658 adapter->netdev->stats.tx_dropped++;
1659 }
1660 ind_bufp->index = 0;
1661 if (atomic_sub_return(entries, &tx_scrq->used) <=
1662 (adapter->req_tx_entries_per_subcrq / 2) &&
1663 __netif_subqueue_stopped(adapter->netdev, queue_num) &&
1664 !test_bit(0, &adapter->resetting)) {
1665 netif_wake_subqueue(adapter->netdev, queue_num);
1666 netdev_dbg(adapter->netdev, "Started queue %d\n",
1667 queue_num);
1668 }
1669 }
1670
1671 static int ibmvnic_tx_scrq_flush(struct ibmvnic_adapter *adapter,
1672 struct ibmvnic_sub_crq_queue *tx_scrq)
1673 {
1674 struct ibmvnic_ind_xmit_queue *ind_bufp;
1675 u64 dma_addr;
1676 u64 entries;
1677 u64 handle;
1678 int rc;
1679
1680 ind_bufp = &tx_scrq->ind_buf;
1681 dma_addr = (u64)ind_bufp->indir_dma;
1682 entries = (u64)ind_bufp->index;
1683 handle = tx_scrq->handle;
1684
1685 if (!entries)
1686 return 0;
1687 rc = send_subcrq_indirect(adapter, handle, dma_addr, entries);
1688 if (rc)
1689 ibmvnic_tx_scrq_clean_buffer(adapter, tx_scrq);
1690 else
1691 ind_bufp->index = 0;
1692 return 0;
1693 }
1694
1695 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
1696 {
1697 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1698 int queue_num = skb_get_queue_mapping(skb);
1699 u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req;
1700 struct device *dev = &adapter->vdev->dev;
1701 struct ibmvnic_ind_xmit_queue *ind_bufp;
1702 struct ibmvnic_tx_buff *tx_buff = NULL;
1703 struct ibmvnic_sub_crq_queue *tx_scrq;
1704 struct ibmvnic_tx_pool *tx_pool;
1705 unsigned int tx_send_failed = 0;
1706 netdev_tx_t ret = NETDEV_TX_OK;
1707 unsigned int tx_map_failed = 0;
1708 union sub_crq indir_arr[16];
1709 unsigned int tx_dropped = 0;
1710 unsigned int tx_packets = 0;
1711 unsigned int tx_bytes = 0;
1712 dma_addr_t data_dma_addr;
1713 struct netdev_queue *txq;
1714 unsigned long lpar_rc;
1715 union sub_crq tx_crq;
1716 unsigned int offset;
1717 int num_entries = 1;
1718 unsigned char *dst;
1719 int index = 0;
1720 u8 proto = 0;
1721
1722 tx_scrq = adapter->tx_scrq[queue_num];
1723 txq = netdev_get_tx_queue(netdev, queue_num);
1724 ind_bufp = &tx_scrq->ind_buf;
1725
1726 if (test_bit(0, &adapter->resetting)) {
1727 dev_kfree_skb_any(skb);
1728
1729 tx_send_failed++;
1730 tx_dropped++;
1731 ret = NETDEV_TX_OK;
1732 goto out;
1733 }
1734
1735 if (ibmvnic_xmit_workarounds(skb, netdev)) {
1736 tx_dropped++;
1737 tx_send_failed++;
1738 ret = NETDEV_TX_OK;
1739 ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1740 goto out;
1741 }
1742 if (skb_is_gso(skb))
1743 tx_pool = &adapter->tso_pool[queue_num];
1744 else
1745 tx_pool = &adapter->tx_pool[queue_num];
1746
1747 index = tx_pool->free_map[tx_pool->consumer_index];
1748
1749 if (index == IBMVNIC_INVALID_MAP) {
1750 dev_kfree_skb_any(skb);
1751 tx_send_failed++;
1752 tx_dropped++;
1753 ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1754 ret = NETDEV_TX_OK;
1755 goto out;
1756 }
1757
1758 tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP;
1759
1760 offset = index * tx_pool->buf_size;
1761 dst = tx_pool->long_term_buff.buff + offset;
1762 memset(dst, 0, tx_pool->buf_size);
1763 data_dma_addr = tx_pool->long_term_buff.addr + offset;
1764
1765 if (skb_shinfo(skb)->nr_frags) {
1766 int cur, i;
1767
1768 /* Copy the head */
1769 skb_copy_from_linear_data(skb, dst, skb_headlen(skb));
1770 cur = skb_headlen(skb);
1771
1772 /* Copy the frags */
1773 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1774 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1775
1776 memcpy(dst + cur, skb_frag_address(frag),
1777 skb_frag_size(frag));
1778 cur += skb_frag_size(frag);
1779 }
1780 } else {
1781 skb_copy_from_linear_data(skb, dst, skb->len);
1782 }
1783
1784 /* post changes to long_term_buff *dst before VIOS accessing it */
1785 dma_wmb();
1786
1787 tx_pool->consumer_index =
1788 (tx_pool->consumer_index + 1) % tx_pool->num_buffers;
1789
1790 tx_buff = &tx_pool->tx_buff[index];
1791 tx_buff->skb = skb;
1792 tx_buff->index = index;
1793 tx_buff->pool_index = queue_num;
1794
1795 memset(&tx_crq, 0, sizeof(tx_crq));
1796 tx_crq.v1.first = IBMVNIC_CRQ_CMD;
1797 tx_crq.v1.type = IBMVNIC_TX_DESC;
1798 tx_crq.v1.n_crq_elem = 1;
1799 tx_crq.v1.n_sge = 1;
1800 tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED;
1801
1802 if (skb_is_gso(skb))
1803 tx_crq.v1.correlator =
1804 cpu_to_be32(index | IBMVNIC_TSO_POOL_MASK);
1805 else
1806 tx_crq.v1.correlator = cpu_to_be32(index);
1807 tx_crq.v1.dma_reg = cpu_to_be16(tx_pool->long_term_buff.map_id);
1808 tx_crq.v1.sge_len = cpu_to_be32(skb->len);
1809 tx_crq.v1.ioba = cpu_to_be64(data_dma_addr);
1810
1811 if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) {
1812 tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT;
1813 tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci);
1814 }
1815
1816 if (skb->protocol == htons(ETH_P_IP)) {
1817 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4;
1818 proto = ip_hdr(skb)->protocol;
1819 } else if (skb->protocol == htons(ETH_P_IPV6)) {
1820 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6;
1821 proto = ipv6_hdr(skb)->nexthdr;
1822 }
1823
1824 if (proto == IPPROTO_TCP)
1825 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP;
1826 else if (proto == IPPROTO_UDP)
1827 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP;
1828
1829 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1830 tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD;
1831 hdrs += 2;
1832 }
1833 if (skb_is_gso(skb)) {
1834 tx_crq.v1.flags1 |= IBMVNIC_TX_LSO;
1835 tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size);
1836 hdrs += 2;
1837 }
1838
1839 if ((*hdrs >> 7) & 1)
1840 build_hdr_descs_arr(skb, indir_arr, &num_entries, *hdrs);
1841
1842 tx_crq.v1.n_crq_elem = num_entries;
1843 tx_buff->num_entries = num_entries;
1844 /* flush buffer if current entry can not fit */
1845 if (num_entries + ind_bufp->index > IBMVNIC_MAX_IND_DESCS) {
1846 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1847 if (lpar_rc != H_SUCCESS)
1848 goto tx_flush_err;
1849 }
1850
1851 indir_arr[0] = tx_crq;
1852 memcpy(&ind_bufp->indir_arr[ind_bufp->index], &indir_arr[0],
1853 num_entries * sizeof(struct ibmvnic_generic_scrq));
1854 ind_bufp->index += num_entries;
1855 if (__netdev_tx_sent_queue(txq, skb->len,
1856 netdev_xmit_more() &&
1857 ind_bufp->index < IBMVNIC_MAX_IND_DESCS)) {
1858 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1859 if (lpar_rc != H_SUCCESS)
1860 goto tx_err;
1861 }
1862
1863 if (atomic_add_return(num_entries, &tx_scrq->used)
1864 >= adapter->req_tx_entries_per_subcrq) {
1865 netdev_dbg(netdev, "Stopping queue %d\n", queue_num);
1866 netif_stop_subqueue(netdev, queue_num);
1867 }
1868
1869 tx_packets++;
1870 tx_bytes += skb->len;
1871 txq->trans_start = jiffies;
1872 ret = NETDEV_TX_OK;
1873 goto out;
1874
1875 tx_flush_err:
1876 dev_kfree_skb_any(skb);
1877 tx_buff->skb = NULL;
1878 tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
1879 tx_pool->num_buffers - 1 :
1880 tx_pool->consumer_index - 1;
1881 tx_dropped++;
1882 tx_err:
1883 if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER)
1884 dev_err_ratelimited(dev, "tx: send failed\n");
1885
1886 if (lpar_rc == H_CLOSED || adapter->failover_pending) {
1887 /* Disable TX and report carrier off if queue is closed
1888 * or pending failover.
1889 * Firmware guarantees that a signal will be sent to the
1890 * driver, triggering a reset or some other action.
1891 */
1892 netif_tx_stop_all_queues(netdev);
1893 netif_carrier_off(netdev);
1894 }
1895 out:
1896 netdev->stats.tx_dropped += tx_dropped;
1897 netdev->stats.tx_bytes += tx_bytes;
1898 netdev->stats.tx_packets += tx_packets;
1899 adapter->tx_send_failed += tx_send_failed;
1900 adapter->tx_map_failed += tx_map_failed;
1901 adapter->tx_stats_buffers[queue_num].packets += tx_packets;
1902 adapter->tx_stats_buffers[queue_num].bytes += tx_bytes;
1903 adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped;
1904
1905 return ret;
1906 }
1907
1908 static void ibmvnic_set_multi(struct net_device *netdev)
1909 {
1910 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1911 struct netdev_hw_addr *ha;
1912 union ibmvnic_crq crq;
1913
1914 memset(&crq, 0, sizeof(crq));
1915 crq.request_capability.first = IBMVNIC_CRQ_CMD;
1916 crq.request_capability.cmd = REQUEST_CAPABILITY;
1917
1918 if (netdev->flags & IFF_PROMISC) {
1919 if (!adapter->promisc_supported)
1920 return;
1921 } else {
1922 if (netdev->flags & IFF_ALLMULTI) {
1923 /* Accept all multicast */
1924 memset(&crq, 0, sizeof(crq));
1925 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1926 crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1927 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL;
1928 ibmvnic_send_crq(adapter, &crq);
1929 } else if (netdev_mc_empty(netdev)) {
1930 /* Reject all multicast */
1931 memset(&crq, 0, sizeof(crq));
1932 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1933 crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1934 crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL;
1935 ibmvnic_send_crq(adapter, &crq);
1936 } else {
1937 /* Accept one or more multicast(s) */
1938 netdev_for_each_mc_addr(ha, netdev) {
1939 memset(&crq, 0, sizeof(crq));
1940 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1941 crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1942 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC;
1943 ether_addr_copy(&crq.multicast_ctrl.mac_addr[0],
1944 ha->addr);
1945 ibmvnic_send_crq(adapter, &crq);
1946 }
1947 }
1948 }
1949 }
1950
1951 static int __ibmvnic_set_mac(struct net_device *netdev, u8 *dev_addr)
1952 {
1953 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1954 union ibmvnic_crq crq;
1955 int rc;
1956
1957 if (!is_valid_ether_addr(dev_addr)) {
1958 rc = -EADDRNOTAVAIL;
1959 goto err;
1960 }
1961
1962 memset(&crq, 0, sizeof(crq));
1963 crq.change_mac_addr.first = IBMVNIC_CRQ_CMD;
1964 crq.change_mac_addr.cmd = CHANGE_MAC_ADDR;
1965 ether_addr_copy(&crq.change_mac_addr.mac_addr[0], dev_addr);
1966
1967 mutex_lock(&adapter->fw_lock);
1968 adapter->fw_done_rc = 0;
1969 reinit_completion(&adapter->fw_done);
1970
1971 rc = ibmvnic_send_crq(adapter, &crq);
1972 if (rc) {
1973 rc = -EIO;
1974 mutex_unlock(&adapter->fw_lock);
1975 goto err;
1976 }
1977
1978 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1979 /* netdev->dev_addr is changed in handle_change_mac_rsp function */
1980 if (rc || adapter->fw_done_rc) {
1981 rc = -EIO;
1982 mutex_unlock(&adapter->fw_lock);
1983 goto err;
1984 }
1985 mutex_unlock(&adapter->fw_lock);
1986 return 0;
1987 err:
1988 ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
1989 return rc;
1990 }
1991
1992 static int ibmvnic_set_mac(struct net_device *netdev, void *p)
1993 {
1994 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1995 struct sockaddr *addr = p;
1996 int rc;
1997
1998 rc = 0;
1999 if (!is_valid_ether_addr(addr->sa_data))
2000 return -EADDRNOTAVAIL;
2001
2002 ether_addr_copy(adapter->mac_addr, addr->sa_data);
2003 if (adapter->state != VNIC_PROBED)
2004 rc = __ibmvnic_set_mac(netdev, addr->sa_data);
2005
2006 return rc;
2007 }
2008
2009 static const char *reset_reason_to_string(enum ibmvnic_reset_reason reason)
2010 {
2011 switch (reason) {
2012 case VNIC_RESET_FAILOVER:
2013 return "FAILOVER";
2014 case VNIC_RESET_MOBILITY:
2015 return "MOBILITY";
2016 case VNIC_RESET_FATAL:
2017 return "FATAL";
2018 case VNIC_RESET_NON_FATAL:
2019 return "NON_FATAL";
2020 case VNIC_RESET_TIMEOUT:
2021 return "TIMEOUT";
2022 case VNIC_RESET_CHANGE_PARAM:
2023 return "CHANGE_PARAM";
2024 case VNIC_RESET_PASSIVE_INIT:
2025 return "PASSIVE_INIT";
2026 }
2027 return "UNKNOWN";
2028 }
2029
2030 /*
2031 * do_reset returns zero if we are able to keep processing reset events, or
2032 * non-zero if we hit a fatal error and must halt.
2033 */
2034 static int do_reset(struct ibmvnic_adapter *adapter,
2035 struct ibmvnic_rwi *rwi, u32 reset_state)
2036 {
2037 u64 old_num_rx_queues, old_num_tx_queues;
2038 u64 old_num_rx_slots, old_num_tx_slots;
2039 struct net_device *netdev = adapter->netdev;
2040 int rc;
2041
2042 netdev_dbg(adapter->netdev,
2043 "[S:%s FOP:%d] Reset reason: %s, reset_state: %s\n",
2044 adapter_state_to_string(adapter->state),
2045 adapter->failover_pending,
2046 reset_reason_to_string(rwi->reset_reason),
2047 adapter_state_to_string(reset_state));
2048
2049 adapter->reset_reason = rwi->reset_reason;
2050 /* requestor of VNIC_RESET_CHANGE_PARAM already has the rtnl lock */
2051 if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
2052 rtnl_lock();
2053
2054 /* Now that we have the rtnl lock, clear any pending failover.
2055 * This will ensure ibmvnic_open() has either completed or will
2056 * block until failover is complete.
2057 */
2058 if (rwi->reset_reason == VNIC_RESET_FAILOVER)
2059 adapter->failover_pending = false;
2060
2061 /* read the state and check (again) after getting rtnl */
2062 reset_state = adapter->state;
2063
2064 if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
2065 rc = -EBUSY;
2066 goto out;
2067 }
2068
2069 netif_carrier_off(netdev);
2070
2071 old_num_rx_queues = adapter->req_rx_queues;
2072 old_num_tx_queues = adapter->req_tx_queues;
2073 old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq;
2074 old_num_tx_slots = adapter->req_tx_entries_per_subcrq;
2075
2076 ibmvnic_cleanup(netdev);
2077
2078 if (reset_state == VNIC_OPEN &&
2079 adapter->reset_reason != VNIC_RESET_MOBILITY &&
2080 adapter->reset_reason != VNIC_RESET_FAILOVER) {
2081 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2082 rc = __ibmvnic_close(netdev);
2083 if (rc)
2084 goto out;
2085 } else {
2086 adapter->state = VNIC_CLOSING;
2087
2088 /* Release the RTNL lock before link state change and
2089 * re-acquire after the link state change to allow
2090 * linkwatch_event to grab the RTNL lock and run during
2091 * a reset.
2092 */
2093 rtnl_unlock();
2094 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
2095 rtnl_lock();
2096 if (rc)
2097 goto out;
2098
2099 if (adapter->state == VNIC_OPEN) {
2100 /* When we dropped rtnl, ibmvnic_open() got
2101 * it and noticed that we are resetting and
2102 * set the adapter state to OPEN. Update our
2103 * new "target" state, and resume the reset
2104 * from VNIC_CLOSING state.
2105 */
2106 netdev_dbg(netdev,
2107 "Open changed state from %s, updating.\n",
2108 adapter_state_to_string(reset_state));
2109 reset_state = VNIC_OPEN;
2110 adapter->state = VNIC_CLOSING;
2111 }
2112
2113 if (adapter->state != VNIC_CLOSING) {
2114 /* If someone else changed the adapter state
2115 * when we dropped the rtnl, fail the reset
2116 */
2117 rc = -1;
2118 goto out;
2119 }
2120 adapter->state = VNIC_CLOSED;
2121 }
2122 }
2123
2124 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2125 release_resources(adapter);
2126 release_sub_crqs(adapter, 1);
2127 release_crq_queue(adapter);
2128 }
2129
2130 if (adapter->reset_reason != VNIC_RESET_NON_FATAL) {
2131 /* remove the closed state so when we call open it appears
2132 * we are coming from the probed state.
2133 */
2134 adapter->state = VNIC_PROBED;
2135
2136 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2137 rc = init_crq_queue(adapter);
2138 } else if (adapter->reset_reason == VNIC_RESET_MOBILITY) {
2139 rc = ibmvnic_reenable_crq_queue(adapter);
2140 release_sub_crqs(adapter, 1);
2141 } else {
2142 rc = ibmvnic_reset_crq(adapter);
2143 if (rc == H_CLOSED || rc == H_SUCCESS) {
2144 rc = vio_enable_interrupts(adapter->vdev);
2145 if (rc)
2146 netdev_err(adapter->netdev,
2147 "Reset failed to enable interrupts. rc=%d\n",
2148 rc);
2149 }
2150 }
2151
2152 if (rc) {
2153 netdev_err(adapter->netdev,
2154 "Reset couldn't initialize crq. rc=%d\n", rc);
2155 goto out;
2156 }
2157
2158 rc = ibmvnic_reset_init(adapter, true);
2159 if (rc) {
2160 rc = IBMVNIC_INIT_FAILED;
2161 goto out;
2162 }
2163
2164 /* If the adapter was in PROBE or DOWN state prior to the reset,
2165 * exit here.
