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/irqdomain.h>
57 #include <linux/kthread.h>
58 #include <linux/seq_file.h>
59 #include <linux/interrupt.h>
60 #include <net/net_namespace.h>
61 #include <asm/hvcall.h>
62 #include <linux/atomic.h>
63 #include <asm/vio.h>
64 #include <asm/xive.h>
65 #include <asm/iommu.h>
66 #include <linux/uaccess.h>
67 #include <asm/firmware.h>
68 #include <linux/workqueue.h>
69 #include <linux/if_vlan.h>
70 #include <linux/utsname.h>
71 #include <linux/cpu.h>
72
73 #include "ibmvnic.h"
74
75 static const char ibmvnic_driver_name[] = "ibmvnic";
76 static const char ibmvnic_driver_string[] = "IBM System i/p Virtual NIC Driver";
77
78 MODULE_AUTHOR("Santiago Leon");
79 MODULE_DESCRIPTION("IBM System i/p Virtual NIC Driver");
80 MODULE_LICENSE("GPL");
81 MODULE_VERSION(IBMVNIC_DRIVER_VERSION);
82
83 static int ibmvnic_version = IBMVNIC_INITIAL_VERSION;
84 static void release_sub_crqs(struct ibmvnic_adapter *, bool);
85 static int ibmvnic_reset_crq(struct ibmvnic_adapter *);
86 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *);
87 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *);
88 static int ibmvnic_send_crq(struct ibmvnic_adapter *, union ibmvnic_crq *);
89 static int send_subcrq_indirect(struct ibmvnic_adapter *, u64, u64, u64);
90 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance);
91 static int enable_scrq_irq(struct ibmvnic_adapter *,
92 struct ibmvnic_sub_crq_queue *);
93 static int disable_scrq_irq(struct ibmvnic_adapter *,
94 struct ibmvnic_sub_crq_queue *);
95 static int pending_scrq(struct ibmvnic_adapter *,
96 struct ibmvnic_sub_crq_queue *);
97 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *,
98 struct ibmvnic_sub_crq_queue *);
99 static int ibmvnic_poll(struct napi_struct *napi, int data);
100 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter);
101 static inline void reinit_init_done(struct ibmvnic_adapter *adapter);
102 static void send_query_map(struct ibmvnic_adapter *adapter);
103 static int send_request_map(struct ibmvnic_adapter *, dma_addr_t, u32, u8);
104 static int send_request_unmap(struct ibmvnic_adapter *, u8);
105 static int send_login(struct ibmvnic_adapter *adapter);
106 static void send_query_cap(struct ibmvnic_adapter *adapter);
107 static int init_sub_crqs(struct ibmvnic_adapter *);
108 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter);
109 static int ibmvnic_reset_init(struct ibmvnic_adapter *, bool reset);
110 static void release_crq_queue(struct ibmvnic_adapter *);
111 static int __ibmvnic_set_mac(struct net_device *, u8 *);
112 static int init_crq_queue(struct ibmvnic_adapter *adapter);
113 static int send_query_phys_parms(struct ibmvnic_adapter *adapter);
114 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
115 struct ibmvnic_sub_crq_queue *tx_scrq);
116 static void free_long_term_buff(struct ibmvnic_adapter *adapter,
117 struct ibmvnic_long_term_buff *ltb);
118 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter);
119 static void flush_reset_queue(struct ibmvnic_adapter *adapter);
120 static void print_subcrq_error(struct device *dev, int rc, const char *func);
121
122 struct ibmvnic_stat {
123 char name[ETH_GSTRING_LEN];
124 int offset;
125 };
126
127 #define IBMVNIC_STAT_OFF(stat) (offsetof(struct ibmvnic_adapter, stats) + \
128 offsetof(struct ibmvnic_statistics, stat))
129 #define IBMVNIC_GET_STAT(a, off) (*((u64 *)(((unsigned long)(a)) + (off))))
130
131 static const struct ibmvnic_stat ibmvnic_stats[] = {
132 {"rx_packets", IBMVNIC_STAT_OFF(rx_packets)},
133 {"rx_bytes", IBMVNIC_STAT_OFF(rx_bytes)},
134 {"tx_packets", IBMVNIC_STAT_OFF(tx_packets)},
135 {"tx_bytes", IBMVNIC_STAT_OFF(tx_bytes)},
136 {"ucast_tx_packets", IBMVNIC_STAT_OFF(ucast_tx_packets)},
137 {"ucast_rx_packets", IBMVNIC_STAT_OFF(ucast_rx_packets)},
138 {"mcast_tx_packets", IBMVNIC_STAT_OFF(mcast_tx_packets)},
139 {"mcast_rx_packets", IBMVNIC_STAT_OFF(mcast_rx_packets)},
140 {"bcast_tx_packets", IBMVNIC_STAT_OFF(bcast_tx_packets)},
141 {"bcast_rx_packets", IBMVNIC_STAT_OFF(bcast_rx_packets)},
142 {"align_errors", IBMVNIC_STAT_OFF(align_errors)},
143 {"fcs_errors", IBMVNIC_STAT_OFF(fcs_errors)},
144 {"single_collision_frames", IBMVNIC_STAT_OFF(single_collision_frames)},
145 {"multi_collision_frames", IBMVNIC_STAT_OFF(multi_collision_frames)},
146 {"sqe_test_errors", IBMVNIC_STAT_OFF(sqe_test_errors)},
147 {"deferred_tx", IBMVNIC_STAT_OFF(deferred_tx)},
148 {"late_collisions", IBMVNIC_STAT_OFF(late_collisions)},
149 {"excess_collisions", IBMVNIC_STAT_OFF(excess_collisions)},
150 {"internal_mac_tx_errors", IBMVNIC_STAT_OFF(internal_mac_tx_errors)},
151 {"carrier_sense", IBMVNIC_STAT_OFF(carrier_sense)},
152 {"too_long_frames", IBMVNIC_STAT_OFF(too_long_frames)},
153 {"internal_mac_rx_errors", IBMVNIC_STAT_OFF(internal_mac_rx_errors)},
154 };
155
send_crq_init_complete(struct ibmvnic_adapter * adapter)156 static int send_crq_init_complete(struct ibmvnic_adapter *adapter)
157 {
158 union ibmvnic_crq crq;
159
160 memset(&crq, 0, sizeof(crq));
161 crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
162 crq.generic.cmd = IBMVNIC_CRQ_INIT_COMPLETE;
163
164 return ibmvnic_send_crq(adapter, &crq);
165 }
166
send_version_xchg(struct ibmvnic_adapter * adapter)167 static int send_version_xchg(struct ibmvnic_adapter *adapter)
168 {
169 union ibmvnic_crq crq;
170
171 memset(&crq, 0, sizeof(crq));
172 crq.version_exchange.first = IBMVNIC_CRQ_CMD;
173 crq.version_exchange.cmd = VERSION_EXCHANGE;
174 crq.version_exchange.version = cpu_to_be16(ibmvnic_version);
175
176 return ibmvnic_send_crq(adapter, &crq);
177 }
178
ibmvnic_clean_queue_affinity(struct ibmvnic_adapter * adapter,struct ibmvnic_sub_crq_queue * queue)179 static void ibmvnic_clean_queue_affinity(struct ibmvnic_adapter *adapter,
180 struct ibmvnic_sub_crq_queue *queue)
181 {
182 if (!(queue && queue->irq))
183 return;
184
185 cpumask_clear(queue->affinity_mask);
186
187 if (irq_set_affinity_and_hint(queue->irq, NULL))
188 netdev_warn(adapter->netdev,
189 "%s: Clear affinity failed, queue addr = %p, IRQ = %d\n",
190 __func__, queue, queue->irq);
191 }
192
ibmvnic_clean_affinity(struct ibmvnic_adapter * adapter)193 static void ibmvnic_clean_affinity(struct ibmvnic_adapter *adapter)
194 {
195 struct ibmvnic_sub_crq_queue **rxqs;
196 struct ibmvnic_sub_crq_queue **txqs;
197 int num_rxqs, num_txqs;
198 int i;
199
200 rxqs = adapter->rx_scrq;
201 txqs = adapter->tx_scrq;
202 num_txqs = adapter->num_active_tx_scrqs;
203 num_rxqs = adapter->num_active_rx_scrqs;
204
205 netdev_dbg(adapter->netdev, "%s: Cleaning irq affinity hints", __func__);
206 if (txqs) {
207 for (i = 0; i < num_txqs; i++)
208 ibmvnic_clean_queue_affinity(adapter, txqs[i]);
209 }
210 if (rxqs) {
211 for (i = 0; i < num_rxqs; i++)
212 ibmvnic_clean_queue_affinity(adapter, rxqs[i]);
213 }
214 }
215
ibmvnic_set_queue_affinity(struct ibmvnic_sub_crq_queue * queue,unsigned int * cpu,int * stragglers,int stride)216 static int ibmvnic_set_queue_affinity(struct ibmvnic_sub_crq_queue *queue,
217 unsigned int *cpu, int *stragglers,
218 int stride)
219 {
220 cpumask_var_t mask;
221 int i;
222 int rc = 0;
223
224 if (!(queue && queue->irq))
225 return rc;
226
227 /* cpumask_var_t is either a pointer or array, allocation works here */
228 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
229 return -ENOMEM;
230
231 /* while we have extra cpu give one extra to this irq */
232 if (*stragglers) {
233 stride++;
234 (*stragglers)--;
235 }
236 /* atomic write is safer than writing bit by bit directly */
237 for (i = 0; i < stride; i++) {
238 cpumask_set_cpu(*cpu, mask);
239 *cpu = cpumask_next_wrap(*cpu, cpu_online_mask,
240 nr_cpu_ids, false);
241 }
242 /* set queue affinity mask */
243 cpumask_copy(queue->affinity_mask, mask);
244 rc = irq_set_affinity_and_hint(queue->irq, queue->affinity_mask);
245 free_cpumask_var(mask);
246
247 return rc;
248 }
249
250 /* assumes cpu read lock is held */
ibmvnic_set_affinity(struct ibmvnic_adapter * adapter)251 static void ibmvnic_set_affinity(struct ibmvnic_adapter *adapter)
252 {
253 struct ibmvnic_sub_crq_queue **rxqs = adapter->rx_scrq;
254 struct ibmvnic_sub_crq_queue **txqs = adapter->tx_scrq;
255 struct ibmvnic_sub_crq_queue *queue;
256 int num_rxqs = adapter->num_active_rx_scrqs, i_rxqs = 0;
257 int num_txqs = adapter->num_active_tx_scrqs, i_txqs = 0;
258 int total_queues, stride, stragglers, i;
259 unsigned int num_cpu, cpu;
260 bool is_rx_queue;
261 int rc = 0;
262
263 netdev_dbg(adapter->netdev, "%s: Setting irq affinity hints", __func__);
264 if (!(adapter->rx_scrq && adapter->tx_scrq)) {
265 netdev_warn(adapter->netdev,
266 "%s: Set affinity failed, queues not allocated\n",
267 __func__);
268 return;
269 }
270
271 total_queues = num_rxqs + num_txqs;
272 num_cpu = num_online_cpus();
273 /* number of cpu's assigned per irq */
274 stride = max_t(int, num_cpu / total_queues, 1);
275 /* number of leftover cpu's */
276 stragglers = num_cpu >= total_queues ? num_cpu % total_queues : 0;
277 /* next available cpu to assign irq to */
278 cpu = cpumask_next(-1, cpu_online_mask);
279
280 for (i = 0; i < total_queues; i++) {
281 is_rx_queue = false;
282 /* balance core load by alternating rx and tx assignments
283 * ex: TX0 -> RX0 -> TX1 -> RX1 etc.
284 */
285 if ((i % 2 == 1 && i_rxqs < num_rxqs) || i_txqs == num_txqs) {
286 queue = rxqs[i_rxqs++];
287 is_rx_queue = true;
288 } else {
289 queue = txqs[i_txqs++];
290 }
291
292 rc = ibmvnic_set_queue_affinity(queue, &cpu, &stragglers,
293 stride);
294 if (rc)
295 goto out;
296
297 if (!queue || is_rx_queue)
298 continue;
299
300 rc = __netif_set_xps_queue(adapter->netdev,
301 cpumask_bits(queue->affinity_mask),
302 i_txqs - 1, XPS_CPUS);
303 if (rc)
304 netdev_warn(adapter->netdev, "%s: Set XPS on queue %d failed, rc = %d.\n",
305 __func__, i_txqs - 1, rc);
306 }
307
308 out:
309 if (rc) {
310 netdev_warn(adapter->netdev,
311 "%s: Set affinity failed, queue addr = %p, IRQ = %d, rc = %d.\n",
312 __func__, queue, queue->irq, rc);
313 ibmvnic_clean_affinity(adapter);
314 }
315 }
316
ibmvnic_cpu_online(unsigned int cpu,struct hlist_node * node)317 static int ibmvnic_cpu_online(unsigned int cpu, struct hlist_node *node)
318 {
319 struct ibmvnic_adapter *adapter;
320
321 adapter = hlist_entry_safe(node, struct ibmvnic_adapter, node);
322 ibmvnic_set_affinity(adapter);
323 return 0;
324 }
325
ibmvnic_cpu_dead(unsigned int cpu,struct hlist_node * node)326 static int ibmvnic_cpu_dead(unsigned int cpu, struct hlist_node *node)
327 {
328 struct ibmvnic_adapter *adapter;
329
330 adapter = hlist_entry_safe(node, struct ibmvnic_adapter, node_dead);
331 ibmvnic_set_affinity(adapter);
332 return 0;
333 }
334
ibmvnic_cpu_down_prep(unsigned int cpu,struct hlist_node * node)335 static int ibmvnic_cpu_down_prep(unsigned int cpu, struct hlist_node *node)
336 {
337 struct ibmvnic_adapter *adapter;
338
339 adapter = hlist_entry_safe(node, struct ibmvnic_adapter, node);
340 ibmvnic_clean_affinity(adapter);
341 return 0;
342 }
343
344 static enum cpuhp_state ibmvnic_online;
345
ibmvnic_cpu_notif_add(struct ibmvnic_adapter * adapter)346 static int ibmvnic_cpu_notif_add(struct ibmvnic_adapter *adapter)
347 {
348 int ret;
349
350 ret = cpuhp_state_add_instance_nocalls(ibmvnic_online, &adapter->node);
351 if (ret)
352 return ret;
353 ret = cpuhp_state_add_instance_nocalls(CPUHP_IBMVNIC_DEAD,
354 &adapter->node_dead);
355 if (!ret)
356 return ret;
357 cpuhp_state_remove_instance_nocalls(ibmvnic_online, &adapter->node);
358 return ret;
359 }
360
ibmvnic_cpu_notif_remove(struct ibmvnic_adapter * adapter)361 static void ibmvnic_cpu_notif_remove(struct ibmvnic_adapter *adapter)
362 {
363 cpuhp_state_remove_instance_nocalls(ibmvnic_online, &adapter->node);
364 cpuhp_state_remove_instance_nocalls(CPUHP_IBMVNIC_DEAD,
365 &adapter->node_dead);
366 }
367
h_reg_sub_crq(unsigned long unit_address,unsigned long token,unsigned long length,unsigned long * number,unsigned long * irq)368 static long h_reg_sub_crq(unsigned long unit_address, unsigned long token,
369 unsigned long length, unsigned long *number,
370 unsigned long *irq)
371 {
372 unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
373 long rc;
374
375 rc = plpar_hcall(H_REG_SUB_CRQ, retbuf, unit_address, token, length);
376 *number = retbuf[0];
377 *irq = retbuf[1];
378
379 return rc;
380 }
381
382 /**
383 * ibmvnic_wait_for_completion - Check device state and wait for completion
384 * @adapter: private device data
385 * @comp_done: completion structure to wait for
386 * @timeout: time to wait in milliseconds
387 *
388 * Wait for a completion signal or until the timeout limit is reached
389 * while checking that the device is still active.
390 */
ibmvnic_wait_for_completion(struct ibmvnic_adapter * adapter,struct completion * comp_done,unsigned long timeout)391 static int ibmvnic_wait_for_completion(struct ibmvnic_adapter *adapter,
392 struct completion *comp_done,
393 unsigned long timeout)
394 {
395 struct net_device *netdev;
396 unsigned long div_timeout;
397 u8 retry;
398
399 netdev = adapter->netdev;
400 retry = 5;
401 div_timeout = msecs_to_jiffies(timeout / retry);
402 while (true) {
403 if (!adapter->crq.active) {
404 netdev_err(netdev, "Device down!\n");
405 return -ENODEV;
406 }
407 if (!retry--)
408 break;
409 if (wait_for_completion_timeout(comp_done, div_timeout))
410 return 0;
411 }
412 netdev_err(netdev, "Operation timed out.\n");
413 return -ETIMEDOUT;
414 }
415
416 /**
417 * reuse_ltb() - Check if a long term buffer can be reused
418 * @ltb: The long term buffer to be checked
419 * @size: The size of the long term buffer.
420 *
421 * An LTB can be reused unless its size has changed.
422 *
423 * Return: Return true if the LTB can be reused, false otherwise.
424 */
reuse_ltb(struct ibmvnic_long_term_buff * ltb,int size)425 static bool reuse_ltb(struct ibmvnic_long_term_buff *ltb, int size)
426 {
427 return (ltb->buff && ltb->size == size);
428 }
429
430 /**
431 * alloc_long_term_buff() - Allocate a long term buffer (LTB)
432 *
433 * @adapter: ibmvnic adapter associated to the LTB
434 * @ltb: container object for the LTB
435 * @size: size of the LTB
436 *
437 * Allocate an LTB of the specified size and notify VIOS.
438 *
439 * If the given @ltb already has the correct size, reuse it. Otherwise if
440 * its non-NULL, free it. Then allocate a new one of the correct size.
441 * Notify the VIOS either way since we may now be working with a new VIOS.
442 *
443 * Allocating larger chunks of memory during resets, specially LPM or under
444 * low memory situations can cause resets to fail/timeout and for LPAR to
445 * lose connectivity. So hold onto the LTB even if we fail to communicate
446 * with the VIOS and reuse it on next open. Free LTB when adapter is closed.
447 *
448 * Return: 0 if we were able to allocate the LTB and notify the VIOS and
449 * a negative value otherwise.
450 */
alloc_long_term_buff(struct ibmvnic_adapter * adapter,struct ibmvnic_long_term_buff * ltb,int size)451 static int alloc_long_term_buff(struct ibmvnic_adapter *adapter,
452 struct ibmvnic_long_term_buff *ltb, int size)
453 {
454 struct device *dev = &adapter->vdev->dev;
455 u64 prev = 0;
456 int rc;
457
458 if (!reuse_ltb(ltb, size)) {
459 dev_dbg(dev,
460 "LTB size changed from 0x%llx to 0x%x, reallocating\n",
461 ltb->size, size);
462 prev = ltb->size;
463 free_long_term_buff(adapter, ltb);
464 }
465
466 if (ltb->buff) {
467 dev_dbg(dev, "Reusing LTB [map %d, size 0x%llx]\n",
468 ltb->map_id, ltb->size);
469 } else {
470 ltb->buff = dma_alloc_coherent(dev, size, <b->addr,
471 GFP_KERNEL);
472 if (!ltb->buff) {
473 dev_err(dev, "Couldn't alloc long term buffer\n");
474 return -ENOMEM;
475 }
476 ltb->size = size;
477
478 ltb->map_id = find_first_zero_bit(adapter->map_ids,
479 MAX_MAP_ID);
480 bitmap_set(adapter->map_ids, ltb->map_id, 1);
481
482 dev_dbg(dev,
483 "Allocated new LTB [map %d, size 0x%llx was 0x%llx]\n",
484 ltb->map_id, ltb->size, prev);
485 }
486
487 /* Ensure ltb is zeroed - specially when reusing it. */
488 memset(ltb->buff, 0, ltb->size);
489
490 mutex_lock(&adapter->fw_lock);
491 adapter->fw_done_rc = 0;
492 reinit_completion(&adapter->fw_done);
493
494 rc = send_request_map(adapter, ltb->addr, ltb->size, ltb->map_id);
495 if (rc) {
496 dev_err(dev, "send_request_map failed, rc = %d\n", rc);
497 goto out;
498 }
499
500 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
501 if (rc) {
502 dev_err(dev, "LTB map request aborted or timed out, rc = %d\n",
503 rc);
504 goto out;
505 }
506
507 if (adapter->fw_done_rc) {
508 dev_err(dev, "Couldn't map LTB, rc = %d\n",
509 adapter->fw_done_rc);
510 rc = -EIO;
511 goto out;
512 }
513 rc = 0;
514 out:
515 /* don't free LTB on communication error - see function header */
516 mutex_unlock(&adapter->fw_lock);
517 return rc;
518 }
519
free_long_term_buff(struct ibmvnic_adapter * adapter,struct ibmvnic_long_term_buff * ltb)520 static void free_long_term_buff(struct ibmvnic_adapter *adapter,
521 struct ibmvnic_long_term_buff *ltb)
522 {
523 struct device *dev = &adapter->vdev->dev;
524
525 if (!ltb->buff)
526 return;
527
528 /* VIOS automatically unmaps the long term buffer at remote
529 * end for the following resets:
530 * FAILOVER, MOBILITY, TIMEOUT.