2166 */
2167 if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN) {
2168 rc = 0;
2169 goto out;
2170 }
2171
2172 rc = ibmvnic_login(netdev);
2173 if (rc)
2174 goto out;
2175
2176 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2177 rc = init_resources(adapter);
2178 if (rc)
2179 goto out;
2180 } else if (adapter->req_rx_queues != old_num_rx_queues ||
2181 adapter->req_tx_queues != old_num_tx_queues ||
2182 adapter->req_rx_add_entries_per_subcrq !=
2183 old_num_rx_slots ||
2184 adapter->req_tx_entries_per_subcrq !=
2185 old_num_tx_slots ||
2186 !adapter->rx_pool ||
2187 !adapter->tso_pool ||
2188 !adapter->tx_pool) {
2189 release_rx_pools(adapter);
2190 release_tx_pools(adapter);
2191 release_napi(adapter);
2192 release_vpd_data(adapter);
2193
2194 rc = init_resources(adapter);
2195 if (rc)
2196 goto out;
2197
2198 } else {
2199 rc = reset_tx_pools(adapter);
2200 if (rc) {
2201 netdev_dbg(adapter->netdev, "reset tx pools failed (%d)\n",
2202 rc);
2203 goto out;
2204 }
2205
2206 rc = reset_rx_pools(adapter);
2207 if (rc) {
2208 netdev_dbg(adapter->netdev, "reset rx pools failed (%d)\n",
2209 rc);
2210 goto out;
2211 }
2212 }
2213 ibmvnic_disable_irqs(adapter);
2214 }
2215 adapter->state = VNIC_CLOSED;
2216
2217 if (reset_state == VNIC_CLOSED) {
2218 rc = 0;
2219 goto out;
2220 }
2221
2222 rc = __ibmvnic_open(netdev);
2223 if (rc) {
2224 rc = IBMVNIC_OPEN_FAILED;
2225 goto out;
2226 }
2227
2228 /* refresh device's multicast list */
2229 ibmvnic_set_multi(netdev);
2230
2231 if (adapter->reset_reason == VNIC_RESET_FAILOVER ||
2232 adapter->reset_reason == VNIC_RESET_MOBILITY)
2233 __netdev_notify_peers(netdev);
2234
2235 rc = 0;
2236
2237 out:
2238 /* restore the adapter state if reset failed */
2239 if (rc)
2240 adapter->state = reset_state;
2241 /* requestor of VNIC_RESET_CHANGE_PARAM should still hold the rtnl lock */
2242 if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
2243 rtnl_unlock();
2244
2245 netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Reset done, rc %d\n",
2246 adapter_state_to_string(adapter->state),
2247 adapter->failover_pending, rc);
2248 return rc;
2249 }
2250
2251 static int do_hard_reset(struct ibmvnic_adapter *adapter,
2252 struct ibmvnic_rwi *rwi, u32 reset_state)
2253 {
2254 struct net_device *netdev = adapter->netdev;
2255 int rc;
2256
2257 netdev_dbg(adapter->netdev, "Hard resetting driver (%s)\n",
2258 reset_reason_to_string(rwi->reset_reason));
2259
2260 /* read the state and check (again) after getting rtnl */
2261 reset_state = adapter->state;
2262
2263 if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
2264 rc = -EBUSY;
2265 goto out;
2266 }
2267
2268 netif_carrier_off(netdev);
2269 adapter->reset_reason = rwi->reset_reason;
2270
2271 ibmvnic_cleanup(netdev);
2272 release_resources(adapter);
2273 release_sub_crqs(adapter, 0);
2274 release_crq_queue(adapter);
2275
2276 /* remove the closed state so when we call open it appears
2277 * we are coming from the probed state.
2278 */
2279 adapter->state = VNIC_PROBED;
2280
2281 reinit_completion(&adapter->init_done);
2282 rc = init_crq_queue(adapter);
2283 if (rc) {
2284 netdev_err(adapter->netdev,
2285 "Couldn't initialize crq. rc=%d\n", rc);
2286 goto out;
2287 }
2288
2289 rc = ibmvnic_reset_init(adapter, false);
2290 if (rc)
2291 goto out;
2292
2293 /* If the adapter was in PROBE or DOWN state prior to the reset,
2294 * exit here.
2295 */
2296 if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN)
2297 goto out;
2298
2299 rc = ibmvnic_login(netdev);
2300 if (rc)
2301 goto out;
2302
2303 rc = init_resources(adapter);
2304 if (rc)
2305 goto out;
2306
2307 ibmvnic_disable_irqs(adapter);
2308 adapter->state = VNIC_CLOSED;
2309
2310 if (reset_state == VNIC_CLOSED)
2311 goto out;
2312
2313 rc = __ibmvnic_open(netdev);
2314 if (rc) {
2315 rc = IBMVNIC_OPEN_FAILED;
2316 goto out;
2317 }
2318
2319 __netdev_notify_peers(netdev);
2320 out:
2321 /* restore adapter state if reset failed */
2322 if (rc)
2323 adapter->state = reset_state;
2324 netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Hard reset done, rc %d\n",
2325 adapter_state_to_string(adapter->state),
2326 adapter->failover_pending, rc);
2327 return rc;
2328 }
2329
2330 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter)
2331 {
2332 struct ibmvnic_rwi *rwi;
2333 unsigned long flags;
2334
2335 spin_lock_irqsave(&adapter->rwi_lock, flags);
2336
2337 if (!list_empty(&adapter->rwi_list)) {
2338 rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi,
2339 list);
2340 list_del(&rwi->list);
2341 } else {
2342 rwi = NULL;
2343 }
2344
2345 spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2346 return rwi;
2347 }
2348
2349 /**
2350 * do_passive_init - complete probing when partner device is detected.
2351 * @adapter: ibmvnic_adapter struct
2352 *
2353 * If the ibmvnic device does not have a partner device to communicate with at boot
2354 * and that partner device comes online at a later time, this function is called
2355 * to complete the initialization process of ibmvnic device.
2356 * Caller is expected to hold rtnl_lock().
2357 *
2358 * Returns non-zero if sub-CRQs are not initialized properly leaving the device
2359 * in the down state.
2360 * Returns 0 upon success and the device is in PROBED state.
2361 */
2362
2363 static int do_passive_init(struct ibmvnic_adapter *adapter)
2364 {
2365 unsigned long timeout = msecs_to_jiffies(30000);
2366 struct net_device *netdev = adapter->netdev;
2367 struct device *dev = &adapter->vdev->dev;
2368 int rc;
2369
2370 netdev_dbg(netdev, "Partner device found, probing.\n");
2371
2372 adapter->state = VNIC_PROBING;
2373 reinit_completion(&adapter->init_done);
2374 adapter->init_done_rc = 0;
2375 adapter->crq.active = true;
2376
2377 rc = send_crq_init_complete(adapter);
2378 if (rc)
2379 goto out;
2380
2381 rc = send_version_xchg(adapter);
2382 if (rc)
2383 netdev_dbg(adapter->netdev, "send_version_xchg failed, rc=%d\n", rc);
2384
2385 if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
2386 dev_err(dev, "Initialization sequence timed out\n");
2387 rc = -ETIMEDOUT;
2388 goto out;
2389 }
2390
2391 rc = init_sub_crqs(adapter);
2392 if (rc) {
2393 dev_err(dev, "Initialization of sub crqs failed, rc=%d\n", rc);
2394 goto out;
2395 }
2396
2397 rc = init_sub_crq_irqs(adapter);
2398 if (rc) {
2399 dev_err(dev, "Failed to initialize sub crq irqs\n, rc=%d", rc);
2400 goto init_failed;
2401 }
2402
2403 netdev->mtu = adapter->req_mtu - ETH_HLEN;
2404 netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
2405 netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
2406
2407 adapter->state = VNIC_PROBED;
2408 netdev_dbg(netdev, "Probed successfully. Waiting for signal from partner device.\n");
2409
2410 return 0;
2411
2412 init_failed:
2413 release_sub_crqs(adapter, 1);
2414 out:
2415 adapter->state = VNIC_DOWN;
2416 return rc;
2417 }
2418
2419 static void __ibmvnic_reset(struct work_struct *work)
2420 {
2421 struct ibmvnic_adapter *adapter;
2422 bool saved_state = false;
2423 struct ibmvnic_rwi *tmprwi;
2424 struct ibmvnic_rwi *rwi;
2425 unsigned long flags;
2426 u32 reset_state;
2427 int num_fails = 0;
2428 int rc = 0;
2429
2430 adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset);
2431
2432 if (test_and_set_bit_lock(0, &adapter->resetting)) {
2433 queue_delayed_work(system_long_wq,
2434 &adapter->ibmvnic_delayed_reset,
2435 IBMVNIC_RESET_DELAY);
2436 return;
2437 }
2438
2439 rwi = get_next_rwi(adapter);
2440 while (rwi) {
2441 spin_lock_irqsave(&adapter->state_lock, flags);
2442
2443 if (adapter->state == VNIC_REMOVING ||
2444 adapter->state == VNIC_REMOVED) {
2445 spin_unlock_irqrestore(&adapter->state_lock, flags);
2446 kfree(rwi);
2447 rc = EBUSY;
2448 break;
2449 }
2450
2451 if (!saved_state) {
2452 reset_state = adapter->state;
2453 saved_state = true;
2454 }
2455 spin_unlock_irqrestore(&adapter->state_lock, flags);
2456
2457 if (rwi->reset_reason == VNIC_RESET_PASSIVE_INIT) {
2458 rtnl_lock();
2459 rc = do_passive_init(adapter);
2460 rtnl_unlock();
2461 if (!rc)
2462 netif_carrier_on(adapter->netdev);
2463 } else if (adapter->force_reset_recovery) {
2464 /* Since we are doing a hard reset now, clear the
2465 * failover_pending flag so we don't ignore any
2466 * future MOBILITY or other resets.
2467 */
2468 adapter->failover_pending = false;
2469
2470 /* Transport event occurred during previous reset */
2471 if (adapter->wait_for_reset) {
2472 /* Previous was CHANGE_PARAM; caller locked */
2473 adapter->force_reset_recovery = false;
2474 rc = do_hard_reset(adapter, rwi, reset_state);
2475 } else {
2476 rtnl_lock();
2477 adapter->force_reset_recovery = false;
2478 rc = do_hard_reset(adapter, rwi, reset_state);
2479 rtnl_unlock();
2480 }
2481 if (rc)
2482 num_fails++;
2483 else
2484 num_fails = 0;
2485
2486 /* If auto-priority-failover is enabled we can get
2487 * back to back failovers during resets, resulting
2488 * in at least two failed resets (from high-priority
2489 * backing device to low-priority one and then back)
2490 * If resets continue to fail beyond that, give the
2491 * adapter some time to settle down before retrying.
2492 */
2493 if (num_fails >= 3) {
2494 netdev_dbg(adapter->netdev,
2495 "[S:%s] Hard reset failed %d times, waiting 60 secs\n",
2496 adapter_state_to_string(adapter->state),
2497 num_fails);
2498 set_current_state(TASK_UNINTERRUPTIBLE);
2499 schedule_timeout(60 * HZ);
2500 }
2501 } else {
2502 rc = do_reset(adapter, rwi, reset_state);
2503 }
2504 tmprwi = rwi;
2505 adapter->last_reset_time = jiffies;
2506
2507 if (rc)
2508 netdev_dbg(adapter->netdev, "Reset failed, rc=%d\n", rc);
2509
2510 rwi = get_next_rwi(adapter);
2511
2512 /*
2513 * If there is another reset queued, free the previous rwi
2514 * and process the new reset even if previous reset failed
2515 * (the previous reset could have failed because of a fail
2516 * over for instance, so process the fail over).
2517 *
2518 * If there are no resets queued and the previous reset failed,
2519 * the adapter would be in an undefined state. So retry the
2520 * previous reset as a hard reset.
2521 */
2522 if (rwi)
2523 kfree(tmprwi);
2524 else if (rc)
2525 rwi = tmprwi;
2526
2527 if (rwi && (rwi->reset_reason == VNIC_RESET_FAILOVER ||
2528 rwi->reset_reason == VNIC_RESET_MOBILITY || rc))
2529 adapter->force_reset_recovery = true;
2530 }
2531
2532 if (adapter->wait_for_reset) {
2533 adapter->reset_done_rc = rc;
2534 complete(&adapter->reset_done);
2535 }
2536
2537 clear_bit_unlock(0, &adapter->resetting);
2538
2539 netdev_dbg(adapter->netdev,
2540 "[S:%s FRR:%d WFR:%d] Done processing resets\n",
2541 adapter_state_to_string(adapter->state),
2542 adapter->force_reset_recovery,
2543 adapter->wait_for_reset);
2544 }
2545
2546 static void __ibmvnic_delayed_reset(struct work_struct *work)
2547 {
2548 struct ibmvnic_adapter *adapter;
2549
2550 adapter = container_of(work, struct ibmvnic_adapter,
2551 ibmvnic_delayed_reset.work);
2552 __ibmvnic_reset(&adapter->ibmvnic_reset);
2553 }
2554
2555 static int ibmvnic_reset(struct ibmvnic_adapter *adapter,
2556 enum ibmvnic_reset_reason reason)
2557 {
2558 struct list_head *entry, *tmp_entry;
2559 struct ibmvnic_rwi *rwi, *tmp;
2560 struct net_device *netdev = adapter->netdev;
2561 unsigned long flags;
2562 int ret;
2563
2564 spin_lock_irqsave(&adapter->rwi_lock, flags);
2565
2566 /* If failover is pending don't schedule any other reset.
2567 * Instead let the failover complete. If there is already a
2568 * a failover reset scheduled, we will detect and drop the
2569 * duplicate reset when walking the ->rwi_list below.
2570 */
2571 if (adapter->state == VNIC_REMOVING ||
2572 adapter->state == VNIC_REMOVED ||
2573 (adapter->failover_pending && reason != VNIC_RESET_FAILOVER)) {
2574 ret = EBUSY;
2575 netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n");
2576 goto err;
2577 }
2578
2579 if (adapter->state == VNIC_PROBING) {
2580 netdev_warn(netdev, "Adapter reset during probe\n");
2581 adapter->init_done_rc = -EAGAIN;
2582 ret = EAGAIN;
2583 goto err;
2584 }
2585
2586 list_for_each_entry(tmp, &adapter->rwi_list, list) {
2587 if (tmp->reset_reason == reason) {
2588 netdev_dbg(netdev, "Skipping matching reset, reason=%s\n",
2589 reset_reason_to_string(reason));
2590 ret = EBUSY;
2591 goto err;
2592 }
2593 }
2594
2595 rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC);
2596 if (!rwi) {
2597 ret = ENOMEM;
2598 goto err;
2599 }
2600 /* if we just received a transport event,
2601 * flush reset queue and process this reset
2602 */
2603 if (adapter->force_reset_recovery && !list_empty(&adapter->rwi_list)) {
2604 list_for_each_safe(entry, tmp_entry, &adapter->rwi_list)
2605 list_del(entry);
2606 }
2607 rwi->reset_reason = reason;
2608 list_add_tail(&rwi->list, &adapter->rwi_list);
2609 netdev_dbg(adapter->netdev, "Scheduling reset (reason %s)\n",
2610 reset_reason_to_string(reason));
2611 queue_work(system_long_wq, &adapter->ibmvnic_reset);
2612
2613 ret = 0;
2614 err:
2615 /* ibmvnic_close() below can block, so drop the lock first */
2616 spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2617
2618 if (ret == ENOMEM)
2619 ibmvnic_close(netdev);
2620
2621 return -ret;
2622 }
2623
2624 static void ibmvnic_tx_timeout(struct net_device *dev, unsigned int txqueue)
2625 {
2626 struct ibmvnic_adapter *adapter = netdev_priv(dev);
2627
2628 if (test_bit(0, &adapter->resetting)) {
2629 netdev_err(adapter->netdev,
2630 "Adapter is resetting, skip timeout reset\n");
2631 return;
2632 }
2633 /* No queuing up reset until at least 5 seconds (default watchdog val)
2634 * after last reset
2635 */
2636 if (time_before(jiffies, (adapter->last_reset_time + dev->watchdog_timeo))) {
2637 netdev_dbg(dev, "Not yet time to tx timeout.\n");
2638 return;
2639 }
2640 ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT);
2641 }
2642
2643 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter,
2644 struct ibmvnic_rx_buff *rx_buff)
2645 {
2646 struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index];
2647
2648 rx_buff->skb = NULL;
2649
2650 pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff);
2651 pool->next_alloc = (pool->next_alloc + 1) % pool->size;
2652
2653 atomic_dec(&pool->available);
2654 }
2655
2656 static int ibmvnic_poll(struct napi_struct *napi, int budget)
2657 {
2658 struct ibmvnic_sub_crq_queue *rx_scrq;
2659 struct ibmvnic_adapter *adapter;
2660 struct net_device *netdev;
2661 int frames_processed;
2662 int scrq_num;
2663
2664 netdev = napi->dev;
2665 adapter = netdev_priv(netdev);
2666 scrq_num = (int)(napi - adapter->napi);
2667 frames_processed = 0;
2668 rx_scrq = adapter->rx_scrq[scrq_num];
2669
2670 restart_poll:
2671 while (frames_processed < budget) {
2672 struct sk_buff *skb;
2673 struct ibmvnic_rx_buff *rx_buff;
2674 union sub_crq *next;
2675 u32 length;
2676 u16 offset;
2677 u8 flags = 0;
2678
2679 if (unlikely(test_bit(0, &adapter->resetting) &&
2680 adapter->reset_reason != VNIC_RESET_NON_FATAL)) {
2681 enable_scrq_irq(adapter, rx_scrq);
2682 napi_complete_done(napi, frames_processed);
2683 return frames_processed;
2684 }
2685
2686 if (!pending_scrq(adapter, rx_scrq))
2687 break;
2688 next = ibmvnic_next_scrq(adapter, rx_scrq);
2689 rx_buff = (struct ibmvnic_rx_buff *)
2690 be64_to_cpu(next->rx_comp.correlator);
2691 /* do error checking */
2692 if (next->rx_comp.rc) {
2693 netdev_dbg(netdev, "rx buffer returned with rc %x\n",
2694 be16_to_cpu(next->rx_comp.rc));
2695 /* free the entry */
2696 next->rx_comp.first = 0;
2697 dev_kfree_skb_any(rx_buff->skb);
2698 remove_buff_from_pool(adapter, rx_buff);
2699 continue;
2700 } else if (!rx_buff->skb) {
2701 /* free the entry */
2702 next->rx_comp.first = 0;
2703 remove_buff_from_pool(adapter, rx_buff);
2704 continue;
2705 }
2706
2707 length = be32_to_cpu(next->rx_comp.len);
2708 offset = be16_to_cpu(next->rx_comp.off_frame_data);
2709 flags = next->rx_comp.flags;
2710 skb = rx_buff->skb;
2711 /* load long_term_buff before copying to skb */
2712 dma_rmb();
2713 skb_copy_to_linear_data(skb, rx_buff->data + offset,
2714 length);
2715
2716 /* VLAN Header has been stripped by the system firmware and
2717 * needs to be inserted by the driver
2718 */
2719 if (adapter->rx_vlan_header_insertion &&
2720 (flags & IBMVNIC_VLAN_STRIPPED))
2721 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
2722 ntohs(next->rx_comp.vlan_tci));
2723
2724 /* free the entry */
2725 next->rx_comp.first = 0;
2726 remove_buff_from_pool(adapter, rx_buff);
2727
2728 skb_put(skb, length);
2729 skb->protocol = eth_type_trans(skb, netdev);
2730 skb_record_rx_queue(skb, scrq_num);
2731
2732 if (flags & IBMVNIC_IP_CHKSUM_GOOD &&
2733 flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) {
2734 skb->ip_summed = CHECKSUM_UNNECESSARY;
2735 }
2736
2737 length = skb->len;
2738 napi_gro_receive(napi, skb); /* send it up */
2739 netdev->stats.rx_packets++;
2740 netdev->stats.rx_bytes += length;
2741 adapter->rx_stats_buffers[scrq_num].packets++;
2742 adapter->rx_stats_buffers[scrq_num].bytes += length;
2743 frames_processed++;
2744 }
2745
2746 if (adapter->state != VNIC_CLOSING &&
2747 ((atomic_read(&adapter->rx_pool[scrq_num].available) <
2748 adapter->req_rx_add_entries_per_subcrq / 2) ||
2749 frames_processed < budget))
2750 replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]);
2751 if (frames_processed < budget) {
2752 if (napi_complete_done(napi, frames_processed)) {
2753 enable_scrq_irq(adapter, rx_scrq);
2754 if (pending_scrq(adapter, rx_scrq)) {
2755 if (napi_reschedule(napi)) {
2756 disable_scrq_irq(adapter, rx_scrq);
2757 goto restart_poll;
2758 }
2759 }
2760 }
2761 }
2762 return frames_processed;
2763 }
2764
2765 static int wait_for_reset(struct ibmvnic_adapter *adapter)
2766 {
2767 int rc, ret;
2768
2769 adapter->fallback.mtu = adapter->req_mtu;
2770 adapter->fallback.rx_queues = adapter->req_rx_queues;
2771 adapter->fallback.tx_queues = adapter->req_tx_queues;
2772 adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq;
2773 adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq;
2774
2775 reinit_completion(&adapter->reset_done);
2776 adapter->wait_for_reset = true;
2777 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2778
2779 if (rc) {
2780 ret = rc;
2781 goto out;
2782 }
2783 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 60000);
2784 if (rc) {
2785 ret = -ENODEV;
2786 goto out;
2787 }
2788
2789 ret = 0;
2790 if (adapter->reset_done_rc) {
2791 ret = -EIO;
2792 adapter->desired.mtu = adapter->fallback.mtu;
2793 adapter->desired.rx_queues = adapter->fallback.rx_queues;
2794 adapter->desired.tx_queues = adapter->fallback.tx_queues;
2795 adapter->desired.rx_entries = adapter->fallback.rx_entries;
2796 adapter->desired.tx_entries = adapter->fallback.tx_entries;
2797
2798 reinit_completion(&adapter->reset_done);
2799 adapter->wait_for_reset = true;
2800 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2801 if (rc) {
2802 ret = rc;
2803 goto out;
2804 }
2805 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done,
2806 60000);
2807 if (rc) {
2808 ret = -ENODEV;
2809 goto out;
2810 }
2811 }
2812 out:
2813 adapter->wait_for_reset = false;
2814
2815 return ret;
2816 }
2817
2818 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu)
2819 {
2820 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2821
2822 adapter->desired.mtu = new_mtu + ETH_HLEN;
2823
2824 return wait_for_reset(adapter);
2825 }
2826
2827 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb,
2828 struct net_device *dev,
2829 netdev_features_t features)
2830 {
2831 /* Some backing hardware adapters can not
2832 * handle packets with a MSS less than 224
2833 * or with only one segment.