531 */
532 if (adapter->reset_reason != VNIC_RESET_FAILOVER &&
533 adapter->reset_reason != VNIC_RESET_MOBILITY &&
534 adapter->reset_reason != VNIC_RESET_TIMEOUT)
535 send_request_unmap(adapter, ltb->map_id);
536
537 dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
538
539 ltb->buff = NULL;
540 /* mark this map_id free */
541 bitmap_clear(adapter->map_ids, ltb->map_id, 1);
542 ltb->map_id = 0;
543 }
544
545 /**
546 * free_ltb_set - free the given set of long term buffers (LTBS)
547 * @adapter: The ibmvnic adapter containing this ltb set
548 * @ltb_set: The ltb_set to be freed
549 *
550 * Free the set of LTBs in the given set.
551 */
552
free_ltb_set(struct ibmvnic_adapter * adapter,struct ibmvnic_ltb_set * ltb_set)553 static void free_ltb_set(struct ibmvnic_adapter *adapter,
554 struct ibmvnic_ltb_set *ltb_set)
555 {
556 int i;
557
558 for (i = 0; i < ltb_set->num_ltbs; i++)
559 free_long_term_buff(adapter, <b_set->ltbs[i]);
560
561 kfree(ltb_set->ltbs);
562 ltb_set->ltbs = NULL;
563 ltb_set->num_ltbs = 0;
564 }
565
566 /**
567 * alloc_ltb_set() - Allocate a set of long term buffers (LTBs)
568 *
569 * @adapter: ibmvnic adapter associated to the LTB
570 * @ltb_set: container object for the set of LTBs
571 * @num_buffs: Number of buffers in the LTB
572 * @buff_size: Size of each buffer in the LTB
573 *
574 * Allocate a set of LTBs to accommodate @num_buffs buffers of @buff_size
575 * each. We currently cap size each LTB to IBMVNIC_ONE_LTB_SIZE. If the
576 * new set of LTBs have fewer LTBs than the old set, free the excess LTBs.
577 * If new set needs more than in old set, allocate the remaining ones.
578 * Try and reuse as many LTBs as possible and avoid reallocation.
579 *
580 * Any changes to this allocation strategy must be reflected in
581 * map_rxpool_buff_to_ltb() and map_txpool_buff_to_ltb().
582 */
alloc_ltb_set(struct ibmvnic_adapter * adapter,struct ibmvnic_ltb_set * ltb_set,int num_buffs,int buff_size)583 static int alloc_ltb_set(struct ibmvnic_adapter *adapter,
584 struct ibmvnic_ltb_set *ltb_set, int num_buffs,
585 int buff_size)
586 {
587 struct device *dev = &adapter->vdev->dev;
588 struct ibmvnic_ltb_set old_set;
589 struct ibmvnic_ltb_set new_set;
590 int rem_size;
591 int tot_size; /* size of all ltbs */
592 int ltb_size; /* size of one ltb */
593 int nltbs;
594 int rc;
595 int n;
596 int i;
597
598 dev_dbg(dev, "%s() num_buffs %d, buff_size %d\n", __func__, num_buffs,
599 buff_size);
600
601 ltb_size = rounddown(IBMVNIC_ONE_LTB_SIZE, buff_size);
602 tot_size = num_buffs * buff_size;
603
604 if (ltb_size > tot_size)
605 ltb_size = tot_size;
606
607 nltbs = tot_size / ltb_size;
608 if (tot_size % ltb_size)
609 nltbs++;
610
611 old_set = *ltb_set;
612
613 if (old_set.num_ltbs == nltbs) {
614 new_set = old_set;
615 } else {
616 int tmp = nltbs * sizeof(struct ibmvnic_long_term_buff);
617
618 new_set.ltbs = kzalloc(tmp, GFP_KERNEL);
619 if (!new_set.ltbs)
620 return -ENOMEM;
621
622 new_set.num_ltbs = nltbs;
623
624 /* Free any excess ltbs in old set */
625 for (i = new_set.num_ltbs; i < old_set.num_ltbs; i++)
626 free_long_term_buff(adapter, &old_set.ltbs[i]);
627
628 /* Copy remaining ltbs to new set. All LTBs except the
629 * last one are of the same size. alloc_long_term_buff()
630 * will realloc if the size changes.
631 */
632 n = min(old_set.num_ltbs, new_set.num_ltbs);
633 for (i = 0; i < n; i++)
634 new_set.ltbs[i] = old_set.ltbs[i];
635
636 /* Any additional ltbs in new set will have NULL ltbs for
637 * now and will be allocated in alloc_long_term_buff().
638 */
639
640 /* We no longer need the old_set so free it. Note that we
641 * may have reused some ltbs from old set and freed excess
642 * ltbs above. So we only need to free the container now
643 * not the LTBs themselves. (i.e. dont free_ltb_set()!)
644 */
645 kfree(old_set.ltbs);
646 old_set.ltbs = NULL;
647 old_set.num_ltbs = 0;
648
649 /* Install the new set. If allocations fail below, we will
650 * retry later and know what size LTBs we need.
651 */
652 *ltb_set = new_set;
653 }
654
655 i = 0;
656 rem_size = tot_size;
657 while (rem_size) {
658 if (ltb_size > rem_size)
659 ltb_size = rem_size;
660
661 rem_size -= ltb_size;
662
663 rc = alloc_long_term_buff(adapter, &new_set.ltbs[i], ltb_size);
664 if (rc)
665 goto out;
666 i++;
667 }
668
669 WARN_ON(i != new_set.num_ltbs);
670
671 return 0;
672 out:
673 /* We may have allocated one/more LTBs before failing and we
674 * want to try and reuse on next reset. So don't free ltb set.
675 */
676 return rc;
677 }
678
679 /**
680 * map_rxpool_buf_to_ltb - Map given rxpool buffer to offset in an LTB.
681 * @rxpool: The receive buffer pool containing buffer
682 * @bufidx: Index of buffer in rxpool
683 * @ltbp: (Output) pointer to the long term buffer containing the buffer
684 * @offset: (Output) offset of buffer in the LTB from @ltbp
685 *
686 * Map the given buffer identified by [rxpool, bufidx] to an LTB in the
687 * pool and its corresponding offset. Assume for now that each LTB is of
688 * different size but could possibly be optimized based on the allocation
689 * strategy in alloc_ltb_set().
690 */
map_rxpool_buf_to_ltb(struct ibmvnic_rx_pool * rxpool,unsigned int bufidx,struct ibmvnic_long_term_buff ** ltbp,unsigned int * offset)691 static void map_rxpool_buf_to_ltb(struct ibmvnic_rx_pool *rxpool,
692 unsigned int bufidx,
693 struct ibmvnic_long_term_buff **ltbp,
694 unsigned int *offset)
695 {
696 struct ibmvnic_long_term_buff *ltb;
697 int nbufs; /* # of buffers in one ltb */
698 int i;
699
700 WARN_ON(bufidx >= rxpool->size);
701
702 for (i = 0; i < rxpool->ltb_set.num_ltbs; i++) {
703 ltb = &rxpool->ltb_set.ltbs[i];
704 nbufs = ltb->size / rxpool->buff_size;
705 if (bufidx < nbufs)
706 break;
707 bufidx -= nbufs;
708 }
709
710 *ltbp = ltb;
711 *offset = bufidx * rxpool->buff_size;
712 }
713
714 /**
715 * map_txpool_buf_to_ltb - Map given txpool buffer to offset in an LTB.
716 * @txpool: The transmit buffer pool containing buffer
717 * @bufidx: Index of buffer in txpool
718 * @ltbp: (Output) pointer to the long term buffer (LTB) containing the buffer
719 * @offset: (Output) offset of buffer in the LTB from @ltbp
720 *
721 * Map the given buffer identified by [txpool, bufidx] to an LTB in the
722 * pool and its corresponding offset.
723 */
map_txpool_buf_to_ltb(struct ibmvnic_tx_pool * txpool,unsigned int bufidx,struct ibmvnic_long_term_buff ** ltbp,unsigned int * offset)724 static void map_txpool_buf_to_ltb(struct ibmvnic_tx_pool *txpool,
725 unsigned int bufidx,
726 struct ibmvnic_long_term_buff **ltbp,
727 unsigned int *offset)
728 {
729 struct ibmvnic_long_term_buff *ltb;
730 int nbufs; /* # of buffers in one ltb */
731 int i;
732
733 WARN_ON_ONCE(bufidx >= txpool->num_buffers);
734
735 for (i = 0; i < txpool->ltb_set.num_ltbs; i++) {
736 ltb = &txpool->ltb_set.ltbs[i];
737 nbufs = ltb->size / txpool->buf_size;
738 if (bufidx < nbufs)
739 break;
740 bufidx -= nbufs;
741 }
742
743 *ltbp = ltb;
744 *offset = bufidx * txpool->buf_size;
745 }
746
deactivate_rx_pools(struct ibmvnic_adapter * adapter)747 static void deactivate_rx_pools(struct ibmvnic_adapter *adapter)
748 {
749 int i;
750
751 for (i = 0; i < adapter->num_active_rx_pools; i++)
752 adapter->rx_pool[i].active = 0;
753 }
754
replenish_rx_pool(struct ibmvnic_adapter * adapter,struct ibmvnic_rx_pool * pool)755 static void replenish_rx_pool(struct ibmvnic_adapter *adapter,
756 struct ibmvnic_rx_pool *pool)
757 {
758 int count = pool->size - atomic_read(&pool->available);
759 u64 handle = adapter->rx_scrq[pool->index]->handle;
760 struct device *dev = &adapter->vdev->dev;
761 struct ibmvnic_ind_xmit_queue *ind_bufp;
762 struct ibmvnic_sub_crq_queue *rx_scrq;
763 struct ibmvnic_long_term_buff *ltb;
764 union sub_crq *sub_crq;
765 int buffers_added = 0;
766 unsigned long lpar_rc;
767 struct sk_buff *skb;
768 unsigned int offset;
769 dma_addr_t dma_addr;
770 unsigned char *dst;
771 int shift = 0;
772 int bufidx;
773 int i;
774
775 if (!pool->active)
776 return;
777
778 rx_scrq = adapter->rx_scrq[pool->index];
779 ind_bufp = &rx_scrq->ind_buf;
780
781 /* netdev_skb_alloc() could have failed after we saved a few skbs
782 * in the indir_buf and we would not have sent them to VIOS yet.