2834 */
2835 if (skb_is_gso(skb)) {
2836 if (skb_shinfo(skb)->gso_size < 224 ||
2837 skb_shinfo(skb)->gso_segs == 1)
2838 features &= ~NETIF_F_GSO_MASK;
2839 }
2840
2841 return features;
2842 }
2843
2844 static const struct net_device_ops ibmvnic_netdev_ops = {
2845 .ndo_open = ibmvnic_open,
2846 .ndo_stop = ibmvnic_close,
2847 .ndo_start_xmit = ibmvnic_xmit,
2848 .ndo_set_rx_mode = ibmvnic_set_multi,
2849 .ndo_set_mac_address = ibmvnic_set_mac,
2850 .ndo_validate_addr = eth_validate_addr,
2851 .ndo_tx_timeout = ibmvnic_tx_timeout,
2852 .ndo_change_mtu = ibmvnic_change_mtu,
2853 .ndo_features_check = ibmvnic_features_check,
2854 };
2855
2856 /* ethtool functions */
2857
2858 static int ibmvnic_get_link_ksettings(struct net_device *netdev,
2859 struct ethtool_link_ksettings *cmd)
2860 {
2861 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2862 int rc;
2863
2864 rc = send_query_phys_parms(adapter);
2865 if (rc) {
2866 adapter->speed = SPEED_UNKNOWN;
2867 adapter->duplex = DUPLEX_UNKNOWN;
2868 }
2869 cmd->base.speed = adapter->speed;
2870 cmd->base.duplex = adapter->duplex;
2871 cmd->base.port = PORT_FIBRE;
2872 cmd->base.phy_address = 0;
2873 cmd->base.autoneg = AUTONEG_ENABLE;
2874
2875 return 0;
2876 }
2877
2878 static void ibmvnic_get_drvinfo(struct net_device *netdev,
2879 struct ethtool_drvinfo *info)
2880 {
2881 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2882
2883 strscpy(info->driver, ibmvnic_driver_name, sizeof(info->driver));
2884 strscpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version));
2885 strscpy(info->fw_version, adapter->fw_version,
2886 sizeof(info->fw_version));
2887 }
2888
2889 static u32 ibmvnic_get_msglevel(struct net_device *netdev)
2890 {
2891 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2892
2893 return adapter->msg_enable;
2894 }
2895
2896 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data)
2897 {
2898 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2899
2900 adapter->msg_enable = data;
2901 }
2902
2903 static u32 ibmvnic_get_link(struct net_device *netdev)
2904 {
2905 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2906
2907 /* Don't need to send a query because we request a logical link up at
2908 * init and then we wait for link state indications
2909 */
2910 return adapter->logical_link_state;
2911 }
2912
2913 static void ibmvnic_get_ringparam(struct net_device *netdev,
2914 struct ethtool_ringparam *ring)
2915 {
2916 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2917
2918 if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
2919 ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq;
2920 ring->tx_max_pending = adapter->max_tx_entries_per_subcrq;
2921 } else {
2922 ring->rx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
2923 ring->tx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
2924 }
2925 ring->rx_mini_max_pending = 0;
2926 ring->rx_jumbo_max_pending = 0;
2927 ring->rx_pending = adapter->req_rx_add_entries_per_subcrq;
2928 ring->tx_pending = adapter->req_tx_entries_per_subcrq;
2929 ring->rx_mini_pending = 0;
2930 ring->rx_jumbo_pending = 0;
2931 }
2932
2933 static int ibmvnic_set_ringparam(struct net_device *netdev,
2934 struct ethtool_ringparam *ring)
2935 {
2936 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2937 int ret;
2938
2939 ret = 0;
2940 adapter->desired.rx_entries = ring->rx_pending;
2941 adapter->desired.tx_entries = ring->tx_pending;
2942
2943 ret = wait_for_reset(adapter);
2944
2945 if (!ret &&
2946 (adapter->req_rx_add_entries_per_subcrq != ring->rx_pending ||
2947 adapter->req_tx_entries_per_subcrq != ring->tx_pending))
2948 netdev_info(netdev,
2949 "Could not match full ringsize request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
2950 ring->rx_pending, ring->tx_pending,
2951 adapter->req_rx_add_entries_per_subcrq,
2952 adapter->req_tx_entries_per_subcrq);
2953 return ret;
2954 }
2955
2956 static void ibmvnic_get_channels(struct net_device *netdev,
2957 struct ethtool_channels *channels)
2958 {
2959 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2960
2961 if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
2962 channels->max_rx = adapter->max_rx_queues;
2963 channels->max_tx = adapter->max_tx_queues;
2964 } else {
2965 channels->max_rx = IBMVNIC_MAX_QUEUES;
2966 channels->max_tx = IBMVNIC_MAX_QUEUES;
2967 }
2968
2969 channels->max_other = 0;
2970 channels->max_combined = 0;
2971 channels->rx_count = adapter->req_rx_queues;
2972 channels->tx_count = adapter->req_tx_queues;
2973 channels->other_count = 0;
2974 channels->combined_count = 0;
2975 }
2976
2977 static int ibmvnic_set_channels(struct net_device *netdev,
2978 struct ethtool_channels *channels)
2979 {
2980 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2981 int ret;
2982
2983 ret = 0;
2984 adapter->desired.rx_queues = channels->rx_count;
2985 adapter->desired.tx_queues = channels->tx_count;
2986
2987 ret = wait_for_reset(adapter);
2988
2989 if (!ret &&
2990 (adapter->req_rx_queues != channels->rx_count ||
2991 adapter->req_tx_queues != channels->tx_count))
2992 netdev_info(netdev,
2993 "Could not match full channels request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
2994 channels->rx_count, channels->tx_count,
2995 adapter->req_rx_queues, adapter->req_tx_queues);
2996 return ret;
2997 }
2998
2999 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data)
3000 {
3001 struct ibmvnic_adapter *adapter = netdev_priv(dev);
3002 int i;
3003
3004 switch (stringset) {
3005 case ETH_SS_STATS:
3006 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats);
3007 i++, data += ETH_GSTRING_LEN)
3008 memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN);
3009
3010 for (i = 0; i < adapter->req_tx_queues; i++) {
3011 snprintf(data, ETH_GSTRING_LEN, "tx%d_packets", i);
3012 data += ETH_GSTRING_LEN;
3013
3014 snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i);
3015 data += ETH_GSTRING_LEN;
3016
3017 snprintf(data, ETH_GSTRING_LEN,
3018 "tx%d_dropped_packets", i);
3019 data += ETH_GSTRING_LEN;
3020 }
3021
3022 for (i = 0; i < adapter->req_rx_queues; i++) {
3023 snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i);
3024 data += ETH_GSTRING_LEN;
3025
3026 snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i);
3027 data += ETH_GSTRING_LEN;
3028
3029 snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i);
3030 data += ETH_GSTRING_LEN;
3031 }
3032 break;
3033
3034 case ETH_SS_PRIV_FLAGS:
3035 for (i = 0; i < ARRAY_SIZE(ibmvnic_priv_flags); i++)
3036 strcpy(data + i * ETH_GSTRING_LEN,
3037 ibmvnic_priv_flags[i]);
3038 break;
3039 default:
3040 return;
3041 }
3042 }
3043
3044 static int ibmvnic_get_sset_count(struct net_device *dev, int sset)
3045 {
3046 struct ibmvnic_adapter *adapter = netdev_priv(dev);
3047
3048 switch (sset) {
3049 case ETH_SS_STATS:
3050 return ARRAY_SIZE(ibmvnic_stats) +
3051 adapter->req_tx_queues * NUM_TX_STATS +
3052 adapter->req_rx_queues * NUM_RX_STATS;
3053 case ETH_SS_PRIV_FLAGS:
3054 return ARRAY_SIZE(ibmvnic_priv_flags);
3055 default:
3056 return -EOPNOTSUPP;
3057 }
3058 }
3059
3060 static void ibmvnic_get_ethtool_stats(struct net_device *dev,
3061 struct ethtool_stats *stats, u64 *data)
3062 {
3063 struct ibmvnic_adapter *adapter = netdev_priv(dev);
3064 union ibmvnic_crq crq;
3065 int i, j;
3066 int rc;
3067
3068 memset(&crq, 0, sizeof(crq));
3069 crq.request_statistics.first = IBMVNIC_CRQ_CMD;
3070 crq.request_statistics.cmd = REQUEST_STATISTICS;
3071 crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token);
3072 crq.request_statistics.len =
3073 cpu_to_be32(sizeof(struct ibmvnic_statistics));
3074
3075 /* Wait for data to be written */
3076 reinit_completion(&adapter->stats_done);
3077 rc = ibmvnic_send_crq(adapter, &crq);
3078 if (rc)
3079 return;
3080 rc = ibmvnic_wait_for_completion(adapter, &adapter->stats_done, 10000);
3081 if (rc)
3082 return;
3083
3084 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++)
3085 data[i] = be64_to_cpu(IBMVNIC_GET_STAT
3086 (adapter, ibmvnic_stats[i].offset));
3087
3088 for (j = 0; j < adapter->req_tx_queues; j++) {
3089 data[i] = adapter->tx_stats_buffers[j].packets;
3090 i++;
3091 data[i] = adapter->tx_stats_buffers[j].bytes;
3092 i++;
3093 data[i] = adapter->tx_stats_buffers[j].dropped_packets;
3094 i++;
3095 }
3096
3097 for (j = 0; j < adapter->req_rx_queues; j++) {
3098 data[i] = adapter->rx_stats_buffers[j].packets;
3099 i++;
3100 data[i] = adapter->rx_stats_buffers[j].bytes;
3101 i++;
3102 data[i] = adapter->rx_stats_buffers[j].interrupts;
3103 i++;
3104 }
3105 }
3106
3107 static u32 ibmvnic_get_priv_flags(struct net_device *netdev)
3108 {
3109 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3110
3111 return adapter->priv_flags;
3112 }
3113
3114 static int ibmvnic_set_priv_flags(struct net_device *netdev, u32 flags)
3115 {
3116 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3117 bool which_maxes = !!(flags & IBMVNIC_USE_SERVER_MAXES);
3118
3119 if (which_maxes)
3120 adapter->priv_flags |= IBMVNIC_USE_SERVER_MAXES;
3121 else
3122 adapter->priv_flags &= ~IBMVNIC_USE_SERVER_MAXES;
3123
3124 return 0;
3125 }
3126
3127 static const struct ethtool_ops ibmvnic_ethtool_ops = {
3128 .get_drvinfo = ibmvnic_get_drvinfo,
3129 .get_msglevel = ibmvnic_get_msglevel,
3130 .set_msglevel = ibmvnic_set_msglevel,
3131 .get_link = ibmvnic_get_link,
3132 .get_ringparam = ibmvnic_get_ringparam,
3133 .set_ringparam = ibmvnic_set_ringparam,
3134 .get_channels = ibmvnic_get_channels,
3135 .set_channels = ibmvnic_set_channels,
3136 .get_strings = ibmvnic_get_strings,
3137 .get_sset_count = ibmvnic_get_sset_count,
3138 .get_ethtool_stats = ibmvnic_get_ethtool_stats,
3139 .get_link_ksettings = ibmvnic_get_link_ksettings,
3140 .get_priv_flags = ibmvnic_get_priv_flags,
3141 .set_priv_flags = ibmvnic_set_priv_flags,
3142 };
3143
3144 /* Routines for managing CRQs/sCRQs */
3145
3146 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter,
3147 struct ibmvnic_sub_crq_queue *scrq)
3148 {
3149 int rc;
3150
3151 if (!scrq) {
3152 netdev_dbg(adapter->netdev, "Invalid scrq reset.\n");
3153 return -EINVAL;
3154 }
3155
3156 if (scrq->irq) {
3157 free_irq(scrq->irq, scrq);
3158 irq_dispose_mapping(scrq->irq);
3159 scrq->irq = 0;
3160 }
3161
3162 if (scrq->msgs) {
3163 memset(scrq->msgs, 0, 4 * PAGE_SIZE);
3164 atomic_set(&scrq->used, 0);
3165 scrq->cur = 0;
3166 scrq->ind_buf.index = 0;
3167 } else {
3168 netdev_dbg(adapter->netdev, "Invalid scrq reset\n");
3169 return -EINVAL;
3170 }
3171
3172 rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
3173 4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
3174 return rc;
3175 }
3176
3177 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter)
3178 {
3179 int i, rc;
3180
3181 if (!adapter->tx_scrq || !adapter->rx_scrq)
3182 return -EINVAL;
3183
3184 for (i = 0; i < adapter->req_tx_queues; i++) {
3185 netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i);
3186 rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]);
3187 if (rc)
3188 return rc;
3189 }
3190
3191 for (i = 0; i < adapter->req_rx_queues; i++) {
3192 netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i);
3193 rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]);
3194 if (rc)
3195 return rc;
3196 }
3197
3198 return rc;
3199 }
3200
3201 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter,
3202 struct ibmvnic_sub_crq_queue *scrq,
3203 bool do_h_free)
3204 {
3205 struct device *dev = &adapter->vdev->dev;
3206 long rc;
3207
3208 netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n");
3209
3210 if (do_h_free) {
3211 /* Close the sub-crqs */
3212 do {
3213 rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3214 adapter->vdev->unit_address,
3215 scrq->crq_num);
3216 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
3217
3218 if (rc) {
3219 netdev_err(adapter->netdev,
3220 "Failed to release sub-CRQ %16lx, rc = %ld\n",
3221 scrq->crq_num, rc);
3222 }
3223 }
3224
3225 dma_free_coherent(dev,
3226 IBMVNIC_IND_ARR_SZ,
3227 scrq->ind_buf.indir_arr,
3228 scrq->ind_buf.indir_dma);
3229
3230 dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3231 DMA_BIDIRECTIONAL);
3232 free_pages((unsigned long)scrq->msgs, 2);
3233 kfree(scrq);
3234 }
3235
3236 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter
3237 *adapter)
3238 {
3239 struct device *dev = &adapter->vdev->dev;
3240 struct ibmvnic_sub_crq_queue *scrq;
3241 int rc;
3242
3243 scrq = kzalloc(sizeof(*scrq), GFP_KERNEL);
3244 if (!scrq)
3245 return NULL;
3246
3247 scrq->msgs =
3248 (union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2);
3249 if (!scrq->msgs) {
3250 dev_warn(dev, "Couldn't allocate crq queue messages page\n");
3251 goto zero_page_failed;
3252 }
3253
3254 scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE,
3255 DMA_BIDIRECTIONAL);
3256 if (dma_mapping_error(dev, scrq->msg_token)) {
3257 dev_warn(dev, "Couldn't map crq queue messages page\n");
3258 goto map_failed;
3259 }
3260
3261 rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
3262 4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
3263
3264 if (rc == H_RESOURCE)
3265 rc = ibmvnic_reset_crq(adapter);
3266
3267 if (rc == H_CLOSED) {
3268 dev_warn(dev, "Partner adapter not ready, waiting.\n");
3269 } else if (rc) {
3270 dev_warn(dev, "Error %d registering sub-crq\n", rc);
3271 goto reg_failed;
3272 }
3273
3274 scrq->adapter = adapter;
3275 scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs);
3276 scrq->ind_buf.index = 0;
3277
3278 scrq->ind_buf.indir_arr =
3279 dma_alloc_coherent(dev,
3280 IBMVNIC_IND_ARR_SZ,
3281 &scrq->ind_buf.indir_dma,
3282 GFP_KERNEL);
3283
3284 if (!scrq->ind_buf.indir_arr)
3285 goto indir_failed;
3286
3287 spin_lock_init(&scrq->lock);
3288
3289 netdev_dbg(adapter->netdev,
3290 "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n",
3291 scrq->crq_num, scrq->hw_irq, scrq->irq);
3292
3293 return scrq;
3294
3295 indir_failed:
3296 do {
3297 rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3298 adapter->vdev->unit_address,
3299 scrq->crq_num);
3300 } while (rc == H_BUSY || rc == H_IS_LONG_BUSY(rc));
3301 reg_failed:
3302 dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3303 DMA_BIDIRECTIONAL);
3304 map_failed:
3305 free_pages((unsigned long)scrq->msgs, 2);
3306 zero_page_failed:
3307 kfree(scrq);
3308
3309 return NULL;
3310 }
3311
3312 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free)
3313 {
3314 int i;
3315
3316 if (adapter->tx_scrq) {
3317 for (i = 0; i < adapter->num_active_tx_scrqs; i++) {
3318 if (!adapter->tx_scrq[i])
3319 continue;
3320
3321 netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n",
3322 i);
3323 ibmvnic_tx_scrq_clean_buffer(adapter, adapter->tx_scrq[i]);
3324 if (adapter->tx_scrq[i]->irq) {
3325 free_irq(adapter->tx_scrq[i]->irq,
3326 adapter->tx_scrq[i]);
3327 irq_dispose_mapping(adapter->tx_scrq[i]->irq);
3328 adapter->tx_scrq[i]->irq = 0;
3329 }
3330
3331 release_sub_crq_queue(adapter, adapter->tx_scrq[i],
3332 do_h_free);
3333 }
3334
3335 kfree(adapter->tx_scrq);
3336 adapter->tx_scrq = NULL;
3337 adapter->num_active_tx_scrqs = 0;
3338 }
3339
3340 if (adapter->rx_scrq) {
3341 for (i = 0; i < adapter->num_active_rx_scrqs; i++) {
3342 if (!adapter->rx_scrq[i])
3343 continue;
3344
3345 netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n",
3346 i);
3347 if (adapter->rx_scrq[i]->irq) {
3348 free_irq(adapter->rx_scrq[i]->irq,
3349 adapter->rx_scrq[i]);
3350 irq_dispose_mapping(adapter->rx_scrq[i]->irq);
3351 adapter->rx_scrq[i]->irq = 0;
3352 }
3353
3354 release_sub_crq_queue(adapter, adapter->rx_scrq[i],
3355 do_h_free);
3356 }
3357
3358 kfree(adapter->rx_scrq);
3359 adapter->rx_scrq = NULL;
3360 adapter->num_active_rx_scrqs = 0;
3361 }
3362 }
3363
3364 static int disable_scrq_irq(struct ibmvnic_adapter *adapter,
3365 struct ibmvnic_sub_crq_queue *scrq)
3366 {
3367 struct device *dev = &adapter->vdev->dev;
3368 unsigned long rc;
3369
3370 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3371 H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3372 if (rc)
3373 dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n",
3374 scrq->hw_irq, rc);
3375 return rc;
3376 }
3377
3378 static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
3379 struct ibmvnic_sub_crq_queue *scrq)
3380 {
3381 struct device *dev = &adapter->vdev->dev;
3382 unsigned long rc;
3383
3384 if (scrq->hw_irq > 0x100000000ULL) {
3385 dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq);
3386 return 1;
3387 }
3388
3389 if (test_bit(0, &adapter->resetting) &&
3390 adapter->reset_reason == VNIC_RESET_MOBILITY) {
3391 u64 val = (0xff000000) | scrq->hw_irq;
3392
3393 rc = plpar_hcall_norets(H_EOI, val);
3394 /* H_EOI would fail with rc = H_FUNCTION when running
3395 * in XIVE mode which is expected, but not an error.