783 * To account for them, start the loop at ind_bufp->index rather
784 * than 0. If we pushed all the skbs to VIOS, ind_bufp->index will
785 * be 0.
786 */
787 for (i = ind_bufp->index; i < count; ++i) {
788 bufidx = pool->free_map[pool->next_free];
789
790 /* We maybe reusing the skb from earlier resets. Allocate
791 * only if necessary. But since the LTB may have changed
792 * during reset (see init_rx_pools()), update LTB below
793 * even if reusing skb.
794 */
795 skb = pool->rx_buff[bufidx].skb;
796 if (!skb) {
797 skb = netdev_alloc_skb(adapter->netdev,
798 pool->buff_size);
799 if (!skb) {
800 dev_err(dev, "Couldn't replenish rx buff\n");
801 adapter->replenish_no_mem++;
802 break;
803 }
804 }
805
806 pool->free_map[pool->next_free] = IBMVNIC_INVALID_MAP;
807 pool->next_free = (pool->next_free + 1) % pool->size;
808
809 /* Copy the skb to the long term mapped DMA buffer */
810 map_rxpool_buf_to_ltb(pool, bufidx, <b, &offset);
811 dst = ltb->buff + offset;
812 memset(dst, 0, pool->buff_size);
813 dma_addr = ltb->addr + offset;
814
815 /* add the skb to an rx_buff in the pool */
816 pool->rx_buff[bufidx].data = dst;
817 pool->rx_buff[bufidx].dma = dma_addr;
818 pool->rx_buff[bufidx].skb = skb;
819 pool->rx_buff[bufidx].pool_index = pool->index;
820 pool->rx_buff[bufidx].size = pool->buff_size;
821
822 /* queue the rx_buff for the next send_subcrq_indirect */
823 sub_crq = &ind_bufp->indir_arr[ind_bufp->index++];
824 memset(sub_crq, 0, sizeof(*sub_crq));
825 sub_crq->rx_add.first = IBMVNIC_CRQ_CMD;
826 sub_crq->rx_add.correlator =
827 cpu_to_be64((u64)&pool->rx_buff[bufidx]);
828 sub_crq->rx_add.ioba = cpu_to_be32(dma_addr);
829 sub_crq->rx_add.map_id = ltb->map_id;
830
831 /* The length field of the sCRQ is defined to be 24 bits so the
832 * buffer size needs to be left shifted by a byte before it is
833 * converted to big endian to prevent the last byte from being
834 * truncated.
835 */
836 #ifdef __LITTLE_ENDIAN__
837 shift = 8;
838 #endif
839 sub_crq->rx_add.len = cpu_to_be32(pool->buff_size << shift);
840
841 /* if send_subcrq_indirect queue is full, flush to VIOS */
842 if (ind_bufp->index == IBMVNIC_MAX_IND_DESCS ||
843 i == count - 1) {
844 lpar_rc =
845 send_subcrq_indirect(adapter, handle,
846 (u64)ind_bufp->indir_dma,
847 (u64)ind_bufp->index);
848 if (lpar_rc != H_SUCCESS)
849 goto failure;
850 buffers_added += ind_bufp->index;
851 adapter->replenish_add_buff_success += ind_bufp->index;
852 ind_bufp->index = 0;
853 }
854 }
855 atomic_add(buffers_added, &pool->available);
856 return;
857
858 failure:
859 if (lpar_rc != H_PARAMETER && lpar_rc != H_CLOSED)
860 dev_err_ratelimited(dev, "rx: replenish packet buffer failed\n");
861 for (i = ind_bufp->index - 1; i >= 0; --i) {
862 struct ibmvnic_rx_buff *rx_buff;
863
864 pool->next_free = pool->next_free == 0 ?
865 pool->size - 1 : pool->next_free - 1;
866 sub_crq = &ind_bufp->indir_arr[i];
867 rx_buff = (struct ibmvnic_rx_buff *)
868 be64_to_cpu(sub_crq->rx_add.correlator);
869 bufidx = (int)(rx_buff - pool->rx_buff);
870 pool->free_map[pool->next_free] = bufidx;
871 dev_kfree_skb_any(pool->rx_buff[bufidx].skb);
872 pool->rx_buff[bufidx].skb = NULL;
873 }
874 adapter->replenish_add_buff_failure += ind_bufp->index;
875 atomic_add(buffers_added, &pool->available);
876 ind_bufp->index = 0;
877 if (lpar_rc == H_CLOSED || adapter->failover_pending) {
878 /* Disable buffer pool replenishment and report carrier off if
879 * queue is closed or pending failover.
880 * Firmware guarantees that a signal will be sent to the
881 * driver, triggering a reset.
882 */
883 deactivate_rx_pools(adapter);
884 netif_carrier_off(adapter->netdev);
885 }
886 }
887
replenish_pools(struct ibmvnic_adapter * adapter)888 static void replenish_pools(struct ibmvnic_adapter *adapter)
889 {
890 int i;
891
892 adapter->replenish_task_cycles++;
893 for (i = 0; i < adapter->num_active_rx_pools; i++) {
894 if (adapter->rx_pool[i].active)
895 replenish_rx_pool(adapter, &adapter->rx_pool[i]);
896 }
897
898 netdev_dbg(adapter->netdev, "Replenished %d pools\n", i);
899 }
900
release_stats_buffers(struct ibmvnic_adapter * adapter)901 static void release_stats_buffers(struct ibmvnic_adapter *adapter)
902 {
903 kfree(adapter->tx_stats_buffers);
904 kfree(adapter->rx_stats_buffers);
905 adapter->tx_stats_buffers = NULL;
906 adapter->rx_stats_buffers = NULL;
907 }
908
init_stats_buffers(struct ibmvnic_adapter * adapter)909 static int init_stats_buffers(struct ibmvnic_adapter *adapter)
910 {
911 adapter->tx_stats_buffers =
912 kcalloc(IBMVNIC_MAX_QUEUES,
913 sizeof(struct ibmvnic_tx_queue_stats),
914 GFP_KERNEL);
915 if (!adapter->tx_stats_buffers)
916 return -ENOMEM;
917
918 adapter->rx_stats_buffers =
919 kcalloc(IBMVNIC_MAX_QUEUES,
920 sizeof(struct ibmvnic_rx_queue_stats),
921 GFP_KERNEL);
922 if (!adapter->rx_stats_buffers)
923 return -ENOMEM;
924
925 return 0;
926 }
927
release_stats_token(struct ibmvnic_adapter * adapter)928 static void release_stats_token(struct ibmvnic_adapter *adapter)
929 {
930 struct device *dev = &adapter->vdev->dev;
931
932 if (!adapter->stats_token)
933 return;
934
935 dma_unmap_single(dev, adapter->stats_token,
936 sizeof(struct ibmvnic_statistics),
937 DMA_FROM_DEVICE);
938 adapter->stats_token = 0;
939 }
940
init_stats_token(struct ibmvnic_adapter * adapter)941 static int init_stats_token(struct ibmvnic_adapter *adapter)
942 {
943 struct device *dev = &adapter->vdev->dev;
944 dma_addr_t stok;
945 int rc;
946
947 stok = dma_map_single(dev, &adapter->stats,
948 sizeof(struct ibmvnic_statistics),
949 DMA_FROM_DEVICE);
950 rc = dma_mapping_error(dev, stok);
951 if (rc) {
952 dev_err(dev, "Couldn't map stats buffer, rc = %d\n", rc);
953 return rc;
954 }
955
956 adapter->stats_token = stok;
957 netdev_dbg(adapter->netdev, "Stats token initialized (%llx)\n", stok);
958 return 0;
959 }
960
961 /**
962 * release_rx_pools() - Release any rx pools attached to @adapter.
963 * @adapter: ibmvnic adapter
964 *
965 * Safe to call this multiple times - even if no pools are attached.
966 */
release_rx_pools(struct ibmvnic_adapter * adapter)967 static void release_rx_pools(struct ibmvnic_adapter *adapter)
968 {
969 struct ibmvnic_rx_pool *rx_pool;
970 int i, j;
971
972 if (!adapter->rx_pool)
973 return;
974
975 for (i = 0; i < adapter->num_active_rx_pools; i++) {
976 rx_pool = &adapter->rx_pool[i];
977
978 netdev_dbg(adapter->netdev, "Releasing rx_pool[%d]\n", i);
979
980 kfree(rx_pool->free_map);
981
982 free_ltb_set(adapter, &rx_pool->ltb_set);
983
984 if (!rx_pool->rx_buff)
985 continue;
986
987 for (j = 0; j < rx_pool->size; j++) {
988 if (rx_pool->rx_buff[j].skb) {
989 dev_kfree_skb_any(rx_pool->rx_buff[j].skb);
990 rx_pool->rx_buff[j].skb = NULL;
991 }
992 }
993
994 kfree(rx_pool->rx_buff);
995 }
996
997 kfree(adapter->rx_pool);
998 adapter->rx_pool = NULL;
999 adapter->num_active_rx_pools = 0;
1000 adapter->prev_rx_pool_size = 0;
1001 }
1002
1003 /**
1004 * reuse_rx_pools() - Check if the existing rx pools can be reused.
1005 * @adapter: ibmvnic adapter
1006 *
1007 * Check if the existing rx pools in the adapter can be reused. The
1008 * pools can be reused if the pool parameters (number of pools,
1009 * number of buffers in the pool and size of each buffer) have not
1010 * changed.
1011 *
1012 * NOTE: This assumes that all pools have the same number of buffers
1013 * which is the case currently. If that changes, we must fix this.
1014 *
1015 * Return: true if the rx pools can be reused, false otherwise.
1016 */
reuse_rx_pools(struct ibmvnic_adapter * adapter)1017 static bool reuse_rx_pools(struct ibmvnic_adapter *adapter)
1018 {
1019 u64 old_num_pools, new_num_pools;
1020 u64 old_pool_size, new_pool_size;
1021 u64 old_buff_size, new_buff_size;
1022
1023 if (!adapter->rx_pool)
1024 return false;
1025
1026 old_num_pools = adapter->num_active_rx_pools;
1027 new_num_pools = adapter->req_rx_queues;
1028
1029 old_pool_size = adapter->prev_rx_pool_size;
1030 new_pool_size = adapter->req_rx_add_entries_per_subcrq;
1031
1032 old_buff_size = adapter->prev_rx_buf_sz;
1033 new_buff_size = adapter->cur_rx_buf_sz;
1034
1035 if (old_buff_size != new_buff_size ||
1036 old_num_pools != new_num_pools ||
1037 old_pool_size != new_pool_size)
1038 return false;
1039
1040 return true;
1041 }
1042
1043 /**
1044 * init_rx_pools(): Initialize the set of receiver pools in the adapter.
1045 * @netdev: net device associated with the vnic interface
1046 *
1047 * Initialize the set of receiver pools in the ibmvnic adapter associated
1048 * with the net_device @netdev. If possible, reuse the existing rx pools.
1049 * Otherwise free any existing pools and allocate a new set of pools
1050 * before initializing them.
1051 *
1052 * Return: 0 on success and negative value on error.
1053 */
init_rx_pools(struct net_device * netdev)1054 static int init_rx_pools(struct net_device *netdev)
1055 {
1056 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1057 struct device *dev = &adapter->vdev->dev;
1058 struct ibmvnic_rx_pool *rx_pool;
1059 u64 num_pools;
1060 u64 pool_size; /* # of buffers in one pool */
1061 u64 buff_size;
1062 int i, j, rc;
1063
1064 pool_size = adapter->req_rx_add_entries_per_subcrq;
1065 num_pools = adapter->req_rx_queues;
1066 buff_size = adapter->cur_rx_buf_sz;
1067
1068 if (reuse_rx_pools(adapter)) {
1069 dev_dbg(dev, "Reusing rx pools\n");
1070 goto update_ltb;
1071 }
1072
1073 /* Allocate/populate the pools. */
1074 release_rx_pools(adapter);
1075
1076 adapter->rx_pool = kcalloc(num_pools,
1077 sizeof(struct ibmvnic_rx_pool),
1078 GFP_KERNEL);
1079 if (!adapter->rx_pool) {
1080 dev_err(dev, "Failed to allocate rx pools\n");
1081 return -ENOMEM;
1082 }
1083
1084 /* Set num_active_rx_pools early. If we fail below after partial
1085 * allocation, release_rx_pools() will know how many to look for.
1086 */
1087 adapter->num_active_rx_pools = num_pools;
1088
1089 for (i = 0; i < num_pools; i++) {
1090 rx_pool = &adapter->rx_pool[i];
1091
1092 netdev_dbg(adapter->netdev,
1093 "Initializing rx_pool[%d], %lld buffs, %lld bytes each\n",
1094 i, pool_size, buff_size);
1095
1096 rx_pool->size = pool_size;
1097 rx_pool->index = i;
1098 rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
1099
1100 rx_pool->free_map = kcalloc(rx_pool->size, sizeof(int),
1101 GFP_KERNEL);
1102 if (!rx_pool->free_map) {
1103 dev_err(dev, "Couldn't alloc free_map %d\n", i);
1104 rc = -ENOMEM;
1105 goto out_release;
1106 }
1107
1108 rx_pool->rx_buff = kcalloc(rx_pool->size,
1109 sizeof(struct ibmvnic_rx_buff),
1110 GFP_KERNEL);
1111 if (!rx_pool->rx_buff) {
1112 dev_err(dev, "Couldn't alloc rx buffers\n");
1113 rc = -ENOMEM;
1114 goto out_release;
1115 }
1116 }
1117
1118 adapter->prev_rx_pool_size = pool_size;
1119 adapter->prev_rx_buf_sz = adapter->cur_rx_buf_sz;
1120
1121 update_ltb:
1122 for (i = 0; i < num_pools; i++) {
1123 rx_pool = &adapter->rx_pool[i];
1124 dev_dbg(dev, "Updating LTB for rx pool %d [%d, %d]\n",
1125 i, rx_pool->size, rx_pool->buff_size);
1126
1127 rc = alloc_ltb_set(adapter, &rx_pool->ltb_set,
1128 rx_pool->size, rx_pool->buff_size);
1129 if (rc)
1130 goto out;
1131
1132 for (j = 0; j < rx_pool->size; ++j) {
1133 struct ibmvnic_rx_buff *rx_buff;
1134
1135 rx_pool->free_map[j] = j;
1136
1137 /* NOTE: Don't clear rx_buff->skb here - will leak
1138 * memory! replenish_rx_pool() will reuse skbs or
1139 * allocate as necessary.
1140 */
1141 rx_buff = &rx_pool->rx_buff[j];
1142 rx_buff->dma = 0;
1143 rx_buff->data = 0;
1144 rx_buff->size = 0;
1145 rx_buff->pool_index = 0;
1146 }
1147
1148 /* Mark pool "empty" so replenish_rx_pools() will
1149 * update the LTB info for each buffer
1150 */
1151 atomic_set(&rx_pool->available, 0);
1152 rx_pool->next_alloc = 0;
1153 rx_pool->next_free = 0;
1154 /* replenish_rx_pool() may have called deactivate_rx_pools()
1155 * on failover. Ensure pool is active now.
1156 */
1157 rx_pool->active = 1;
1158 }
1159 return 0;
1160 out_release:
1161 release_rx_pools(adapter);
1162 out:
1163 /* We failed to allocate one or more LTBs or map them on the VIOS.
1164 * Hold onto the pools and any LTBs that we did allocate/map.
1165 */
1166 return rc;
1167 }
1168
release_vpd_data(struct ibmvnic_adapter * adapter)1169 static void release_vpd_data(struct ibmvnic_adapter *adapter)
1170 {
1171 if (!adapter->vpd)
1172 return;
1173
1174 kfree(adapter->vpd->buff);
1175 kfree(adapter->vpd);
1176
1177 adapter->vpd = NULL;
1178 }
1179
release_one_tx_pool(struct ibmvnic_adapter * adapter,struct ibmvnic_tx_pool * tx_pool)1180 static void release_one_tx_pool(struct ibmvnic_adapter *adapter,
1181 struct ibmvnic_tx_pool *tx_pool)
1182 {
1183 kfree(tx_pool->tx_buff);
1184 kfree(tx_pool->free_map);
1185 free_ltb_set(adapter, &tx_pool->ltb_set);
1186 }
1187
1188 /**
1189 * release_tx_pools() - Release any tx pools attached to @adapter.
1190 * @adapter: ibmvnic adapter
1191 *
1192 * Safe to call this multiple times - even if no pools are attached.
1193 */
release_tx_pools(struct ibmvnic_adapter * adapter)1194 static void release_tx_pools(struct ibmvnic_adapter *adapter)
1195 {
1196 int i;
1197
1198 /* init_tx_pools() ensures that ->tx_pool and ->tso_pool are
1199 * both NULL or both non-NULL. So we only need to check one.
1200 */
1201 if (!adapter->tx_pool)
1202 return;
1203
1204 for (i = 0; i < adapter->num_active_tx_pools; i++) {
1205 release_one_tx_pool(adapter, &adapter->tx_pool[i]);
1206 release_one_tx_pool(adapter, &adapter->tso_pool[i]);
1207 }
1208
1209 kfree(adapter->tx_pool);
1210 adapter->tx_pool = NULL;
1211 kfree(adapter->tso_pool);
1212 adapter->tso_pool = NULL;
1213 adapter->num_active_tx_pools = 0;
1214 adapter->prev_tx_pool_size = 0;
1215 }
1216
init_one_tx_pool(struct net_device * netdev,struct ibmvnic_tx_pool * tx_pool,int pool_size,int buf_size)1217 static int init_one_tx_pool(struct net_device *netdev,
1218 struct ibmvnic_tx_pool *tx_pool,
1219 int pool_size, int buf_size)
1220 {
1221 int i;
1222
1223 tx_pool->tx_buff = kcalloc(pool_size,
1224 sizeof(struct ibmvnic_tx_buff),
1225 GFP_KERNEL);
1226 if (!tx_pool->tx_buff)
1227 return -ENOMEM;
1228
1229 tx_pool->free_map = kcalloc(pool_size, sizeof(int), GFP_KERNEL);
1230 if (!tx_pool->free_map) {
1231 kfree(tx_pool->tx_buff);
1232 tx_pool->tx_buff = NULL;
1233 return -ENOMEM;
1234 }
1235
1236 for (i = 0; i < pool_size; i++)
1237 tx_pool->free_map[i] = i;
1238
1239 tx_pool->consumer_index = 0;
1240 tx_pool->producer_index = 0;
1241 tx_pool->num_buffers = pool_size;
1242 tx_pool->buf_size = buf_size;
1243
1244 return 0;
1245 }
1246
1247 /**
1248 * reuse_tx_pools() - Check if the existing tx pools can be reused.
1249 * @adapter: ibmvnic adapter
1250 *
1251 * Check if the existing tx pools in the adapter can be reused. The
1252 * pools can be reused if the pool parameters (number of pools,
1253 * number of buffers in the pool and mtu) have not changed.
1254 *
1255 * NOTE: This assumes that all pools have the same number of buffers
1256 * which is the case currently. If that changes, we must fix this.
1257 *
1258 * Return: true if the tx pools can be reused, false otherwise.
1259 */
reuse_tx_pools(struct ibmvnic_adapter * adapter)1260 static bool reuse_tx_pools(struct ibmvnic_adapter *adapter)
1261 {
1262 u64 old_num_pools, new_num_pools;
1263 u64 old_pool_size, new_pool_size;
1264 u64 old_mtu, new_mtu;
1265
1266 if (!adapter->tx_pool)
1267 return false;
1268
1269 old_num_pools = adapter->num_active_tx_pools;
1270 new_num_pools = adapter->num_active_tx_scrqs;
1271 old_pool_size = adapter->prev_tx_pool_size;
1272 new_pool_size = adapter->req_tx_entries_per_subcrq;
1273 old_mtu = adapter->prev_mtu;
1274 new_mtu = adapter->req_mtu;
1275
1276 if (old_mtu != new_mtu ||
1277 old_num_pools != new_num_pools ||
1278 old_pool_size != new_pool_size)
1279 return false;
1280
1281 return true;
1282 }
1283
1284 /**
1285 * init_tx_pools(): Initialize the set of transmit pools in the adapter.