3396 */
3397 if (rc && (rc != H_FUNCTION))
3398 dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n",
3399 val, rc);
3400 }
3401
3402 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3403 H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3404 if (rc)
3405 dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n",
3406 scrq->hw_irq, rc);
3407 return rc;
3408 }
3409
3410 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter,
3411 struct ibmvnic_sub_crq_queue *scrq)
3412 {
3413 struct device *dev = &adapter->vdev->dev;
3414 struct ibmvnic_tx_pool *tx_pool;
3415 struct ibmvnic_tx_buff *txbuff;
3416 struct netdev_queue *txq;
3417 union sub_crq *next;
3418 int index;
3419 int i;
3420
3421 restart_loop:
3422 while (pending_scrq(adapter, scrq)) {
3423 unsigned int pool = scrq->pool_index;
3424 int num_entries = 0;
3425 int total_bytes = 0;
3426 int num_packets = 0;
3427
3428 next = ibmvnic_next_scrq(adapter, scrq);
3429 for (i = 0; i < next->tx_comp.num_comps; i++) {
3430 index = be32_to_cpu(next->tx_comp.correlators[i]);
3431 if (index & IBMVNIC_TSO_POOL_MASK) {
3432 tx_pool = &adapter->tso_pool[pool];
3433 index &= ~IBMVNIC_TSO_POOL_MASK;
3434 } else {
3435 tx_pool = &adapter->tx_pool[pool];
3436 }
3437
3438 txbuff = &tx_pool->tx_buff[index];
3439 num_packets++;
3440 num_entries += txbuff->num_entries;
3441 if (txbuff->skb) {
3442 total_bytes += txbuff->skb->len;
3443 if (next->tx_comp.rcs[i]) {
3444 dev_err(dev, "tx error %x\n",
3445 next->tx_comp.rcs[i]);
3446 dev_kfree_skb_irq(txbuff->skb);
3447 } else {
3448 dev_consume_skb_irq(txbuff->skb);
3449 }
3450 txbuff->skb = NULL;
3451 } else {
3452 netdev_warn(adapter->netdev,
3453 "TX completion received with NULL socket buffer\n");
3454 }
3455 tx_pool->free_map[tx_pool->producer_index] = index;
3456 tx_pool->producer_index =
3457 (tx_pool->producer_index + 1) %
3458 tx_pool->num_buffers;
3459 }
3460 /* remove tx_comp scrq*/
3461 next->tx_comp.first = 0;
3462
3463 txq = netdev_get_tx_queue(adapter->netdev, scrq->pool_index);
3464 netdev_tx_completed_queue(txq, num_packets, total_bytes);
3465
3466 if (atomic_sub_return(num_entries, &scrq->used) <=
3467 (adapter->req_tx_entries_per_subcrq / 2) &&
3468 __netif_subqueue_stopped(adapter->netdev,
3469 scrq->pool_index)) {
3470 netif_wake_subqueue(adapter->netdev, scrq->pool_index);
3471 netdev_dbg(adapter->netdev, "Started queue %d\n",
3472 scrq->pool_index);
3473 }
3474 }
3475
3476 enable_scrq_irq(adapter, scrq);
3477
3478 if (pending_scrq(adapter, scrq)) {
3479 disable_scrq_irq(adapter, scrq);
3480 goto restart_loop;
3481 }
3482
3483 return 0;
3484 }
3485
3486 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance)
3487 {
3488 struct ibmvnic_sub_crq_queue *scrq = instance;
3489 struct ibmvnic_adapter *adapter = scrq->adapter;
3490
3491 disable_scrq_irq(adapter, scrq);
3492 ibmvnic_complete_tx(adapter, scrq);
3493
3494 return IRQ_HANDLED;
3495 }
3496
3497 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance)
3498 {
3499 struct ibmvnic_sub_crq_queue *scrq = instance;
3500 struct ibmvnic_adapter *adapter = scrq->adapter;
3501
3502 /* When booting a kdump kernel we can hit pending interrupts
3503 * prior to completing driver initialization.
3504 */
3505 if (unlikely(adapter->state != VNIC_OPEN))
3506 return IRQ_NONE;
3507
3508 adapter->rx_stats_buffers[scrq->scrq_num].interrupts++;
3509
3510 if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) {
3511 disable_scrq_irq(adapter, scrq);
3512 __napi_schedule(&adapter->napi[scrq->scrq_num]);
3513 }
3514
3515 return IRQ_HANDLED;
3516 }
3517
3518 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter)
3519 {
3520 struct device *dev = &adapter->vdev->dev;
3521 struct ibmvnic_sub_crq_queue *scrq;
3522 int i = 0, j = 0;
3523 int rc = 0;
3524
3525 for (i = 0; i < adapter->req_tx_queues; i++) {
3526 netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n",
3527 i);
3528 scrq = adapter->tx_scrq[i];
3529 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3530
3531 if (!scrq->irq) {
3532 rc = -EINVAL;
3533 dev_err(dev, "Error mapping irq\n");
3534 goto req_tx_irq_failed;
3535 }
3536
3537 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-tx%d",
3538 adapter->vdev->unit_address, i);
3539 rc = request_irq(scrq->irq, ibmvnic_interrupt_tx,
3540 0, scrq->name, scrq);
3541
3542 if (rc) {
3543 dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n",
3544 scrq->irq, rc);
3545 irq_dispose_mapping(scrq->irq);
3546 goto req_tx_irq_failed;
3547 }
3548 }
3549
3550 for (i = 0; i < adapter->req_rx_queues; i++) {
3551 netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n",
3552 i);
3553 scrq = adapter->rx_scrq[i];
3554 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3555 if (!scrq->irq) {
3556 rc = -EINVAL;
3557 dev_err(dev, "Error mapping irq\n");
3558 goto req_rx_irq_failed;
3559 }
3560 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-rx%d",
3561 adapter->vdev->unit_address, i);
3562 rc = request_irq(scrq->irq, ibmvnic_interrupt_rx,
3563 0, scrq->name, scrq);
3564 if (rc) {
3565 dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n",
3566 scrq->irq, rc);
3567 irq_dispose_mapping(scrq->irq);
3568 goto req_rx_irq_failed;
3569 }
3570 }
3571 return rc;
3572
3573 req_rx_irq_failed:
3574 for (j = 0; j < i; j++) {
3575 free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]);
3576 irq_dispose_mapping(adapter->rx_scrq[j]->irq);
3577 }
3578 i = adapter->req_tx_queues;
3579 req_tx_irq_failed:
3580 for (j = 0; j < i; j++) {
3581 free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]);
3582 irq_dispose_mapping(adapter->tx_scrq[j]->irq);
3583 }
3584 release_sub_crqs(adapter, 1);
3585 return rc;
3586 }
3587
3588 static int init_sub_crqs(struct ibmvnic_adapter *adapter)
3589 {
3590 struct device *dev = &adapter->vdev->dev;
3591 struct ibmvnic_sub_crq_queue **allqueues;
3592 int registered_queues = 0;
3593 int total_queues;
3594 int more = 0;
3595 int i;
3596
3597 total_queues = adapter->req_tx_queues + adapter->req_rx_queues;
3598
3599 allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL);
3600 if (!allqueues)
3601 return -1;
3602
3603 for (i = 0; i < total_queues; i++) {
3604 allqueues[i] = init_sub_crq_queue(adapter);
3605 if (!allqueues[i]) {
3606 dev_warn(dev, "Couldn't allocate all sub-crqs\n");
3607 break;
3608 }
3609 registered_queues++;
3610 }
3611
3612 /* Make sure we were able to register the minimum number of queues */
3613 if (registered_queues <
3614 adapter->min_tx_queues + adapter->min_rx_queues) {
3615 dev_err(dev, "Fatal: Couldn't init min number of sub-crqs\n");
3616 goto tx_failed;
3617 }
3618
3619 /* Distribute the failed allocated queues*/
3620 for (i = 0; i < total_queues - registered_queues + more ; i++) {
3621 netdev_dbg(adapter->netdev, "Reducing number of queues\n");
3622 switch (i % 3) {
3623 case 0:
3624 if (adapter->req_rx_queues > adapter->min_rx_queues)
3625 adapter->req_rx_queues--;
3626 else
3627 more++;
3628 break;
3629 case 1:
3630 if (adapter->req_tx_queues > adapter->min_tx_queues)
3631 adapter->req_tx_queues--;
3632 else
3633 more++;
3634 break;
3635 }
3636 }
3637
3638 adapter->tx_scrq = kcalloc(adapter->req_tx_queues,
3639 sizeof(*adapter->tx_scrq), GFP_KERNEL);
3640 if (!adapter->tx_scrq)
3641 goto tx_failed;
3642
3643 for (i = 0; i < adapter->req_tx_queues; i++) {
3644 adapter->tx_scrq[i] = allqueues[i];
3645 adapter->tx_scrq[i]->pool_index = i;
3646 adapter->num_active_tx_scrqs++;
3647 }
3648
3649 adapter->rx_scrq = kcalloc(adapter->req_rx_queues,
3650 sizeof(*adapter->rx_scrq), GFP_KERNEL);
3651 if (!adapter->rx_scrq)
3652 goto rx_failed;
3653
3654 for (i = 0; i < adapter->req_rx_queues; i++) {
3655 adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues];
3656 adapter->rx_scrq[i]->scrq_num = i;
3657 adapter->num_active_rx_scrqs++;
3658 }
3659
3660 kfree(allqueues);
3661 return 0;
3662
3663 rx_failed:
3664 kfree(adapter->tx_scrq);
3665 adapter->tx_scrq = NULL;
3666 tx_failed:
3667 for (i = 0; i < registered_queues; i++)
3668 release_sub_crq_queue(adapter, allqueues[i], 1);
3669 kfree(allqueues);
3670 return -1;
3671 }
3672
3673 static void send_request_cap(struct ibmvnic_adapter *adapter, int retry)
3674 {
3675 struct device *dev = &adapter->vdev->dev;
3676 union ibmvnic_crq crq;
3677 int max_entries;
3678
3679 if (!retry) {
3680 /* Sub-CRQ entries are 32 byte long */
3681 int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4);
3682
3683 if (adapter->min_tx_entries_per_subcrq > entries_page ||
3684 adapter->min_rx_add_entries_per_subcrq > entries_page) {
3685 dev_err(dev, "Fatal, invalid entries per sub-crq\n");
3686 return;
3687 }
3688
3689 if (adapter->desired.mtu)
3690 adapter->req_mtu = adapter->desired.mtu;
3691 else
3692 adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN;
3693
3694 if (!adapter->desired.tx_entries)
3695 adapter->desired.tx_entries =
3696 adapter->max_tx_entries_per_subcrq;
3697 if (!adapter->desired.rx_entries)
3698 adapter->desired.rx_entries =
3699 adapter->max_rx_add_entries_per_subcrq;
3700
3701 max_entries = IBMVNIC_MAX_LTB_SIZE /
3702 (adapter->req_mtu + IBMVNIC_BUFFER_HLEN);
3703
3704 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3705 adapter->desired.tx_entries > IBMVNIC_MAX_LTB_SIZE) {
3706 adapter->desired.tx_entries = max_entries;
3707 }
3708
3709 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3710 adapter->desired.rx_entries > IBMVNIC_MAX_LTB_SIZE) {
3711 adapter->desired.rx_entries = max_entries;
3712 }
3713
3714 if (adapter->desired.tx_entries)
3715 adapter->req_tx_entries_per_subcrq =
3716 adapter->desired.tx_entries;
3717 else
3718 adapter->req_tx_entries_per_subcrq =
3719 adapter->max_tx_entries_per_subcrq;
3720
3721 if (adapter->desired.rx_entries)
3722 adapter->req_rx_add_entries_per_subcrq =
3723 adapter->desired.rx_entries;
3724 else
3725 adapter->req_rx_add_entries_per_subcrq =
3726 adapter->max_rx_add_entries_per_subcrq;
3727
3728 if (adapter->desired.tx_queues)
3729 adapter->req_tx_queues =
3730 adapter->desired.tx_queues;
3731 else
3732 adapter->req_tx_queues =
3733 adapter->opt_tx_comp_sub_queues;
3734
3735 if (adapter->desired.rx_queues)
3736 adapter->req_rx_queues =
3737 adapter->desired.rx_queues;
3738 else
3739 adapter->req_rx_queues =
3740 adapter->opt_rx_comp_queues;
3741
3742 adapter->req_rx_add_queues = adapter->max_rx_add_queues;
3743 }
3744
3745 memset(&crq, 0, sizeof(crq));
3746 crq.request_capability.first = IBMVNIC_CRQ_CMD;
3747 crq.request_capability.cmd = REQUEST_CAPABILITY;
3748
3749 crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES);
3750 crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues);
3751 atomic_inc(&adapter->running_cap_crqs);
3752 ibmvnic_send_crq(adapter, &crq);
3753
3754 crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES);
3755 crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues);
3756 atomic_inc(&adapter->running_cap_crqs);
3757 ibmvnic_send_crq(adapter, &crq);
3758
3759 crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES);
3760 crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues);
3761 atomic_inc(&adapter->running_cap_crqs);
3762 ibmvnic_send_crq(adapter, &crq);
3763
3764 crq.request_capability.capability =
3765 cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ);
3766 crq.request_capability.number =
3767 cpu_to_be64(adapter->req_tx_entries_per_subcrq);
3768 atomic_inc(&adapter->running_cap_crqs);
3769 ibmvnic_send_crq(adapter, &crq);
3770
3771 crq.request_capability.capability =
3772 cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ);
3773 crq.request_capability.number =
3774 cpu_to_be64(adapter->req_rx_add_entries_per_subcrq);
3775 atomic_inc(&adapter->running_cap_crqs);
3776 ibmvnic_send_crq(adapter, &crq);
3777
3778 crq.request_capability.capability = cpu_to_be16(REQ_MTU);
3779 crq.request_capability.number = cpu_to_be64(adapter->req_mtu);
3780 atomic_inc(&adapter->running_cap_crqs);
3781 ibmvnic_send_crq(adapter, &crq);
3782
3783 if (adapter->netdev->flags & IFF_PROMISC) {
3784 if (adapter->promisc_supported) {
3785 crq.request_capability.capability =
3786 cpu_to_be16(PROMISC_REQUESTED);
3787 crq.request_capability.number = cpu_to_be64(1);
3788 atomic_inc(&adapter->running_cap_crqs);
3789 ibmvnic_send_crq(adapter, &crq);
3790 }
3791 } else {
3792 crq.request_capability.capability =
3793 cpu_to_be16(PROMISC_REQUESTED);
3794 crq.request_capability.number = cpu_to_be64(0);
3795 atomic_inc(&adapter->running_cap_crqs);
3796 ibmvnic_send_crq(adapter, &crq);
3797 }
3798 }
3799
3800 static int pending_scrq(struct ibmvnic_adapter *adapter,
3801 struct ibmvnic_sub_crq_queue *scrq)
3802 {
3803 union sub_crq *entry = &scrq->msgs[scrq->cur];
3804 int rc;
3805
3806 rc = !!(entry->generic.first & IBMVNIC_CRQ_CMD_RSP);
3807
3808 /* Ensure that the SCRQ valid flag is loaded prior to loading the
3809 * contents of the SCRQ descriptor
3810 */
3811 dma_rmb();
3812
3813 return rc;
3814 }
3815
3816 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter,
3817 struct ibmvnic_sub_crq_queue *scrq)
3818 {
3819 union sub_crq *entry;
3820 unsigned long flags;
3821
3822 spin_lock_irqsave(&scrq->lock, flags);
3823 entry = &scrq->msgs[scrq->cur];
3824 if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) {
3825 if (++scrq->cur == scrq->size)
3826 scrq->cur = 0;
3827 } else {
3828 entry = NULL;
3829 }
3830 spin_unlock_irqrestore(&scrq->lock, flags);
3831
3832 /* Ensure that the SCRQ valid flag is loaded prior to loading the
3833 * contents of the SCRQ descriptor
3834 */
3835 dma_rmb();
3836
3837 return entry;
3838 }
3839
3840 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter)
3841 {
3842 struct ibmvnic_crq_queue *queue = &adapter->crq;
3843 union ibmvnic_crq *crq;
3844
3845 crq = &queue->msgs[queue->cur];
3846 if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) {
3847 if (++queue->cur == queue->size)
3848 queue->cur = 0;
3849 } else {
3850 crq = NULL;
3851 }
3852
3853 return crq;
3854 }
3855
3856 static void print_subcrq_error(struct device *dev, int rc, const char *func)
3857 {
3858 switch (rc) {
3859 case H_PARAMETER:
3860 dev_warn_ratelimited(dev,
3861 "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n",
3862 func, rc);
3863 break;
3864 case H_CLOSED:
3865 dev_warn_ratelimited(dev,
3866 "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n",
3867 func, rc);
3868 break;
3869 default:
3870 dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc);
3871 break;
3872 }
3873 }
3874
3875 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter,