1286 * @netdev: net device associated with the vnic interface
1287 *
1288 * Initialize the set of transmit pools in the ibmvnic adapter associated
1289 * with the net_device @netdev. If possible, reuse the existing tx pools.
1290 * Otherwise free any existing pools and allocate a new set of pools
1291 * before initializing them.
1292 *
1293 * Return: 0 on success and negative value on error.
1294 */
init_tx_pools(struct net_device * netdev)1295 static int init_tx_pools(struct net_device *netdev)
1296 {
1297 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1298 struct device *dev = &adapter->vdev->dev;
1299 int num_pools;
1300 u64 pool_size; /* # of buffers in pool */
1301 u64 buff_size;
1302 int i, j, rc;
1303
1304 num_pools = adapter->req_tx_queues;
1305
1306 /* We must notify the VIOS about the LTB on all resets - but we only
1307 * need to alloc/populate pools if either the number of buffers or
1308 * size of each buffer in the pool has changed.
1309 */
1310 if (reuse_tx_pools(adapter)) {
1311 netdev_dbg(netdev, "Reusing tx pools\n");
1312 goto update_ltb;
1313 }
1314
1315 /* Allocate/populate the pools. */
1316 release_tx_pools(adapter);
1317
1318 pool_size = adapter->req_tx_entries_per_subcrq;
1319 num_pools = adapter->num_active_tx_scrqs;
1320
1321 adapter->tx_pool = kcalloc(num_pools,
1322 sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
1323 if (!adapter->tx_pool)
1324 return -ENOMEM;
1325
1326 adapter->tso_pool = kcalloc(num_pools,
1327 sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
1328 /* To simplify release_tx_pools() ensure that ->tx_pool and
1329 * ->tso_pool are either both NULL or both non-NULL.
1330 */
1331 if (!adapter->tso_pool) {
1332 kfree(adapter->tx_pool);
1333 adapter->tx_pool = NULL;
1334 return -ENOMEM;
1335 }
1336
1337 /* Set num_active_tx_pools early. If we fail below after partial
1338 * allocation, release_tx_pools() will know how many to look for.
1339 */
1340 adapter->num_active_tx_pools = num_pools;
1341
1342 buff_size = adapter->req_mtu + VLAN_HLEN;
1343 buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
1344
1345 for (i = 0; i < num_pools; i++) {
1346 dev_dbg(dev, "Init tx pool %d [%llu, %llu]\n",
1347 i, adapter->req_tx_entries_per_subcrq, buff_size);
1348
1349 rc = init_one_tx_pool(netdev, &adapter->tx_pool[i],
1350 pool_size, buff_size);
1351 if (rc)
1352 goto out_release;
1353
1354 rc = init_one_tx_pool(netdev, &adapter->tso_pool[i],
1355 IBMVNIC_TSO_BUFS,
1356 IBMVNIC_TSO_BUF_SZ);
1357 if (rc)
1358 goto out_release;
1359 }
1360
1361 adapter->prev_tx_pool_size = pool_size;
1362 adapter->prev_mtu = adapter->req_mtu;
1363
1364 update_ltb:
1365 /* NOTE: All tx_pools have the same number of buffers (which is
1366 * same as pool_size). All tso_pools have IBMVNIC_TSO_BUFS
1367 * buffers (see calls init_one_tx_pool() for these).
1368 * For consistency, we use tx_pool->num_buffers and
1369 * tso_pool->num_buffers below.
1370 */
1371 rc = -1;
1372 for (i = 0; i < num_pools; i++) {
1373 struct ibmvnic_tx_pool *tso_pool;
1374 struct ibmvnic_tx_pool *tx_pool;
1375
1376 tx_pool = &adapter->tx_pool[i];
1377
1378 dev_dbg(dev, "Updating LTB for tx pool %d [%d, %d]\n",
1379 i, tx_pool->num_buffers, tx_pool->buf_size);
1380
1381 rc = alloc_ltb_set(adapter, &tx_pool->ltb_set,
1382 tx_pool->num_buffers, tx_pool->buf_size);
1383 if (rc)
1384 goto out;
1385
1386 tx_pool->consumer_index = 0;
1387 tx_pool->producer_index = 0;
1388
1389 for (j = 0; j < tx_pool->num_buffers; j++)
1390 tx_pool->free_map[j] = j;
1391
1392 tso_pool = &adapter->tso_pool[i];
1393
1394 dev_dbg(dev, "Updating LTB for tso pool %d [%d, %d]\n",
1395 i, tso_pool->num_buffers, tso_pool->buf_size);
1396
1397 rc = alloc_ltb_set(adapter, &tso_pool->ltb_set,
1398 tso_pool->num_buffers, tso_pool->buf_size);
1399 if (rc)
1400 goto out;
1401
1402 tso_pool->consumer_index = 0;
1403 tso_pool->producer_index = 0;
1404
1405 for (j = 0; j < tso_pool->num_buffers; j++)
1406 tso_pool->free_map[j] = j;
1407 }
1408
1409 return 0;
1410 out_release:
1411 release_tx_pools(adapter);
1412 out:
1413 /* We failed to allocate one or more LTBs or map them on the VIOS.
1414 * Hold onto the pools and any LTBs that we did allocate/map.
1415 */
1416 return rc;
1417 }
1418
ibmvnic_napi_enable(struct ibmvnic_adapter * adapter)1419 static void ibmvnic_napi_enable(struct ibmvnic_adapter *adapter)
1420 {
1421 int i;
1422
1423 if (adapter->napi_enabled)
1424 return;
1425
1426 for (i = 0; i < adapter->req_rx_queues; i++)
1427 napi_enable(&adapter->napi[i]);
1428
1429 adapter->napi_enabled = true;
1430 }
1431
ibmvnic_napi_disable(struct ibmvnic_adapter * adapter)1432 static void ibmvnic_napi_disable(struct ibmvnic_adapter *adapter)
1433 {
1434 int i;
1435
1436 if (!adapter->napi_enabled)
1437 return;
1438
1439 for (i = 0; i < adapter->req_rx_queues; i++) {
1440 netdev_dbg(adapter->netdev, "Disabling napi[%d]\n", i);
1441 napi_disable(&adapter->napi[i]);
1442 }
1443
1444 adapter->napi_enabled = false;
1445 }
1446
init_napi(struct ibmvnic_adapter * adapter)1447 static int init_napi(struct ibmvnic_adapter *adapter)
1448 {
1449 int i;
1450
1451 adapter->napi = kcalloc(adapter->req_rx_queues,
1452 sizeof(struct napi_struct), GFP_KERNEL);
1453 if (!adapter->napi)
1454 return -ENOMEM;
1455
1456 for (i = 0; i < adapter->req_rx_queues; i++) {
1457 netdev_dbg(adapter->netdev, "Adding napi[%d]\n", i);
1458 netif_napi_add(adapter->netdev, &adapter->napi[i],
1459 ibmvnic_poll);
1460 }
1461
1462 adapter->num_active_rx_napi = adapter->req_rx_queues;
1463 return 0;
1464 }
1465
release_napi(struct ibmvnic_adapter * adapter)1466 static void release_napi(struct ibmvnic_adapter *adapter)
1467 {
1468 int i;
1469
1470 if (!adapter->napi)
1471 return;
1472
1473 for (i = 0; i < adapter->num_active_rx_napi; i++) {
1474 netdev_dbg(adapter->netdev, "Releasing napi[%d]\n", i);
1475 netif_napi_del(&adapter->napi[i]);
1476 }
1477
1478 kfree(adapter->napi);
1479 adapter->napi = NULL;
1480 adapter->num_active_rx_napi = 0;
1481 adapter->napi_enabled = false;
1482 }
1483
adapter_state_to_string(enum vnic_state state)1484 static const char *adapter_state_to_string(enum vnic_state state)
1485 {
1486 switch (state) {
1487 case VNIC_PROBING:
1488 return "PROBING";
1489 case VNIC_PROBED:
1490 return "PROBED";
1491 case VNIC_OPENING:
1492 return "OPENING";
1493 case VNIC_OPEN:
1494 return "OPEN";
1495 case VNIC_CLOSING:
1496 return "CLOSING";
1497 case VNIC_CLOSED:
1498 return "CLOSED";
1499 case VNIC_REMOVING:
1500 return "REMOVING";
1501 case VNIC_REMOVED:
1502 return "REMOVED";
1503 case VNIC_DOWN:
1504 return "DOWN";
1505 }
1506 return "UNKNOWN";
1507 }
1508
ibmvnic_login(struct net_device * netdev)1509 static int ibmvnic_login(struct net_device *netdev)
1510 {
1511 unsigned long flags, timeout = msecs_to_jiffies(20000);
1512 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1513 int retry_count = 0;
1514 int retries = 10;
1515 bool retry;
1516 int rc;
1517
1518 do {
1519 retry = false;
1520 if (retry_count > retries) {
1521 netdev_warn(netdev, "Login attempts exceeded\n");
1522 return -EACCES;
1523 }
1524
1525 adapter->init_done_rc = 0;
1526 reinit_completion(&adapter->init_done);
1527 rc = send_login(adapter);
1528 if (rc)
1529 return rc;
1530
1531 if (!wait_for_completion_timeout(&adapter->init_done,
1532 timeout)) {
1533 netdev_warn(netdev, "Login timed out\n");
1534 adapter->login_pending = false;
1535 goto partial_reset;
1536 }
1537
1538 if (adapter->init_done_rc == ABORTED) {
1539 netdev_warn(netdev, "Login aborted, retrying...\n");
1540 retry = true;
1541 adapter->init_done_rc = 0;
1542 retry_count++;
1543 /* FW or device may be busy, so
1544 * wait a bit before retrying login
1545 */
1546 msleep(500);
1547 } else if (adapter->init_done_rc == PARTIALSUCCESS) {
1548 retry_count++;
1549 release_sub_crqs(adapter, 1);
1550
1551 retry = true;
1552 netdev_dbg(netdev,
1553 "Received partial success, retrying...\n");
1554 adapter->init_done_rc = 0;
1555 reinit_completion(&adapter->init_done);
1556 send_query_cap(adapter);
1557 if (!wait_for_completion_timeout(&adapter->init_done,
1558 timeout)) {
1559 netdev_warn(netdev,
1560 "Capabilities query timed out\n");
1561 return -ETIMEDOUT;
1562 }
1563
1564 rc = init_sub_crqs(adapter);
1565 if (rc) {
1566 netdev_warn(netdev,
1567 "SCRQ initialization failed\n");
1568 return rc;
1569 }
1570
1571 rc = init_sub_crq_irqs(adapter);
1572 if (rc) {
1573 netdev_warn(netdev,
1574 "SCRQ irq initialization failed\n");
1575 return rc;
1576 }
1577 /* Default/timeout error handling, reset and start fresh */
1578 } else if (adapter->init_done_rc) {
1579 netdev_warn(netdev, "Adapter login failed, init_done_rc = %d\n",
1580 adapter->init_done_rc);
1581
1582 partial_reset:
1583 /* adapter login failed, so free any CRQs or sub-CRQs
1584 * and register again before attempting to login again.
1585 * If we don't do this then the VIOS may think that
1586 * we are already logged in and reject any subsequent
1587 * attempts
1588 */
1589 netdev_warn(netdev,
1590 "Freeing and re-registering CRQs before attempting to login again\n");
1591 retry = true;
1592 adapter->init_done_rc = 0;
1593 release_sub_crqs(adapter, true);
1594 /* Much of this is similar logic as ibmvnic_probe(),
1595 * we are essentially re-initializing communication
1596 * with the server. We really should not run any
1597 * resets/failovers here because this is already a form
1598 * of reset and we do not want parallel resets occurring
1599 */
1600 do {
1601 reinit_init_done(adapter);
1602 /* Clear any failovers we got in the previous
1603 * pass since we are re-initializing the CRQ
1604 */
1605 adapter->failover_pending = false;
1606 release_crq_queue(adapter);
1607 /* If we don't sleep here then we risk an
1608 * unnecessary failover event from the VIOS.
1609 * This is a known VIOS issue caused by a vnic
1610 * device freeing and registering a CRQ too
1611 * quickly.
1612 */
1613 msleep(1500);
1614 /* Avoid any resets, since we are currently
1615 * resetting.
1616 */
1617 spin_lock_irqsave(&adapter->rwi_lock, flags);
1618 flush_reset_queue(adapter);
1619 spin_unlock_irqrestore(&adapter->rwi_lock,
1620 flags);
1621
1622 rc = init_crq_queue(adapter);
1623 if (rc) {
1624 netdev_err(netdev, "login recovery: init CRQ failed %d\n",
1625 rc);
1626 return -EIO;
1627 }
1628
1629 rc = ibmvnic_reset_init(adapter, false);
1630 if (rc)
1631 netdev_err(netdev, "login recovery: Reset init failed %d\n",
1632 rc);
1633 /* IBMVNIC_CRQ_INIT will return EAGAIN if it
1634 * fails, since ibmvnic_reset_init will free
1635 * irq's in failure, we won't be able to receive
1636 * new CRQs so we need to keep trying. probe()
1637 * handles this similarly.
1638 */
1639 } while (rc == -EAGAIN && retry_count++ < retries);
1640 }
1641 } while (retry);
1642
1643 __ibmvnic_set_mac(netdev, adapter->mac_addr);
1644
1645 netdev_dbg(netdev, "[S:%s] Login succeeded\n", adapter_state_to_string(adapter->state));
1646 return 0;
1647 }
1648
release_login_buffer(struct ibmvnic_adapter * adapter)1649 static void release_login_buffer(struct ibmvnic_adapter *adapter)
1650 {
1651 if (!adapter->login_buf)
1652 return;
1653
1654 dma_unmap_single(&adapter->vdev->dev, adapter->login_buf_token,
1655 adapter->login_buf_sz, DMA_TO_DEVICE);
1656 kfree(adapter->login_buf);
1657 adapter->login_buf = NULL;
1658 }
1659
release_login_rsp_buffer(struct ibmvnic_adapter * adapter)1660 static void release_login_rsp_buffer(struct ibmvnic_adapter *adapter)
1661 {
1662 if (!adapter->login_rsp_buf)
1663 return;
1664
1665 dma_unmap_single(&adapter->vdev->dev, adapter->login_rsp_buf_token,
1666 adapter->login_rsp_buf_sz, DMA_FROM_DEVICE);
1667 kfree(adapter->login_rsp_buf);
1668 adapter->login_rsp_buf = NULL;
1669 }
1670
release_resources(struct ibmvnic_adapter * adapter)1671 static void release_resources(struct ibmvnic_adapter *adapter)
1672 {
1673 release_vpd_data(adapter);
1674
1675 release_napi(adapter);
1676 release_login_buffer(adapter);
1677 release_login_rsp_buffer(adapter);
1678 }
1679
set_link_state(struct ibmvnic_adapter * adapter,u8 link_state)1680 static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state)
1681 {
1682 struct net_device *netdev = adapter->netdev;
1683 unsigned long timeout = msecs_to_jiffies(20000);
1684 union ibmvnic_crq crq;
1685 bool resend;
1686 int rc;
1687
1688 netdev_dbg(netdev, "setting link state %d\n", link_state);
1689
1690 memset(&crq, 0, sizeof(crq));
1691 crq.logical_link_state.first = IBMVNIC_CRQ_CMD;
1692 crq.logical_link_state.cmd = LOGICAL_LINK_STATE;
1693 crq.logical_link_state.link_state = link_state;
1694
1695 do {
1696 resend = false;
1697
1698 reinit_completion(&adapter->init_done);
1699 rc = ibmvnic_send_crq(adapter, &crq);
1700 if (rc) {
1701 netdev_err(netdev, "Failed to set link state\n");
1702 return rc;
1703 }
1704
1705 if (!wait_for_completion_timeout(&adapter->init_done,
1706 timeout)) {
1707 netdev_err(netdev, "timeout setting link state\n");
1708 return -ETIMEDOUT;
1709 }
1710
1711 if (adapter->init_done_rc == PARTIALSUCCESS) {
1712 /* Partuial success, delay and re-send */
1713 mdelay(1000);
1714 resend = true;
1715 } else if (adapter->init_done_rc) {
1716 netdev_warn(netdev, "Unable to set link state, rc=%d\n",
1717 adapter->init_done_rc);
1718 return adapter->init_done_rc;
1719 }
1720 } while (resend);
1721
1722 return 0;
1723 }
1724
set_real_num_queues(struct net_device * netdev)1725 static int set_real_num_queues(struct net_device *netdev)
1726 {
1727 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1728 int rc;
1729
1730 netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n",
1731 adapter->req_tx_queues, adapter->req_rx_queues);
1732
1733 rc = netif_set_real_num_tx_queues(netdev, adapter->req_tx_queues);
1734 if (rc) {
1735 netdev_err(netdev, "failed to set the number of tx queues\n");
1736 return rc;
1737 }
1738
1739 rc = netif_set_real_num_rx_queues(netdev, adapter->req_rx_queues);
1740 if (rc)
1741 netdev_err(netdev, "failed to set the number of rx queues\n");
1742
1743 return rc;
1744 }
1745
ibmvnic_get_vpd(struct ibmvnic_adapter * adapter)1746 static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter)
1747 {
1748 struct device *dev = &adapter->vdev->dev;
1749 union ibmvnic_crq crq;
1750 int len = 0;
1751 int rc;
1752
1753 if (adapter->vpd->buff)
1754 len = adapter->vpd->len;
1755
1756 mutex_lock(&adapter->fw_lock);
1757 adapter->fw_done_rc = 0;
1758 reinit_completion(&adapter->fw_done);
1759
1760 crq.get_vpd_size.first = IBMVNIC_CRQ_CMD;
1761 crq.get_vpd_size.cmd = GET_VPD_SIZE;
1762 rc = ibmvnic_send_crq(adapter, &crq);
1763 if (rc) {
1764 mutex_unlock(&adapter->fw_lock);
1765 return rc;
1766 }
1767
1768 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1769 if (rc) {
1770 dev_err(dev, "Could not retrieve VPD size, rc = %d\n", rc);
1771 mutex_unlock(&adapter->fw_lock);
1772 return rc;
1773 }
1774 mutex_unlock(&adapter->fw_lock);
1775
1776 if (!adapter->vpd->len)
1777 return -ENODATA;
1778
1779 if (!adapter->vpd->buff)
1780 adapter->vpd->buff = kzalloc(adapter->vpd->len, GFP_KERNEL);
1781 else if (adapter->vpd->len != len)
1782 adapter->vpd->buff =
1783 krealloc(adapter->vpd->buff,
1784 adapter->vpd->len, GFP_KERNEL);
1785
1786 if (!adapter->vpd->buff) {
1787 dev_err(dev, "Could allocate VPD buffer\n");
1788 return -ENOMEM;
1789 }
1790
1791 adapter->vpd->dma_addr =
1792 dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len,
1793 DMA_FROM_DEVICE);
1794 if (dma_mapping_error(dev, adapter->vpd->dma_addr)) {
1795 dev_err(dev, "Could not map VPD buffer\n");
1796 kfree(adapter->vpd->buff);
1797 adapter->vpd->buff = NULL;
1798 return -ENOMEM;
1799 }
1800
1801 mutex_lock(&adapter->fw_lock);
1802 adapter->fw_done_rc = 0;
1803 reinit_completion(&adapter->fw_done);
1804
1805 crq.get_vpd.first = IBMVNIC_CRQ_CMD;
1806 crq.get_vpd.cmd = GET_VPD;
1807 crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr);
1808 crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len);
1809 rc = ibmvnic_send_crq(adapter, &crq);
1810 if (rc) {
1811 kfree(adapter->vpd->buff);
1812 adapter->vpd->buff = NULL;
1813 mutex_unlock(&adapter->fw_lock);
1814 return rc;
1815 }
1816
1817 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1818 if (rc) {
1819 dev_err(dev, "Unable to retrieve VPD, rc = %d\n", rc);
1820 kfree(adapter->vpd->buff);
1821 adapter->vpd->buff = NULL;
1822 mutex_unlock(&adapter->fw_lock);
1823 return rc;
1824 }
1825
1826 mutex_unlock(&adapter->fw_lock);
1827 return 0;
1828 }
1829
init_resources(struct ibmvnic_adapter * adapter)1830 static int init_resources(struct ibmvnic_adapter *adapter)
1831 {
1832 struct net_device *netdev = adapter->netdev;
1833 int rc;
1834
1835 rc = set_real_num_queues(netdev);
1836 if (rc)
1837 return rc;
1838
1839 adapter->vpd = kzalloc(sizeof(*adapter->vpd), GFP_KERNEL);
1840 if (!adapter->vpd)
1841 return -ENOMEM;
1842
1843 /* Vital Product Data (VPD) */
1844 rc = ibmvnic_get_vpd(adapter);
1845 if (rc) {
1846 netdev_err(netdev, "failed to initialize Vital Product Data (VPD)\n");
1847 return rc;
1848 }
1849
1850 rc = init_napi(adapter);
1851 if (rc)
1852 return rc;
1853
1854 send_query_map(adapter);
1855
1856 rc = init_rx_pools(netdev);
1857 if (rc)
1858 return rc;
1859
1860 rc = init_tx_pools(netdev);
1861 return rc;
1862 }
1863
__ibmvnic_open(struct net_device * netdev)1864 static int __ibmvnic_open(struct net_device *netdev)
1865 {
1866 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1867 enum vnic_state prev_state = adapter->state;
1868 int i, rc;
1869
1870 adapter->state = VNIC_OPENING;
1871 replenish_pools(adapter);
1872 ibmvnic_napi_enable(adapter);
1873
1874 /* We're ready to receive frames, enable the sub-crq interrupts and
1875 * set the logical link state to up
1876 */
1877 for (i = 0; i < adapter->req_rx_queues; i++) {
1878 netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i);
1879 if (prev_state == VNIC_CLOSED)
1880 enable_irq(adapter->rx_scrq[i]->irq);
1881 enable_scrq_irq(adapter, adapter->rx_scrq[i]);
1882 }
1883
1884 for (i = 0; i < adapter->req_tx_queues; i++) {
1885 netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i);
1886 if (prev_state == VNIC_CLOSED)
1887 enable_irq(adapter->tx_scrq[i]->irq);
1888 enable_scrq_irq(adapter, adapter->tx_scrq[i]);
1889 /* netdev_tx_reset_queue will reset dql stats. During NON_FATAL
1890 * resets, don't reset the stats because there could be batched
1891 * skb's waiting to be sent. If we reset dql stats, we risk
1892 * num_completed being greater than num_queued. This will cause
1893 * a BUG_ON in dql_completed().