3876 u64 remote_handle, u64 ioba, u64 num_entries)
3877 {
3878 unsigned int ua = adapter->vdev->unit_address;
3879 struct device *dev = &adapter->vdev->dev;
3880 int rc;
3881
3882 /* Make sure the hypervisor sees the complete request */
3883 dma_wmb();
3884 rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua,
3885 cpu_to_be64(remote_handle),
3886 ioba, num_entries);
3887
3888 if (rc)
3889 print_subcrq_error(dev, rc, __func__);
3890
3891 return rc;
3892 }
3893
3894 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter,
3895 union ibmvnic_crq *crq)
3896 {
3897 unsigned int ua = adapter->vdev->unit_address;
3898 struct device *dev = &adapter->vdev->dev;
3899 u64 *u64_crq = (u64 *)crq;
3900 int rc;
3901
3902 netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n",
3903 (unsigned long)cpu_to_be64(u64_crq[0]),
3904 (unsigned long)cpu_to_be64(u64_crq[1]));
3905
3906 if (!adapter->crq.active &&
3907 crq->generic.first != IBMVNIC_CRQ_INIT_CMD) {
3908 dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n");
3909 return -EINVAL;
3910 }
3911
3912 /* Make sure the hypervisor sees the complete request */
3913 dma_wmb();
3914
3915 rc = plpar_hcall_norets(H_SEND_CRQ, ua,
3916 cpu_to_be64(u64_crq[0]),
3917 cpu_to_be64(u64_crq[1]));
3918
3919 if (rc) {
3920 if (rc == H_CLOSED) {
3921 dev_warn(dev, "CRQ Queue closed\n");
3922 /* do not reset, report the fail, wait for passive init from server */
3923 }
3924
3925 dev_warn(dev, "Send error (rc=%d)\n", rc);
3926 }
3927
3928 return rc;
3929 }
3930
3931 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter)
3932 {
3933 struct device *dev = &adapter->vdev->dev;
3934 union ibmvnic_crq crq;
3935 int retries = 100;
3936 int rc;
3937
3938 memset(&crq, 0, sizeof(crq));
3939 crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
3940 crq.generic.cmd = IBMVNIC_CRQ_INIT;
3941 netdev_dbg(adapter->netdev, "Sending CRQ init\n");
3942
3943 do {
3944 rc = ibmvnic_send_crq(adapter, &crq);
3945 if (rc != H_CLOSED)
3946 break;
3947 retries--;
3948 msleep(50);
3949
3950 } while (retries > 0);
3951
3952 if (rc) {
3953 dev_err(dev, "Failed to send init request, rc = %d\n", rc);
3954 return rc;
3955 }
3956
3957 return 0;
3958 }
3959
3960 struct vnic_login_client_data {
3961 u8 type;
3962 __be16 len;
3963 char name[];
3964 } __packed;
3965
3966 static int vnic_client_data_len(struct ibmvnic_adapter *adapter)
3967 {
3968 int len;
3969
3970 /* Calculate the amount of buffer space needed for the
3971 * vnic client data in the login buffer. There are four entries,
3972 * OS name, LPAR name, device name, and a null last entry.
3973 */
3974 len = 4 * sizeof(struct vnic_login_client_data);
3975 len += 6; /* "Linux" plus NULL */
3976 len += strlen(utsname()->nodename) + 1;
3977 len += strlen(adapter->netdev->name) + 1;
3978
3979 return len;
3980 }
3981
3982 static void vnic_add_client_data(struct ibmvnic_adapter *adapter,
3983 struct vnic_login_client_data *vlcd)
3984 {
3985 const char *os_name = "Linux";
3986 int len;
3987
3988 /* Type 1 - LPAR OS */
3989 vlcd->type = 1;
3990 len = strlen(os_name) + 1;
3991 vlcd->len = cpu_to_be16(len);
3992 strscpy(vlcd->name, os_name, len);
3993 vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
3994
3995 /* Type 2 - LPAR name */
3996 vlcd->type = 2;
3997 len = strlen(utsname()->nodename) + 1;
3998 vlcd->len = cpu_to_be16(len);
3999 strscpy(vlcd->name, utsname()->nodename, len);
4000 vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
4001
4002 /* Type 3 - device name */
4003 vlcd->type = 3;
4004 len = strlen(adapter->netdev->name) + 1;
4005 vlcd->len = cpu_to_be16(len);
4006 strscpy(vlcd->name, adapter->netdev->name, len);
4007 }
4008
4009 static int send_login(struct ibmvnic_adapter *adapter)
4010 {
4011 struct ibmvnic_login_rsp_buffer *login_rsp_buffer;
4012 struct ibmvnic_login_buffer *login_buffer;
4013 struct device *dev = &adapter->vdev->dev;
4014 struct vnic_login_client_data *vlcd;
4015 dma_addr_t rsp_buffer_token;
4016 dma_addr_t buffer_token;
4017 size_t rsp_buffer_size;
4018 union ibmvnic_crq crq;
4019 int client_data_len;
4020 size_t buffer_size;
4021 __be64 *tx_list_p;
4022 __be64 *rx_list_p;
4023 int rc;
4024 int i;
4025
4026 if (!adapter->tx_scrq || !adapter->rx_scrq) {
4027 netdev_err(adapter->netdev,
4028 "RX or TX queues are not allocated, device login failed\n");
4029 return -1;
4030 }
4031
4032 release_login_buffer(adapter);
4033 release_login_rsp_buffer(adapter);
4034
4035 client_data_len = vnic_client_data_len(adapter);
4036
4037 buffer_size =
4038 sizeof(struct ibmvnic_login_buffer) +
4039 sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) +
4040 client_data_len;
4041
4042 login_buffer = kzalloc(buffer_size, GFP_ATOMIC);
4043 if (!login_buffer)
4044 goto buf_alloc_failed;
4045
4046 buffer_token = dma_map_single(dev, login_buffer, buffer_size,
4047 DMA_TO_DEVICE);
4048 if (dma_mapping_error(dev, buffer_token)) {
4049 dev_err(dev, "Couldn't map login buffer\n");
4050 goto buf_map_failed;
4051 }
4052
4053 rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) +
4054 sizeof(u64) * adapter->req_tx_queues +
4055 sizeof(u64) * adapter->req_rx_queues +
4056 sizeof(u64) * adapter->req_rx_queues +
4057 sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS;
4058
4059 login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC);
4060 if (!login_rsp_buffer)
4061 goto buf_rsp_alloc_failed;
4062
4063 rsp_buffer_token = dma_map_single(dev, login_rsp_buffer,
4064 rsp_buffer_size, DMA_FROM_DEVICE);
4065 if (dma_mapping_error(dev, rsp_buffer_token)) {
4066 dev_err(dev, "Couldn't map login rsp buffer\n");
4067 goto buf_rsp_map_failed;
4068 }
4069
4070 adapter->login_buf = login_buffer;
4071 adapter->login_buf_token = buffer_token;
4072 adapter->login_buf_sz = buffer_size;
4073 adapter->login_rsp_buf = login_rsp_buffer;
4074 adapter->login_rsp_buf_token = rsp_buffer_token;
4075 adapter->login_rsp_buf_sz = rsp_buffer_size;
4076
4077 login_buffer->len = cpu_to_be32(buffer_size);
4078 login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB);
4079 login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues);
4080 login_buffer->off_txcomp_subcrqs =
4081 cpu_to_be32(sizeof(struct ibmvnic_login_buffer));
4082 login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues);
4083 login_buffer->off_rxcomp_subcrqs =
4084 cpu_to_be32(sizeof(struct ibmvnic_login_buffer) +
4085 sizeof(u64) * adapter->req_tx_queues);
4086 login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token);
4087 login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size);
4088
4089 tx_list_p = (__be64 *)((char *)login_buffer +
4090 sizeof(struct ibmvnic_login_buffer));
4091 rx_list_p = (__be64 *)((char *)login_buffer +
4092 sizeof(struct ibmvnic_login_buffer) +
4093 sizeof(u64) * adapter->req_tx_queues);
4094
4095 for (i = 0; i < adapter->req_tx_queues; i++) {
4096 if (adapter->tx_scrq[i]) {
4097 tx_list_p[i] =
4098 cpu_to_be64(adapter->tx_scrq[i]->crq_num);
4099 }
4100 }
4101
4102 for (i = 0; i < adapter->req_rx_queues; i++) {
4103 if (adapter->rx_scrq[i]) {
4104 rx_list_p[i] =
4105 cpu_to_be64(adapter->rx_scrq[i]->crq_num);
4106 }
4107 }
4108
4109 /* Insert vNIC login client data */
4110 vlcd = (struct vnic_login_client_data *)
4111 ((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues));
4112 login_buffer->client_data_offset =
4113 cpu_to_be32((char *)vlcd - (char *)login_buffer);
4114 login_buffer->client_data_len = cpu_to_be32(client_data_len);
4115
4116 vnic_add_client_data(adapter, vlcd);
4117
4118 netdev_dbg(adapter->netdev, "Login Buffer:\n");
4119 for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) {
4120 netdev_dbg(adapter->netdev, "%016lx\n",
4121 ((unsigned long *)(adapter->login_buf))[i]);
4122 }
4123
4124 memset(&crq, 0, sizeof(crq));
4125 crq.login.first = IBMVNIC_CRQ_CMD;
4126 crq.login.cmd = LOGIN;
4127 crq.login.ioba = cpu_to_be32(buffer_token);
4128 crq.login.len = cpu_to_be32(buffer_size);
4129
4130 adapter->login_pending = true;
4131 rc = ibmvnic_send_crq(adapter, &crq);
4132 if (rc) {
4133 adapter->login_pending = false;
4134 netdev_err(adapter->netdev, "Failed to send login, rc=%d\n", rc);
4135 goto buf_rsp_map_failed;
4136 }
4137
4138 return 0;
4139
4140 buf_rsp_map_failed:
4141 kfree(login_rsp_buffer);
4142 adapter->login_rsp_buf = NULL;
4143 buf_rsp_alloc_failed:
4144 dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE);
4145 buf_map_failed:
4146 kfree(login_buffer);
4147 adapter->login_buf = NULL;
4148 buf_alloc_failed:
4149 return -1;
4150 }
4151
4152 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr,
4153 u32 len, u8 map_id)
4154 {
4155 union ibmvnic_crq crq;
4156
4157 memset(&crq, 0, sizeof(crq));
4158 crq.request_map.first = IBMVNIC_CRQ_CMD;
4159 crq.request_map.cmd = REQUEST_MAP;
4160 crq.request_map.map_id = map_id;
4161 crq.request_map.ioba = cpu_to_be32(addr);
4162 crq.request_map.len = cpu_to_be32(len);
4163 return ibmvnic_send_crq(adapter, &crq);
4164 }
4165
4166 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id)
4167 {
4168 union ibmvnic_crq crq;
4169
4170 memset(&crq, 0, sizeof(crq));
4171 crq.request_unmap.first = IBMVNIC_CRQ_CMD;
4172 crq.request_unmap.cmd = REQUEST_UNMAP;
4173 crq.request_unmap.map_id = map_id;
4174 return ibmvnic_send_crq(adapter, &crq);
4175 }
4176
4177 static void send_query_map(struct ibmvnic_adapter *adapter)
4178 {
4179 union ibmvnic_crq crq;
4180
4181 memset(&crq, 0, sizeof(crq));
4182 crq.query_map.first = IBMVNIC_CRQ_CMD;
4183 crq.query_map.cmd = QUERY_MAP;
4184 ibmvnic_send_crq(adapter, &crq);
4185 }
4186
4187 /* Send a series of CRQs requesting various capabilities of the VNIC server */
4188 static void send_query_cap(struct ibmvnic_adapter *adapter)
4189 {
4190 union ibmvnic_crq crq;
4191
4192 atomic_set(&adapter->running_cap_crqs, 0);
4193 memset(&crq, 0, sizeof(crq));
4194 crq.query_capability.first = IBMVNIC_CRQ_CMD;
4195 crq.query_capability.cmd = QUERY_CAPABILITY;
4196
4197 crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES);
4198 atomic_inc(&adapter->running_cap_crqs);
4199 ibmvnic_send_crq(adapter, &crq);
4200
4201 crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES);
4202 atomic_inc(&adapter->running_cap_crqs);
4203 ibmvnic_send_crq(adapter, &crq);
4204
4205 crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES);
4206 atomic_inc(&adapter->running_cap_crqs);
4207 ibmvnic_send_crq(adapter, &crq);
4208
4209 crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES);
4210 atomic_inc(&adapter->running_cap_crqs);
4211 ibmvnic_send_crq(adapter, &crq);
4212
4213 crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES);
4214 atomic_inc(&adapter->running_cap_crqs);
4215 ibmvnic_send_crq(adapter, &crq);
4216
4217 crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES);
4218 atomic_inc(&adapter->running_cap_crqs);
4219 ibmvnic_send_crq(adapter, &crq);
4220
4221 crq.query_capability.capability =
4222 cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ);
4223 atomic_inc(&adapter->running_cap_crqs);
4224 ibmvnic_send_crq(adapter, &crq);
4225
4226 crq.query_capability.capability =
4227 cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ);
4228 atomic_inc(&adapter->running_cap_crqs);
4229 ibmvnic_send_crq(adapter, &crq);
4230
4231 crq.query_capability.capability =
4232 cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ);
4233 atomic_inc(&adapter->running_cap_crqs);
4234 ibmvnic_send_crq(adapter, &crq);
4235
4236 crq.query_capability.capability =
4237 cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ);
4238 atomic_inc(&adapter->running_cap_crqs);
4239 ibmvnic_send_crq(adapter, &crq);
4240
4241 crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD);
4242 atomic_inc(&adapter->running_cap_crqs);
4243 ibmvnic_send_crq(adapter, &crq);
4244
4245 crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED);
4246 atomic_inc(&adapter->running_cap_crqs);
4247 ibmvnic_send_crq(adapter, &crq);
4248
4249 crq.query_capability.capability = cpu_to_be16(MIN_MTU);
4250 atomic_inc(&adapter->running_cap_crqs);
4251 ibmvnic_send_crq(adapter, &crq);
4252
4253 crq.query_capability.capability = cpu_to_be16(MAX_MTU);
4254 atomic_inc(&adapter->running_cap_crqs);
4255 ibmvnic_send_crq(adapter, &crq);
4256
4257 crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS);
4258 atomic_inc(&adapter->running_cap_crqs);
4259 ibmvnic_send_crq(adapter, &crq);
4260
4261 crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION);
4262 atomic_inc(&adapter->running_cap_crqs);
4263 ibmvnic_send_crq(adapter, &crq);
4264
4265 crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION);
4266 atomic_inc(&adapter->running_cap_crqs);
4267 ibmvnic_send_crq(adapter, &crq);
4268
4269 crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES);
4270 atomic_inc(&adapter->running_cap_crqs);
4271 ibmvnic_send_crq(adapter, &crq);
4272
4273 crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED);
4274 atomic_inc(&adapter->running_cap_crqs);
4275 ibmvnic_send_crq(adapter, &crq);
4276
4277 crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES);
4278 atomic_inc(&adapter->running_cap_crqs);
4279 ibmvnic_send_crq(adapter, &crq);
4280
4281 crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES);
4282 atomic_inc(&adapter->running_cap_crqs);
4283 ibmvnic_send_crq(adapter, &crq);
4284
4285 crq.query_capability.capability =
4286 cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q);
4287 atomic_inc(&adapter->running_cap_crqs);
4288 ibmvnic_send_crq(adapter, &crq);
4289
4290 crq.query_capability.capability =
4291 cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ);
4292 atomic_inc(&adapter->running_cap_crqs);
4293 ibmvnic_send_crq(adapter, &crq);
4294
4295 crq.query_capability.capability =
4296 cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ);
4297 atomic_inc(&adapter->running_cap_crqs);
4298 ibmvnic_send_crq(adapter, &crq);
4299
4300 crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ);
4301 atomic_inc(&adapter->running_cap_crqs);
4302 ibmvnic_send_crq(adapter, &crq);
4303 }
4304
4305 static void send_query_ip_offload(struct ibmvnic_adapter *adapter)
4306 {
4307 int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer);
4308 struct device *dev = &adapter->vdev->dev;
4309 union ibmvnic_crq crq;
4310
4311 adapter->ip_offload_tok =
4312 dma_map_single(dev,
4313 &adapter->ip_offload_buf,
4314 buf_sz,
4315 DMA_FROM_DEVICE);
4316
4317 if (dma_mapping_error(dev, adapter->ip_offload_tok)) {
4318 if (!firmware_has_feature(FW_FEATURE_CMO))
4319 dev_err(dev, "Couldn't map offload buffer\n");
4320 return;
4321 }
4322
4323 memset(&crq, 0, sizeof(crq));
4324 crq.query_ip_offload.first = IBMVNIC_CRQ_CMD;
4325 crq.query_ip_offload.cmd = QUERY_IP_OFFLOAD;
4326 crq.query_ip_offload.len = cpu_to_be32(buf_sz);
4327 crq.query_ip_offload.ioba =
4328 cpu_to_be32(adapter->ip_offload_tok);
4329
4330 ibmvnic_send_crq(adapter, &crq);
4331 }
4332
4333 static void send_control_ip_offload(struct ibmvnic_adapter *adapter)
4334 {
4335 struct ibmvnic_control_ip_offload_buffer *ctrl_buf = &adapter->ip_offload_ctrl;
4336 struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4337 struct device *dev = &adapter->vdev->dev;
4338 netdev_features_t old_hw_features = 0;
4339 union ibmvnic_crq crq;
4340
4341 adapter->ip_offload_ctrl_tok =
4342 dma_map_single(dev,
4343 ctrl_buf,
4344 sizeof(adapter->ip_offload_ctrl),
4345 DMA_TO_DEVICE);
4346
4347 if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) {
4348 dev_err(dev, "Couldn't map ip offload control buffer\n");
4349 return;
4350 }
4351
4352 ctrl_buf->len = cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4353 ctrl_buf->version = cpu_to_be32(INITIAL_VERSION_IOB);
4354 ctrl_buf->ipv4_chksum = buf->ipv4_chksum;
4355 ctrl_buf->ipv6_chksum = buf->ipv6_chksum;
4356 ctrl_buf->tcp_ipv4_chksum = buf->tcp_ipv4_chksum;
4357 ctrl_buf->udp_ipv4_chksum = buf->udp_ipv4_chksum;
4358 ctrl_buf->tcp_ipv6_chksum = buf->tcp_ipv6_chksum;
4359 ctrl_buf->udp_ipv6_chksum = buf->udp_ipv6_chksum;
4360 ctrl_buf->large_tx_ipv4 = buf->large_tx_ipv4;
4361 ctrl_buf->large_tx_ipv6 = buf->large_tx_ipv6;
4362
4363 /* large_rx disabled for now, additional features needed */
4364 ctrl_buf->large_rx_ipv4 = 0;
4365 ctrl_buf->large_rx_ipv6 = 0;
4366
4367 if (adapter->state != VNIC_PROBING) {
4368 old_hw_features = adapter->netdev->hw_features;
4369 adapter->netdev->hw_features = 0;
4370 }
4371
4372 adapter->netdev->hw_features = NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
4373
4374 if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum)
4375 adapter->netdev->hw_features |= NETIF_F_IP_CSUM;
4376
4377 if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum)
4378 adapter->netdev->hw_features |= NETIF_F_IPV6_CSUM;
4379
4380 if ((adapter->netdev->features &
4381 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
4382 adapter->netdev->hw_features |= NETIF_F_RXCSUM;
4383
4384 if (buf->large_tx_ipv4)
4385 adapter->netdev->hw_features |= NETIF_F_TSO;
4386 if (buf->large_tx_ipv6)
4387 adapter->netdev->hw_features |= NETIF_F_TSO6;
4388
4389 if (adapter->state == VNIC_PROBING) {
4390 adapter->netdev->features |= adapter->netdev->hw_features;
4391 } else if (old_hw_features != adapter->netdev->hw_features) {
4392 netdev_features_t tmp = 0;
4393
4394 /* disable features no longer supported */
4395 adapter->netdev->features &= adapter->netdev->hw_features;
4396 /* turn on features now supported if previously enabled */
4397 tmp = (old_hw_features ^ adapter->netdev->hw_features) &
4398 adapter->netdev->hw_features;
4399 adapter->netdev->features |=
4400 tmp & adapter->netdev->wanted_features;
4401 }
4402
4403 memset(&crq, 0, sizeof(crq));
4404 crq.control_ip_offload.first = IBMVNIC_CRQ_CMD;
4405 crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD;
4406 crq.control_ip_offload.len =
4407 cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4408 crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok);
4409 ibmvnic_send_crq(adapter, &crq);
4410 }
4411
4412 static void handle_vpd_size_rsp(union ibmvnic_crq *crq,
4413 struct ibmvnic_adapter *adapter)
4414 {
4415 struct device *dev = &adapter->vdev->dev;
4416
4417 if (crq->get_vpd_size_rsp.rc.code) {
4418 dev_err(dev, "Error retrieving VPD size, rc=%x\n",
4419 crq->get_vpd_size_rsp.rc.code);
4420 complete(&adapter->fw_done);
4421 return;
4422 }
4423
4424 adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len);
4425 complete(&adapter->fw_done);
4426 }
4427
4428 static void handle_vpd_rsp(union ibmvnic_crq *crq,
4429 struct ibmvnic_adapter *adapter)
4430 {
4431 struct device *dev = &adapter->vdev->dev;
4432 unsigned char *substr = NULL;
4433 u8 fw_level_len = 0;
4434
4435 memset(adapter->fw_version, 0, 32);
4436
4437 dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len,
4438 DMA_FROM_DEVICE);
4439
4440 if (crq->get_vpd_rsp.rc.code) {
4441 dev_err(dev, "Error retrieving VPD from device, rc=%x\n",
4442 crq->get_vpd_rsp.rc.code);
4443 goto complete;
4444 }
4445
4446 /* get the position of the firmware version info
4447 * located after the ASCII 'RM' substring in the buffer
4448 */
4449 substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len);
4450 if (!substr) {
4451 dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n");
4452 goto complete;
4453 }
4454
4455 /* get length of firmware level ASCII substring */
4456 if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) {
4457 fw_level_len = *(substr + 2);
4458 } else {
4459 dev_info(dev, "Length of FW substr extrapolated VDP buff\n");
4460 goto complete;
4461 }
4462
4463 /* copy firmware version string from vpd into adapter */
4464 if ((substr + 3 + fw_level_len) <
4465 (adapter->vpd->buff + adapter->vpd->len)) {
4466 strncpy((char *)adapter->fw_version, substr + 3, fw_level_len);
4467 } else {
4468 dev_info(dev, "FW substr extrapolated VPD buff\n");
4469 }
4470
4471 complete:
4472 if (adapter->fw_version[0] == '\0')
4473 strscpy((char *)adapter->fw_version, "N/A", sizeof(adapter->fw_version));
4474 complete(&adapter->fw_done);
4475 }
4476
4477 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter)
4478 {
4479 struct device *dev = &adapter->vdev->dev;
4480 struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4481 int i;
4482
4483 dma_unmap_single(dev, adapter->ip_offload_tok,
4484 sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE);
4485
4486 netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n");
4487 for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++)
4488 netdev_dbg(adapter->netdev, "%016lx\n",
4489 ((unsigned long *)(buf))[i]);
4490
4491 netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum);
4492 netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum);
4493 netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n",
4494 buf->tcp_ipv4_chksum);
4495 netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n",
4496 buf->tcp_ipv6_chksum);
4497 netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n",
4498 buf->udp_ipv4_chksum);
4499 netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n",
4500 buf->udp_ipv6_chksum);
4501 netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n",
4502 buf->large_tx_ipv4);
4503 netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n",
4504 buf->large_tx_ipv6);
4505 netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n",
4506 buf->large_rx_ipv4);
4507 netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n",
4508 buf->large_rx_ipv6);
4509 netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n",
4510 buf->max_ipv4_header_size);
4511 netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n",
4512 buf->max_ipv6_header_size);
4513 netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n",
4514 buf->max_tcp_header_size);
4515 netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n",
4516 buf->max_udp_header_size);
4517 netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n",
4518 buf->max_large_tx_size);
4519 netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n",
4520 buf->max_large_rx_size);
4521 netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n",
4522 buf->ipv6_extension_header);
4523 netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n",
4524 buf->tcp_pseudosum_req);
4525 netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n",
4526 buf->num_ipv6_ext_headers);
4527 netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n",
4528 buf->off_ipv6_ext_headers);
4529
4530 send_control_ip_offload(adapter);
4531 }
4532
4533 static const char *ibmvnic_fw_err_cause(u16 cause)
4534 {
4535 switch (cause) {
4536 case ADAPTER_PROBLEM:
4537 return "adapter problem";
4538 case BUS_PROBLEM:
4539 return "bus problem";
4540 case FW_PROBLEM:
4541 return "firmware problem";
4542 case DD_PROBLEM:
4543 return "device driver problem";
4544 case EEH_RECOVERY:
4545 return "EEH recovery";
4546 case FW_UPDATED:
4547 return "firmware updated";
4548 case LOW_MEMORY:
4549 return "low Memory";
4550 default:
4551 return "unknown";
4552 }
4553 }
4554
4555 static void handle_error_indication(union ibmvnic_crq *crq,
4556 struct ibmvnic_adapter *adapter)
4557 {
4558 struct device *dev = &adapter->vdev->dev;
4559 u16 cause;
4560
4561 cause = be16_to_cpu(crq->error_indication.error_cause);
4562
4563 dev_warn_ratelimited(dev,
4564 "Firmware reports %serror, cause: %s. Starting recovery...\n",
4565 crq->error_indication.flags
4566 & IBMVNIC_FATAL_ERROR ? "FATAL " : "",
4567 ibmvnic_fw_err_cause(cause));
4568
4569 if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR)
4570 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4571 else
4572 ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL);
4573 }
4574
4575 static int handle_change_mac_rsp(union ibmvnic_crq *crq,
4576 struct ibmvnic_adapter *adapter)
4577 {
4578 struct net_device *netdev = adapter->netdev;
4579 struct device *dev = &adapter->vdev->dev;
4580 long rc;
4581
4582 rc = crq->change_mac_addr_rsp.rc.code;
4583 if (rc) {
4584 dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc);
4585 goto out;
4586 }
4587 /* crq->change_mac_addr.mac_addr is the requested one
4588 * crq->change_mac_addr_rsp.mac_addr is the returned valid one.
4589 */
4590 ether_addr_copy(netdev->dev_addr,
4591 &crq->change_mac_addr_rsp.mac_addr[0]);
4592 ether_addr_copy(adapter->mac_addr,
4593 &crq->change_mac_addr_rsp.mac_addr[0]);
4594 out:
4595 complete(&adapter->fw_done);
4596 return rc;
4597 }
4598
4599 static void handle_request_cap_rsp(union ibmvnic_crq *crq,
4600 struct ibmvnic_adapter *adapter)
4601 {
4602 struct device *dev = &adapter->vdev->dev;
4603 u64 *req_value;
4604 char *name;
4605
4606 atomic_dec(&adapter->running_cap_crqs);
4607 switch (be16_to_cpu(crq->request_capability_rsp.capability)) {
4608 case REQ_TX_QUEUES:
4609 req_value = &adapter->req_tx_queues;
4610 name = "tx";
4611 break;
4612 case REQ_RX_QUEUES:
4613 req_value = &adapter->req_rx_queues;
4614 name = "rx";
4615 break;
4616 case REQ_RX_ADD_QUEUES:
4617 req_value = &adapter->req_rx_add_queues;
4618 name = "rx_add";
4619 break;
4620 case REQ_TX_ENTRIES_PER_SUBCRQ:
4621 req_value = &adapter->req_tx_entries_per_subcrq;
4622 name = "tx_entries_per_subcrq";
4623 break;
4624 case REQ_RX_ADD_ENTRIES_PER_SUBCRQ:
4625 req_value = &adapter->req_rx_add_entries_per_subcrq;
4626 name = "rx_add_entries_per_subcrq";
4627 break;
4628 case REQ_MTU:
4629 req_value = &adapter->req_mtu;
4630 name = "mtu";
4631 break;
4632 case PROMISC_REQUESTED:
4633 req_value = &adapter->promisc;
4634 name = "promisc";
4635 break;
4636 default:
4637 dev_err(dev, "Got invalid cap request rsp %d\n",
4638 crq->request_capability.capability);
4639 return;
4640 }
4641
4642 switch (crq->request_capability_rsp.rc.code) {
4643 case SUCCESS:
4644 break;
4645 case PARTIALSUCCESS:
4646 dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n",
4647 *req_value,
4648 (long)be64_to_cpu(crq->request_capability_rsp.number),
4649 name);
4650
4651 if (be16_to_cpu(crq->request_capability_rsp.capability) ==
4652 REQ_MTU) {
4653 pr_err("mtu of %llu is not supported. Reverting.\n",
4654 *req_value);
4655 *req_value = adapter->fallback.mtu;
4656 } else {
4657 *req_value =
4658 be64_to_cpu(crq->request_capability_rsp.number);
4659 }
4660
4661 send_request_cap(adapter, 1);
4662 return;
4663 default:
4664 dev_err(dev, "Error %d in request cap rsp\n",
4665 crq->request_capability_rsp.rc.code);
4666 return;
4667 }
4668
4669 /* Done receiving requested capabilities, query IP offload support */
4670 if (atomic_read(&adapter->running_cap_crqs) == 0) {
4671 adapter->wait_capability = false;
4672 send_query_ip_offload(adapter);
4673 }
4674 }
4675
4676 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq,
4677 struct ibmvnic_adapter *adapter)
4678 {
4679 struct device *dev = &adapter->vdev->dev;
4680 struct net_device *netdev = adapter->netdev;
4681 struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf;
4682 struct ibmvnic_login_buffer *login = adapter->login_buf;
4683 u64 *tx_handle_array;
4684 u64 *rx_handle_array;
4685 int num_tx_pools;
4686 int num_rx_pools;
4687 u64 *size_array;
4688 int i;
4689
4690 /* CHECK: Test/set of login_pending does not need to be atomic
4691 * because only ibmvnic_tasklet tests/clears this.
4692 */
4693 if (!adapter->login_pending) {
4694 netdev_warn(netdev, "Ignoring unexpected login response\n");
4695 return 0;
4696 }
4697 adapter->login_pending = false;
4698
4699 dma_unmap_single(dev, adapter->login_buf_token, adapter->login_buf_sz,
4700 DMA_TO_DEVICE);
4701 dma_unmap_single(dev, adapter->login_rsp_buf_token,
4702 adapter->login_rsp_buf_sz, DMA_FROM_DEVICE);
4703
4704 /* If the number of queues requested can't be allocated by the
4705 * server, the login response will return with code 1. We will need
4706 * to resend the login buffer with fewer queues requested.
4707 */
4708 if (login_rsp_crq->generic.rc.code) {
4709 adapter->init_done_rc = login_rsp_crq->generic.rc.code;
4710 complete(&adapter->init_done);
4711 return 0;
4712 }
4713
4714 if (adapter->failover_pending) {
4715 adapter->init_done_rc = -EAGAIN;
4716 netdev_dbg(netdev, "Failover pending, ignoring login response\n");
4717 complete(&adapter->init_done);
4718 /* login response buffer will be released on reset */
4719 return 0;
4720 }
4721
4722 netdev->mtu = adapter->req_mtu - ETH_HLEN;
4723
4724 netdev_dbg(adapter->netdev, "Login Response Buffer:\n");
4725 for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) {
4726 netdev_dbg(adapter->netdev, "%016lx\n",
4727 ((unsigned long *)(adapter->login_rsp_buf))[i]);
4728 }
4729
4730 /* Sanity checks */
4731 if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs ||
4732 (be32_to_cpu(login->num_rxcomp_subcrqs) *
4733 adapter->req_rx_add_queues !=
4734 be32_to_cpu(login_rsp->num_rxadd_subcrqs))) {
4735 dev_err(dev, "FATAL: Inconsistent login and login rsp\n");
4736 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4737 return -EIO;
4738 }
4739 size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4740 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size));
4741 /* variable buffer sizes are not supported, so just read the
4742 * first entry.
4743 */
4744 adapter->cur_rx_buf_sz = be64_to_cpu(size_array[0]);
4745
4746 num_tx_pools = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs);
4747 num_rx_pools = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
4748
4749 tx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4750 be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs));
4751 rx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4752 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_subcrqs));
4753
4754 for (i = 0; i < num_tx_pools; i++)
4755 adapter->tx_scrq[i]->handle = tx_handle_array[i];
4756
4757 for (i = 0; i < num_rx_pools; i++)
4758 adapter->rx_scrq[i]->handle = rx_handle_array[i];
4759
4760 adapter->num_active_tx_scrqs = num_tx_pools;
4761 adapter->num_active_rx_scrqs = num_rx_pools;
4762 release_login_rsp_buffer(adapter);
4763 release_login_buffer(adapter);
4764 complete(&adapter->init_done);
4765
4766 return 0;
4767 }
4768
4769 static void handle_request_unmap_rsp(union ibmvnic_crq *crq,
4770 struct ibmvnic_adapter *adapter)
4771 {
4772 struct device *dev = &adapter->vdev->dev;
4773 long rc;
4774
4775 rc = crq->request_unmap_rsp.rc.code;
4776 if (rc)
4777 dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc);
4778 }
4779
4780 static void handle_query_map_rsp(union ibmvnic_crq *crq,
4781 struct ibmvnic_adapter *adapter)
4782 {
4783 struct net_device *netdev = adapter->netdev;
4784 struct device *dev = &adapter->vdev->dev;
4785 long rc;
4786
4787 rc = crq->query_map_rsp.rc.code;
4788 if (rc) {
4789 dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc);
4790 return;
4791 }
4792 netdev_dbg(netdev, "page_size = %d\ntot_pages = %d\nfree_pages = %d\n",
4793 crq->query_map_rsp.page_size, crq->query_map_rsp.tot_pages,
4794 crq->query_map_rsp.free_pages);
4795 }
4796
4797 static void handle_query_cap_rsp(union ibmvnic_crq *crq,
4798 struct ibmvnic_adapter *adapter)
4799 {
4800 struct net_device *netdev = adapter->netdev;
4801 struct device *dev = &adapter->vdev->dev;
4802 long rc;
4803
4804 atomic_dec(&adapter->running_cap_crqs);
4805 netdev_dbg(netdev, "Outstanding queries: %d\n",
4806 atomic_read(&adapter->running_cap_crqs));
4807 rc = crq->query_capability.rc.code;
4808 if (rc) {
4809 dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc);
4810 goto out;
4811 }
4812
4813 switch (be16_to_cpu(crq->query_capability.capability)) {
4814 case MIN_TX_QUEUES:
4815 adapter->min_tx_queues =
4816 be64_to_cpu(crq->query_capability.number);
4817 netdev_dbg(netdev, "min_tx_queues = %lld\n",
4818 adapter->min_tx_queues);
4819 break;
4820 case MIN_RX_QUEUES:
4821 adapter->min_rx_queues =
4822 be64_to_cpu(crq->query_capability.number);
4823 netdev_dbg(netdev, "min_rx_queues = %lld\n",
4824 adapter->min_rx_queues);
4825 break;
4826 case MIN_RX_ADD_QUEUES:
4827 adapter->min_rx_add_queues =
4828 be64_to_cpu(crq->query_capability.number);
4829 netdev_dbg(netdev, "min_rx_add_queues = %lld\n",
4830 adapter->min_rx_add_queues);
4831 break;
4832 case MAX_TX_QUEUES:
4833 adapter->max_tx_queues =
4834 be64_to_cpu(crq->query_capability.number);
4835 netdev_dbg(netdev, "max_tx_queues = %lld\n",
4836 adapter->max_tx_queues);
4837 break;
4838 case MAX_RX_QUEUES:
4839 adapter->max_rx_queues =
4840 be64_to_cpu(crq->query_capability.number);
4841 netdev_dbg(netdev, "max_rx_queues = %lld\n",
4842 adapter->max_rx_queues);
4843 break;
4844 case MAX_RX_ADD_QUEUES:
4845 adapter->max_rx_add_queues =
4846 be64_to_cpu(crq->query_capability.number);
4847 netdev_dbg(netdev, "max_rx_add_queues = %lld\n",
4848 adapter->max_rx_add_queues);
4849 break;
4850 case MIN_TX_ENTRIES_PER_SUBCRQ:
4851 adapter->min_tx_entries_per_subcrq =
4852 be64_to_cpu(crq->query_capability.number);
4853 netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n",
4854 adapter->min_tx_entries_per_subcrq);
4855 break;
4856 case MIN_RX_ADD_ENTRIES_PER_SUBCRQ:
4857 adapter->min_rx_add_entries_per_subcrq =
4858 be64_to_cpu(crq->query_capability.number);
4859 netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n",
4860 adapter->min_rx_add_entries_per_subcrq);
4861 break;
4862 case MAX_TX_ENTRIES_PER_SUBCRQ:
4863 adapter->max_tx_entries_per_subcrq =
4864 be64_to_cpu(crq->query_capability.number);
4865 netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n",
4866 adapter->max_tx_entries_per_subcrq);
4867 break;
4868 case MAX_RX_ADD_ENTRIES_PER_SUBCRQ:
4869 adapter->max_rx_add_entries_per_subcrq =
4870 be64_to_cpu(crq->query_capability.number);
4871 netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n",
4872 adapter->max_rx_add_entries_per_subcrq);
4873 break;
4874 case TCP_IP_OFFLOAD:
4875 adapter->tcp_ip_offload =
4876 be64_to_cpu(crq->query_capability.number);
4877 netdev_dbg(netdev, "tcp_ip_offload = %lld\n",
4878 adapter->tcp_ip_offload);
4879 break;
4880 case PROMISC_SUPPORTED:
4881 adapter->promisc_supported =
4882 be64_to_cpu(crq->query_capability.number);
4883 netdev_dbg(netdev, "promisc_supported = %lld\n",
4884 adapter->promisc_supported);
4885 break;
4886 case MIN_MTU:
4887 adapter->min_mtu = be64_to_cpu(crq->query_capability.number);
4888 netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
4889 netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu);
4890 break;
4891 case MAX_MTU:
4892 adapter->max_mtu = be64_to_cpu(crq->query_capability.number);
4893 netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
4894 netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu);
4895 break;
4896 case MAX_MULTICAST_FILTERS:
4897 adapter->max_multicast_filters =
4898 be64_to_cpu(crq->query_capability.number);
4899 netdev_dbg(netdev, "max_multicast_filters = %lld\n",
4900 adapter->max_multicast_filters);
4901 break;
4902 case VLAN_HEADER_INSERTION:
4903 adapter->vlan_header_insertion =
4904 be64_to_cpu(crq->query_capability.number);
4905 if (adapter->vlan_header_insertion)
4906 netdev->features |= NETIF_F_HW_VLAN_STAG_TX;
4907 netdev_dbg(netdev, "vlan_header_insertion = %lld\n",
4908 adapter->vlan_header_insertion);
4909 break;
4910 case RX_VLAN_HEADER_INSERTION:
4911 adapter->rx_vlan_header_insertion =
4912 be64_to_cpu(crq->query_capability.number);
4913 netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n",
4914 adapter->rx_vlan_header_insertion);
4915 break;
4916 case MAX_TX_SG_ENTRIES:
4917 adapter->max_tx_sg_entries =
4918 be64_to_cpu(crq->query_capability.number);
4919 netdev_dbg(netdev, "max_tx_sg_entries = %lld\n",
4920 adapter->max_tx_sg_entries);
4921 break;
4922 case RX_SG_SUPPORTED:
4923 adapter->rx_sg_supported =
4924 be64_to_cpu(crq->query_capability.number);
4925 netdev_dbg(netdev, "rx_sg_supported = %lld\n",
4926 adapter->rx_sg_supported);
4927 break;
4928 case OPT_TX_COMP_SUB_QUEUES:
4929 adapter->opt_tx_comp_sub_queues =
4930 be64_to_cpu(crq->query_capability.number);
4931 netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n",
4932 adapter->opt_tx_comp_sub_queues);
4933 break;
4934 case OPT_RX_COMP_QUEUES:
4935 adapter->opt_rx_comp_queues =
4936 be64_to_cpu(crq->query_capability.number);
4937 netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n",
4938 adapter->opt_rx_comp_queues);
4939 break;
4940 case OPT_RX_BUFADD_Q_PER_RX_COMP_Q:
4941 adapter->opt_rx_bufadd_q_per_rx_comp_q =
4942 be64_to_cpu(crq->query_capability.number);
4943 netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n",
4944 adapter->opt_rx_bufadd_q_per_rx_comp_q);
4945 break;
4946 case OPT_TX_ENTRIES_PER_SUBCRQ:
4947 adapter->opt_tx_entries_per_subcrq =
4948 be64_to_cpu(crq->query_capability.number);
4949 netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n",
4950 adapter->opt_tx_entries_per_subcrq);
4951 break;
4952 case OPT_RXBA_ENTRIES_PER_SUBCRQ:
4953 adapter->opt_rxba_entries_per_subcrq =
4954 be64_to_cpu(crq->query_capability.number);
4955 netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n",
4956 adapter->opt_rxba_entries_per_subcrq);
4957 break;
4958 case TX_RX_DESC_REQ:
4959 adapter->tx_rx_desc_req = crq->query_capability.number;
4960 netdev_dbg(netdev, "tx_rx_desc_req = %llx\n",
4961 adapter->tx_rx_desc_req);
4962 break;
4963
4964 default:
4965 netdev_err(netdev, "Got invalid cap rsp %d\n",
4966 crq->query_capability.capability);
4967 }
4968
4969 out:
4970 if (atomic_read(&adapter->running_cap_crqs) == 0) {
4971 adapter->wait_capability = false;
4972 send_request_cap(adapter, 0);
4973 }
4974 }
4975
4976 static int send_query_phys_parms(struct ibmvnic_adapter *adapter)
4977 {
4978 union ibmvnic_crq crq;
4979 int rc;
4980
4981 memset(&crq, 0, sizeof(crq));
4982 crq.query_phys_parms.first = IBMVNIC_CRQ_CMD;
4983 crq.query_phys_parms.cmd = QUERY_PHYS_PARMS;
4984
4985 mutex_lock(&adapter->fw_lock);
4986 adapter->fw_done_rc = 0;
4987 reinit_completion(&adapter->fw_done);
4988
4989 rc = ibmvnic_send_crq(adapter, &crq);
4990 if (rc) {
4991 mutex_unlock(&adapter->fw_lock);
4992 return rc;
4993 }
4994
4995 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
4996 if (rc) {
4997 mutex_unlock(&adapter->fw_lock);
4998 return rc;
4999 }
5000
5001 mutex_unlock(&adapter->fw_lock);
5002 return adapter->fw_done_rc ? -EIO : 0;
5003 }
5004
5005 static int handle_query_phys_parms_rsp(union ibmvnic_crq *crq,
5006 struct ibmvnic_adapter *adapter)
5007 {
5008 struct net_device *netdev = adapter->netdev;
5009 int rc;
5010 __be32 rspeed = cpu_to_be32(crq->query_phys_parms_rsp.speed);
5011
5012 rc = crq->query_phys_parms_rsp.rc.code;
5013 if (rc) {
5014 netdev_err(netdev, "Error %d in QUERY_PHYS_PARMS\n", rc);
5015 return rc;
5016 }
5017 switch (rspeed) {
5018 case IBMVNIC_10MBPS:
5019 adapter->speed = SPEED_10;
5020 break;
5021 case IBMVNIC_100MBPS:
5022 adapter->speed = SPEED_100;
5023 break;
5024 case IBMVNIC_1GBPS:
5025 adapter->speed = SPEED_1000;
5026 break;
5027 case IBMVNIC_10GBPS:
5028 adapter->speed = SPEED_10000;
5029 break;
5030 case IBMVNIC_25GBPS:
5031 adapter->speed = SPEED_25000;
5032 break;
5033 case IBMVNIC_40GBPS:
5034 adapter->speed = SPEED_40000;
5035 break;
5036 case IBMVNIC_50GBPS:
5037 adapter->speed = SPEED_50000;
5038 break;
5039 case IBMVNIC_100GBPS:
5040 adapter->speed = SPEED_100000;
5041 break;
5042 case IBMVNIC_200GBPS:
5043 adapter->speed = SPEED_200000;
5044 break;
5045 default:
5046 if (netif_carrier_ok(netdev))
5047 netdev_warn(netdev, "Unknown speed 0x%08x\n", rspeed);
5048 adapter->speed = SPEED_UNKNOWN;
5049 }
5050 if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_FULL_DUPLEX)
5051 adapter->duplex = DUPLEX_FULL;
5052 else if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_HALF_DUPLEX)
5053 adapter->duplex = DUPLEX_HALF;
5054 else
5055 adapter->duplex = DUPLEX_UNKNOWN;
5056
5057 return rc;
5058 }
5059
5060 static void ibmvnic_handle_crq(union ibmvnic_crq *crq,
5061 struct ibmvnic_adapter *adapter)
5062 {
5063 struct ibmvnic_generic_crq *gen_crq = &crq->generic;
5064 struct net_device *netdev = adapter->netdev;
5065 struct device *dev = &adapter->vdev->dev;
5066 u64 *u64_crq = (u64 *)crq;
5067 long rc;
5068
5069 netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n",
5070 (unsigned long)cpu_to_be64(u64_crq[0]),
5071 (unsigned long)cpu_to_be64(u64_crq[1]));
5072 switch (gen_crq->first) {
5073 case IBMVNIC_CRQ_INIT_RSP:
5074 switch (gen_crq->cmd) {
5075 case IBMVNIC_CRQ_INIT:
5076 dev_info(dev, "Partner initialized\n");
5077 adapter->from_passive_init = true;
5078 /* Discard any stale login responses from prev reset.
5079 * CHECK: should we clear even on INIT_COMPLETE?
5080 */
5081 adapter->login_pending = false;
5082
5083 if (adapter->state == VNIC_DOWN)
5084 rc = ibmvnic_reset(adapter, VNIC_RESET_PASSIVE_INIT);
5085 else
5086 rc = ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
5087
5088 if (rc && rc != -EBUSY) {
5089 /* We were unable to schedule the failover
5090 * reset either because the adapter was still
5091 * probing (eg: during kexec) or we could not
5092 * allocate memory. Clear the failover_pending
5093 * flag since no one else will. We ignore
5094 * EBUSY because it means either FAILOVER reset
5095 * is already scheduled or the adapter is
5096 * being removed.
5097 */
5098 netdev_err(netdev,
5099 "Error %ld scheduling failover reset\n",
5100 rc);
5101 adapter->failover_pending = false;
5102 }
5103
5104 if (!completion_done(&adapter->init_done)) {
5105 complete(&adapter->init_done);
5106 if (!adapter->init_done_rc)
5107 adapter->init_done_rc = -EAGAIN;
5108 }
5109
5110 break;
5111 case IBMVNIC_CRQ_INIT_COMPLETE:
5112 dev_info(dev, "Partner initialization complete\n");
5113 adapter->crq.active = true;
5114 send_version_xchg(adapter);
5115 break;
5116 default:
5117 dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd);
5118 }
5119 return;
5120 case IBMVNIC_CRQ_XPORT_EVENT:
5121 netif_carrier_off(netdev);
5122 adapter->crq.active = false;
5123 /* terminate any thread waiting for a response
5124 * from the device
5125 */
5126 if (!completion_done(&adapter->fw_done)) {
5127 adapter->fw_done_rc = -EIO;
5128 complete(&adapter->fw_done);
5129 }
5130 if (!completion_done(&adapter->stats_done))
5131 complete(&adapter->stats_done);
5132 if (test_bit(0, &adapter->resetting))
5133 adapter->force_reset_recovery = true;
5134 if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) {
5135 dev_info(dev, "Migrated, re-enabling adapter\n");
5136 ibmvnic_reset(adapter, VNIC_RESET_MOBILITY);
5137 } else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) {
5138 dev_info(dev, "Backing device failover detected\n");
5139 adapter->failover_pending = true;
5140 } else {
5141 /* The adapter lost the connection */
5142 dev_err(dev, "Virtual Adapter failed (rc=%d)\n",
5143 gen_crq->cmd);
5144 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
5145 }
5146 return;
5147 case IBMVNIC_CRQ_CMD_RSP:
5148 break;
5149 default:
5150 dev_err(dev, "Got an invalid msg type 0x%02x\n",
5151 gen_crq->first);
5152 return;
5153 }
5154
5155 switch (gen_crq->cmd) {
5156 case VERSION_EXCHANGE_RSP:
5157 rc = crq->version_exchange_rsp.rc.code;
5158 if (rc) {
5159 dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc);
5160 break;
5161 }
5162 ibmvnic_version =
5163 be16_to_cpu(crq->version_exchange_rsp.version);
5164 dev_info(dev, "Partner protocol version is %d\n",
5165 ibmvnic_version);
5166 send_query_cap(adapter);
5167 break;
5168 case QUERY_CAPABILITY_RSP:
5169 handle_query_cap_rsp(crq, adapter);
5170 break;
5171 case QUERY_MAP_RSP:
5172 handle_query_map_rsp(crq, adapter);
5173 break;
5174 case REQUEST_MAP_RSP:
5175 adapter->fw_done_rc = crq->request_map_rsp.rc.code;
5176 complete(&adapter->fw_done);
5177 break;
5178 case REQUEST_UNMAP_RSP:
5179 handle_request_unmap_rsp(crq, adapter);
5180 break;
5181 case REQUEST_CAPABILITY_RSP:
5182 handle_request_cap_rsp(crq, adapter);
5183 break;
5184 case LOGIN_RSP:
5185 netdev_dbg(netdev, "Got Login Response\n");
5186 handle_login_rsp(crq, adapter);
5187 break;
5188 case LOGICAL_LINK_STATE_RSP:
5189 netdev_dbg(netdev,
5190 "Got Logical Link State Response, state: %d rc: %d\n",
5191 crq->logical_link_state_rsp.link_state,
5192 crq->logical_link_state_rsp.rc.code);
5193 adapter->logical_link_state =
5194 crq->logical_link_state_rsp.link_state;
5195 adapter->init_done_rc = crq->logical_link_state_rsp.rc.code;
5196 complete(&adapter->init_done);
5197 break;
5198 case LINK_STATE_INDICATION:
5199 netdev_dbg(netdev, "Got Logical Link State Indication\n");
5200 adapter->phys_link_state =
5201 crq->link_state_indication.phys_link_state;
5202 adapter->logical_link_state =
5203 crq->link_state_indication.logical_link_state;
5204 if (adapter->phys_link_state && adapter->logical_link_state)
5205 netif_carrier_on(netdev);
5206 else
5207 netif_carrier_off(netdev);
5208 break;
5209 case CHANGE_MAC_ADDR_RSP:
5210 netdev_dbg(netdev, "Got MAC address change Response\n");
5211 adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter);
5212 break;
5213 case ERROR_INDICATION:
5214 netdev_dbg(netdev, "Got Error Indication\n");
5215 handle_error_indication(crq, adapter);
5216 break;
5217 case REQUEST_STATISTICS_RSP:
5218 netdev_dbg(netdev, "Got Statistics Response\n");
5219 complete(&adapter->stats_done);
5220 break;
5221 case QUERY_IP_OFFLOAD_RSP:
5222 netdev_dbg(netdev, "Got Query IP offload Response\n");
5223 handle_query_ip_offload_rsp(adapter);
5224 break;
5225 case MULTICAST_CTRL_RSP:
5226 netdev_dbg(netdev, "Got multicast control Response\n");
5227 break;
5228 case CONTROL_IP_OFFLOAD_RSP:
5229 netdev_dbg(netdev, "Got Control IP offload Response\n");
5230 dma_unmap_single(dev, adapter->ip_offload_ctrl_tok,
5231 sizeof(adapter->ip_offload_ctrl),
5232 DMA_TO_DEVICE);
5233 complete(&adapter->init_done);
5234 break;
5235 case COLLECT_FW_TRACE_RSP:
5236 netdev_dbg(netdev, "Got Collect firmware trace Response\n");
5237 complete(&adapter->fw_done);
5238 break;
5239 case GET_VPD_SIZE_RSP:
5240 handle_vpd_size_rsp(crq, adapter);
5241 break;
5242 case GET_VPD_RSP:
5243 handle_vpd_rsp(crq, adapter);
5244 break;
5245 case QUERY_PHYS_PARMS_RSP:
5246 adapter->fw_done_rc = handle_query_phys_parms_rsp(crq, adapter);
5247 complete(&adapter->fw_done);
5248 break;
5249 default:
5250 netdev_err(netdev, "Got an invalid cmd type 0x%02x\n",
5251 gen_crq->cmd);
5252 }
5253 }
5254
5255 static irqreturn_t ibmvnic_interrupt(int irq, void *instance)
5256 {
5257 struct ibmvnic_adapter *adapter = instance;
5258
5259 tasklet_schedule(&adapter->tasklet);
5260 return IRQ_HANDLED;
5261 }
5262
5263 static void ibmvnic_tasklet(struct tasklet_struct *t)
5264 {
5265 struct ibmvnic_adapter *adapter = from_tasklet(adapter, t, tasklet);
5266 struct ibmvnic_crq_queue *queue = &adapter->crq;
5267 union ibmvnic_crq *crq;
5268 unsigned long flags;
5269 bool done = false;
5270
5271 spin_lock_irqsave(&queue->lock, flags);
5272 while (!done) {
5273 /* Pull all the valid messages off the CRQ */
5274 while ((crq = ibmvnic_next_crq(adapter)) != NULL) {
5275 /* This barrier makes sure ibmvnic_next_crq()'s
5276 * crq->generic.first & IBMVNIC_CRQ_CMD_RSP is loaded
5277 * before ibmvnic_handle_crq()'s
5278 * switch(gen_crq->first) and switch(gen_crq->cmd).
5279 */
5280 dma_rmb();
5281 ibmvnic_handle_crq(crq, adapter);
5282 crq->generic.first = 0;
5283 }
5284
5285 /* remain in tasklet until all
5286 * capabilities responses are received
5287 */
5288 if (!adapter->wait_capability)
5289 done = true;
5290 }
5291 /* if capabilities CRQ's were sent in this tasklet, the following
5292 * tasklet must wait until all responses are received
5293 */
5294 if (atomic_read(&adapter->running_cap_crqs) != 0)
5295 adapter->wait_capability = true;
5296 spin_unlock_irqrestore(&queue->lock, flags);
5297 }
5298
5299 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter)
5300 {
5301 struct vio_dev *vdev = adapter->vdev;
5302 int rc;
5303
5304 do {
5305 rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
5306 } while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc));
5307
5308 if (rc)
5309 dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc);
5310
5311 return rc;
5312 }
5313
5314 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter)
5315 {
5316 struct ibmvnic_crq_queue *crq = &adapter->crq;
5317 struct device *dev = &adapter->vdev->dev;
5318 struct vio_dev *vdev = adapter->vdev;
5319 int rc;
5320
5321 /* Close the CRQ */
5322 do {
5323 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5324 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5325
5326 /* Clean out the queue */
5327 if (!crq->msgs)
5328 return -EINVAL;
5329
5330 memset(crq->msgs, 0, PAGE_SIZE);
5331 crq->cur = 0;
5332 crq->active = false;
5333
5334 /* And re-open it again */
5335 rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5336 crq->msg_token, PAGE_SIZE);
5337
5338 if (rc == H_CLOSED)
5339 /* Adapter is good, but other end is not ready */
5340 dev_warn(dev, "Partner adapter not ready\n");
5341 else if (rc != 0)
5342 dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc);
5343
5344 return rc;
5345 }
5346
5347 static void release_crq_queue(struct ibmvnic_adapter *adapter)
5348 {
5349 struct ibmvnic_crq_queue *crq = &adapter->crq;
5350 struct vio_dev *vdev = adapter->vdev;
5351 long rc;
5352
5353 if (!crq->msgs)
5354 return;
5355
5356 netdev_dbg(adapter->netdev, "Releasing CRQ\n");
5357 free_irq(vdev->irq, adapter);
5358 tasklet_kill(&adapter->tasklet);
5359 do {
5360 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5361 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5362
5363 dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE,
5364 DMA_BIDIRECTIONAL);
5365 free_page((unsigned long)crq->msgs);
5366 crq->msgs = NULL;
5367 crq->active = false;
5368 }
5369
5370 static int init_crq_queue(struct ibmvnic_adapter *adapter)
5371 {
5372 struct ibmvnic_crq_queue *crq = &adapter->crq;
5373 struct device *dev = &adapter->vdev->dev;
5374 struct vio_dev *vdev = adapter->vdev;
5375 int rc, retrc = -ENOMEM;
5376
5377 if (crq->msgs)
5378 return 0;
5379
5380 crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL);
5381 /* Should we allocate more than one page? */
5382
5383 if (!crq->msgs)
5384 return -ENOMEM;
5385
5386 crq->size = PAGE_SIZE / sizeof(*crq->msgs);
5387 crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE,
5388 DMA_BIDIRECTIONAL);
5389 if (dma_mapping_error(dev, crq->msg_token))
5390 goto map_failed;
5391
5392 rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5393 crq->msg_token, PAGE_SIZE);
5394
5395 if (rc == H_RESOURCE)
5396 /* maybe kexecing and resource is busy. try a reset */
5397 rc = ibmvnic_reset_crq(adapter);
5398 retrc = rc;
5399
5400 if (rc == H_CLOSED) {
5401 dev_warn(dev, "Partner adapter not ready\n");
5402 } else if (rc) {
5403 dev_warn(dev, "Error %d opening adapter\n", rc);
5404 goto reg_crq_failed;
5405 }
5406
5407 retrc = 0;
5408
5409 tasklet_setup(&adapter->tasklet, (void *)ibmvnic_tasklet);
5410
5411 netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq);
5412 snprintf(crq->name, sizeof(crq->name), "ibmvnic-%x",
5413 adapter->vdev->unit_address);
5414 rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, crq->name, adapter);
5415 if (rc) {
5416 dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n",
5417 vdev->irq, rc);
5418 goto req_irq_failed;
5419 }
5420
5421 rc = vio_enable_interrupts(vdev);
5422 if (rc) {
5423 dev_err(dev, "Error %d enabling interrupts\n", rc);
5424 goto req_irq_failed;
5425 }
5426
5427 crq->cur = 0;
5428 spin_lock_init(&crq->lock);
5429
5430 /* process any CRQs that were queued before we enabled interrupts */
5431 tasklet_schedule(&adapter->tasklet);
5432
5433 return retrc;
5434
5435 req_irq_failed:
5436 tasklet_kill(&adapter->tasklet);
5437 do {
5438 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5439 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5440 reg_crq_failed:
5441 dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL);
5442 map_failed:
5443 free_page((unsigned long)crq->msgs);
5444 crq->msgs = NULL;
5445 return retrc;
5446 }
5447
5448 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset)
5449 {
5450 struct device *dev = &adapter->vdev->dev;
5451 unsigned long timeout = msecs_to_jiffies(20000);
5452 u64 old_num_rx_queues = adapter->req_rx_queues;
5453 u64 old_num_tx_queues = adapter->req_tx_queues;
5454 int rc;
5455
5456 adapter->from_passive_init = false;
5457
5458 if (reset)
5459 reinit_completion(&adapter->init_done);
5460
5461 adapter->init_done_rc = 0;
5462 rc = ibmvnic_send_crq_init(adapter);
5463 if (rc) {
5464 dev_err(dev, "Send crq init failed with error %d\n", rc);
5465 return rc;
5466 }
5467
5468 if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
5469 dev_err(dev, "Initialization sequence timed out\n");
5470 return -1;
5471 }
5472
5473 if (adapter->init_done_rc) {
5474 release_crq_queue(adapter);
5475 return adapter->init_done_rc;
5476 }
5477
5478 if (adapter->from_passive_init) {
5479 adapter->state = VNIC_OPEN;
5480 adapter->from_passive_init = false;
5481 return -1;
5482 }
5483
5484 if (reset &&
5485 test_bit(0, &adapter->resetting) && !adapter->wait_for_reset &&
5486 adapter->reset_reason != VNIC_RESET_MOBILITY) {
5487 if (adapter->req_rx_queues != old_num_rx_queues ||
5488 adapter->req_tx_queues != old_num_tx_queues) {
5489 release_sub_crqs(adapter, 0);
5490 rc = init_sub_crqs(adapter);
5491 } else {
5492 rc = reset_sub_crq_queues(adapter);
5493 }
5494 } else {
5495 rc = init_sub_crqs(adapter);
5496 }
5497
5498 if (rc) {
5499 dev_err(dev, "Initialization of sub crqs failed\n");
5500 release_crq_queue(adapter);
5501 return rc;
5502 }
5503
5504 rc = init_sub_crq_irqs(adapter);
5505 if (rc) {
5506 dev_err(dev, "Failed to initialize sub crq irqs\n");
5507 release_crq_queue(adapter);
5508 }
5509
5510 return rc;
5511 }
5512
5513 static struct device_attribute dev_attr_failover;
5514
5515 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
5516 {
5517 struct ibmvnic_adapter *adapter;
5518 struct net_device *netdev;
5519 unsigned char *mac_addr_p;
5520 bool init_success;
5521 int rc;
5522
5523 dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n",
5524 dev->unit_address);
5525
5526 mac_addr_p = (unsigned char *)vio_get_attribute(dev,
5527 VETH_MAC_ADDR, NULL);
5528 if (!mac_addr_p) {
5529 dev_err(&dev->dev,
5530 "(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n",
5531 __FILE__, __LINE__);
5532 return 0;
5533 }
5534
5535 netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter),
5536 IBMVNIC_MAX_QUEUES);
5537 if (!netdev)
5538 return -ENOMEM;
5539
5540 adapter = netdev_priv(netdev);
5541 adapter->state = VNIC_PROBING;
5542 dev_set_drvdata(&dev->dev, netdev);
5543 adapter->vdev = dev;
5544 adapter->netdev = netdev;
5545 adapter->login_pending = false;
5546
5547 ether_addr_copy(adapter->mac_addr, mac_addr_p);
5548 ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
5549 netdev->irq = dev->irq;
5550 netdev->netdev_ops = &ibmvnic_netdev_ops;
5551 netdev->ethtool_ops = &ibmvnic_ethtool_ops;
5552 SET_NETDEV_DEV(netdev, &dev->dev);
5553
5554 INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset);
5555 INIT_DELAYED_WORK(&adapter->ibmvnic_delayed_reset,
5556 __ibmvnic_delayed_reset);
5557 INIT_LIST_HEAD(&adapter->rwi_list);
5558 spin_lock_init(&adapter->rwi_lock);
5559 spin_lock_init(&adapter->state_lock);
5560 mutex_init(&adapter->fw_lock);
5561 init_completion(&adapter->init_done);
5562 init_completion(&adapter->fw_done);
5563 init_completion(&adapter->reset_done);
5564 init_completion(&adapter->stats_done);
5565 clear_bit(0, &adapter->resetting);
5566
5567 init_success = false;
5568 do {
5569 rc = init_crq_queue(adapter);
5570 if (rc) {
5571 dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n",
5572 rc);
5573 goto ibmvnic_init_fail;
5574 }
5575
5576 rc = ibmvnic_reset_init(adapter, false);
5577 } while (rc == -EAGAIN);
5578
5579 /* We are ignoring the error from ibmvnic_reset_init() assuming that the
5580 * partner is not ready. CRQ is not active. When the partner becomes
5581 * ready, we will do the passive init reset.
5582 */
5583
5584 if (!rc)
5585 init_success = true;
5586
5587 rc = init_stats_buffers(adapter);
5588 if (rc)
5589 goto ibmvnic_init_fail;
5590
5591 rc = init_stats_token(adapter);
5592 if (rc)
5593 goto ibmvnic_stats_fail;
5594
5595 rc = device_create_file(&dev->dev, &dev_attr_failover);
5596 if (rc)
5597 goto ibmvnic_dev_file_err;
5598
5599 netif_carrier_off(netdev);
5600 rc = register_netdev(netdev);
5601 if (rc) {
5602 dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc);
5603 goto ibmvnic_register_fail;
5604 }
5605 dev_info(&dev->dev, "ibmvnic registered\n");
5606
5607 if (init_success) {
5608 adapter->state = VNIC_PROBED;
5609 netdev->mtu = adapter->req_mtu - ETH_HLEN;
5610 netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
5611 netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
5612 } else {
5613 adapter->state = VNIC_DOWN;
5614 }
5615
5616 adapter->wait_for_reset = false;
5617 adapter->last_reset_time = jiffies;
5618 return 0;
5619
5620 ibmvnic_register_fail:
5621 device_remove_file(&dev->dev, &dev_attr_failover);
5622
5623 ibmvnic_dev_file_err:
5624 release_stats_token(adapter);
5625
5626 ibmvnic_stats_fail:
5627 release_stats_buffers(adapter);
5628
5629 ibmvnic_init_fail:
5630 release_sub_crqs(adapter, 1);
5631 release_crq_queue(adapter);
5632 mutex_destroy(&adapter->fw_lock);
5633 free_netdev(netdev);
5634
5635 return rc;
5636 }
5637
5638 static void ibmvnic_remove(struct vio_dev *dev)
5639 {
5640 struct net_device *netdev = dev_get_drvdata(&dev->dev);
5641 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5642 unsigned long flags;
5643
5644 spin_lock_irqsave(&adapter->state_lock, flags);
5645
5646 /* If ibmvnic_reset() is scheduling a reset, wait for it to
5647 * finish. Then, set the state to REMOVING to prevent it from
5648 * scheduling any more work and to have reset functions ignore
5649 * any resets that have already been scheduled. Drop the lock
5650 * after setting state, so __ibmvnic_reset() which is called
5651 * from the flush_work() below, can make progress.
5652 */
5653 spin_lock(&adapter->rwi_lock);
5654 adapter->state = VNIC_REMOVING;
5655 spin_unlock(&adapter->rwi_lock);
5656
5657 spin_unlock_irqrestore(&adapter->state_lock, flags);
5658
5659 flush_work(&adapter->ibmvnic_reset);
5660 flush_delayed_work(&adapter->ibmvnic_delayed_reset);
5661
5662 rtnl_lock();
5663 unregister_netdevice(netdev);
5664
5665 release_resources(adapter);
5666 release_sub_crqs(adapter, 1);
5667 release_crq_queue(adapter);
5668
5669 release_stats_token(adapter);
5670 release_stats_buffers(adapter);
5671
5672 adapter->state = VNIC_REMOVED;
5673
5674 rtnl_unlock();
5675 mutex_destroy(&adapter->fw_lock);
5676 device_remove_file(&dev->dev, &dev_attr_failover);
5677 free_netdev(netdev);
5678 dev_set_drvdata(&dev->dev, NULL);
5679 }
5680
5681 static ssize_t failover_store(struct device *dev, struct device_attribute *attr,
5682 const char *buf, size_t count)
5683 {
5684 struct net_device *netdev = dev_get_drvdata(dev);
5685 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5686 unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
5687 __be64 session_token;
5688 long rc;
5689
5690 if (!sysfs_streq(buf, "1"))
5691 return -EINVAL;
5692
5693 rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address,
5694 H_GET_SESSION_TOKEN, 0, 0, 0);
5695 if (rc) {
5696 netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n",
5697 rc);
5698 goto last_resort;
5699 }
5700
5701 session_token = (__be64)retbuf[0];
5702 netdev_dbg(netdev, "Initiating client failover, session id %llx\n",
5703 be64_to_cpu(session_token));
5704 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
5705 H_SESSION_ERR_DETECTED, session_token, 0, 0);
5706 if (rc)
5707 netdev_err(netdev,
5708 "H_VIOCTL initiated failover failed, rc %ld\n",
5709 rc);
5710
5711 last_resort:
5712 netdev_dbg(netdev, "Trying to send CRQ_CMD, the last resort\n");
5713 ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
5714
5715 return count;
5716 }
5717 static DEVICE_ATTR_WO(failover);
5718
5719 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev)
5720 {
5721 struct net_device *netdev = dev_get_drvdata(&vdev->dev);
5722 struct ibmvnic_adapter *adapter;
5723 struct iommu_table *tbl;
5724 unsigned long ret = 0;
5725 int i;
5726
5727 tbl = get_iommu_table_base(&vdev->dev);
5728
5729 /* netdev inits at probe time along with the structures we need below*/
5730 if (!netdev)
5731 return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl);
5732
5733 adapter = netdev_priv(netdev);
5734
5735 ret += PAGE_SIZE; /* the crq message queue */
5736 ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl);
5737
5738 for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++)
5739 ret += 4 * PAGE_SIZE; /* the scrq message queue */
5740
5741 for (i = 0; i < adapter->num_active_rx_pools; i++)
5742 ret += adapter->rx_pool[i].size *
5743 IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl);
5744
5745 return ret;
5746 }
5747
5748 static int ibmvnic_resume(struct device *dev)
5749 {
5750 struct net_device *netdev = dev_get_drvdata(dev);
5751 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5752
5753 if (adapter->state != VNIC_OPEN)
5754 return 0;
5755
5756 tasklet_schedule(&adapter->tasklet);
5757
5758 return 0;
5759 }
5760
5761 static const struct vio_device_id ibmvnic_device_table[] = {
5762 {"network", "IBM,vnic"},
5763 {"", "" }
5764 };
5765 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table);
5766
5767 static const struct dev_pm_ops ibmvnic_pm_ops = {
5768 .resume = ibmvnic_resume
5769 };
5770
5771 static struct vio_driver ibmvnic_driver = {
5772 .id_table = ibmvnic_device_table,
5773 .probe = ibmvnic_probe,
5774 .remove = ibmvnic_remove,
5775 .get_desired_dma = ibmvnic_get_desired_dma,
5776 .name = ibmvnic_driver_name,
5777 .pm = &ibmvnic_pm_ops,
5778 };
5779
5780 /* module functions */
5781 static int __init ibmvnic_module_init(void)
5782 {
5783 pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string,
5784 IBMVNIC_DRIVER_VERSION);
5785
5786 return vio_register_driver(&ibmvnic_driver);
5787 }
5788
5789 static void __exit ibmvnic_module_exit(void)
5790 {
5791 vio_unregister_driver(&ibmvnic_driver);
5792 }
5793
5794 module_init(ibmvnic_module_init);
5795 module_exit(ibmvnic_module_exit);
Ubuntu Jammy Linux kernel mirror