1894 */
1895 if (adapter->reset_reason != VNIC_RESET_NON_FATAL)
1896 netdev_tx_reset_queue(netdev_get_tx_queue(netdev, i));
1897 }
1898
1899 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP);
1900 if (rc) {
1901 ibmvnic_napi_disable(adapter);
1902 ibmvnic_disable_irqs(adapter);
1903 return rc;
1904 }
1905
1906 adapter->tx_queues_active = true;
1907
1908 /* Since queues were stopped until now, there shouldn't be any
1909 * one in ibmvnic_complete_tx() or ibmvnic_xmit() so maybe we
1910 * don't need the synchronize_rcu()? Leaving it for consistency
1911 * with setting ->tx_queues_active = false.
1912 */
1913 synchronize_rcu();
1914
1915 netif_tx_start_all_queues(netdev);
1916
1917 if (prev_state == VNIC_CLOSED) {
1918 for (i = 0; i < adapter->req_rx_queues; i++)
1919 napi_schedule(&adapter->napi[i]);
1920 }
1921
1922 adapter->state = VNIC_OPEN;
1923 return rc;
1924 }
1925
ibmvnic_open(struct net_device * netdev)1926 static int ibmvnic_open(struct net_device *netdev)
1927 {
1928 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1929 int rc;
1930
1931 ASSERT_RTNL();
1932
1933 /* If device failover is pending or we are about to reset, just set
1934 * device state and return. Device operation will be handled by reset
1935 * routine.
1936 *
1937 * It should be safe to overwrite the adapter->state here. Since
1938 * we hold the rtnl, either the reset has not actually started or
1939 * the rtnl got dropped during the set_link_state() in do_reset().
1940 * In the former case, no one else is changing the state (again we
1941 * have the rtnl) and in the latter case, do_reset() will detect and
1942 * honor our setting below.
1943 */
1944 if (adapter->failover_pending || (test_bit(0, &adapter->resetting))) {
1945 netdev_dbg(netdev, "[S:%s FOP:%d] Resetting, deferring open\n",
1946 adapter_state_to_string(adapter->state),
1947 adapter->failover_pending);
1948 adapter->state = VNIC_OPEN;
1949 rc = 0;
1950 goto out;
1951 }
1952
1953 if (adapter->state != VNIC_CLOSED) {
1954 rc = ibmvnic_login(netdev);
1955 if (rc)
1956 goto out;
1957
1958 rc = init_resources(adapter);
1959 if (rc) {
1960 netdev_err(netdev, "failed to initialize resources\n");
1961 goto out;
1962 }
1963 }
1964
1965 rc = __ibmvnic_open(netdev);
1966
1967 out:
1968 /* If open failed and there is a pending failover or in-progress reset,
1969 * set device state and return. Device operation will be handled by
1970 * reset routine. See also comments above regarding rtnl.
1971 */
1972 if (rc &&
1973 (adapter->failover_pending || (test_bit(0, &adapter->resetting)))) {
1974 adapter->state = VNIC_OPEN;
1975 rc = 0;
1976 }
1977
1978 if (rc) {
1979 release_resources(adapter);
1980 release_rx_pools(adapter);
1981 release_tx_pools(adapter);
1982 }
1983
1984 return rc;
1985 }
1986
clean_rx_pools(struct ibmvnic_adapter * adapter)1987 static void clean_rx_pools(struct ibmvnic_adapter *adapter)
1988 {
1989 struct ibmvnic_rx_pool *rx_pool;
1990 struct ibmvnic_rx_buff *rx_buff;
1991 u64 rx_entries;
1992 int rx_scrqs;
1993 int i, j;
1994
1995 if (!adapter->rx_pool)
1996 return;
1997
1998 rx_scrqs = adapter->num_active_rx_pools;
1999 rx_entries = adapter->req_rx_add_entries_per_subcrq;
2000
2001 /* Free any remaining skbs in the rx buffer pools */
2002 for (i = 0; i < rx_scrqs; i++) {
2003 rx_pool = &adapter->rx_pool[i];
2004 if (!rx_pool || !rx_pool->rx_buff)
2005 continue;
2006
2007 netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i);
2008 for (j = 0; j < rx_entries; j++) {
2009 rx_buff = &rx_pool->rx_buff[j];
2010 if (rx_buff && rx_buff->skb) {
2011 dev_kfree_skb_any(rx_buff->skb);
2012 rx_buff->skb = NULL;
2013 }
2014 }
2015 }
2016 }
2017
clean_one_tx_pool(struct ibmvnic_adapter * adapter,struct ibmvnic_tx_pool * tx_pool)2018 static void clean_one_tx_pool(struct ibmvnic_adapter *adapter,
2019 struct ibmvnic_tx_pool *tx_pool)
2020 {
2021 struct ibmvnic_tx_buff *tx_buff;
2022 u64 tx_entries;
2023 int i;
2024
2025 if (!tx_pool || !tx_pool->tx_buff)
2026 return;
2027
2028 tx_entries = tx_pool->num_buffers;
2029
2030 for (i = 0; i < tx_entries; i++) {
2031 tx_buff = &tx_pool->tx_buff[i];
2032 if (tx_buff && tx_buff->skb) {
2033 dev_kfree_skb_any(tx_buff->skb);
2034 tx_buff->skb = NULL;
2035 }
2036 }
2037 }
2038
clean_tx_pools(struct ibmvnic_adapter * adapter)2039 static void clean_tx_pools(struct ibmvnic_adapter *adapter)
2040 {
2041 int tx_scrqs;
2042 int i;
2043
2044 if (!adapter->tx_pool || !adapter->tso_pool)
2045 return;
2046
2047 tx_scrqs = adapter->num_active_tx_pools;
2048
2049 /* Free any remaining skbs in the tx buffer pools */
2050 for (i = 0; i < tx_scrqs; i++) {
2051 netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i);
2052 clean_one_tx_pool(adapter, &adapter->tx_pool[i]);
2053 clean_one_tx_pool(adapter, &adapter->tso_pool[i]);
2054 }
2055 }
2056
ibmvnic_disable_irqs(struct ibmvnic_adapter * adapter)2057 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter)
2058 {
2059 struct net_device *netdev = adapter->netdev;
2060 int i;
2061
2062 if (adapter->tx_scrq) {
2063 for (i = 0; i < adapter->req_tx_queues; i++)
2064 if (adapter->tx_scrq[i]->irq) {
2065 netdev_dbg(netdev,
2066 "Disabling tx_scrq[%d] irq\n", i);
2067 disable_scrq_irq(adapter, adapter->tx_scrq[i]);
2068 disable_irq(adapter->tx_scrq[i]->irq);
2069 }
2070 }
2071
2072 if (adapter->rx_scrq) {
2073 for (i = 0; i < adapter->req_rx_queues; i++) {
2074 if (adapter->rx_scrq[i]->irq) {
2075 netdev_dbg(netdev,
2076 "Disabling rx_scrq[%d] irq\n", i);
2077 disable_scrq_irq(adapter, adapter->rx_scrq[i]);
2078 disable_irq(adapter->rx_scrq[i]->irq);
2079 }
2080 }
2081 }
2082 }
2083
ibmvnic_cleanup(struct net_device * netdev)2084 static void ibmvnic_cleanup(struct net_device *netdev)
2085 {
2086 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2087
2088 /* ensure that transmissions are stopped if called by do_reset */
2089
2090 adapter->tx_queues_active = false;
2091
2092 /* Ensure complete_tx() and ibmvnic_xmit() see ->tx_queues_active
2093 * update so they don't restart a queue after we stop it below.
2094 */
2095 synchronize_rcu();
2096
2097 if (test_bit(0, &adapter->resetting))
2098 netif_tx_disable(netdev);
2099 else
2100 netif_tx_stop_all_queues(netdev);
2101
2102 ibmvnic_napi_disable(adapter);
2103 ibmvnic_disable_irqs(adapter);
2104 }
2105
__ibmvnic_close(struct net_device * netdev)2106 static int __ibmvnic_close(struct net_device *netdev)
2107 {
2108 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2109 int rc = 0;
2110
2111 adapter->state = VNIC_CLOSING;
2112 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
2113 adapter->state = VNIC_CLOSED;
2114 return rc;
2115 }
2116
ibmvnic_close(struct net_device * netdev)2117 static int ibmvnic_close(struct net_device *netdev)
2118 {
2119 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2120 int rc;
2121
2122 netdev_dbg(netdev, "[S:%s FOP:%d FRR:%d] Closing\n",
2123 adapter_state_to_string(adapter->state),
2124 adapter->failover_pending,
2125 adapter->force_reset_recovery);
2126
2127 /* If device failover is pending, just set device state and return.
2128 * Device operation will be handled by reset routine.
2129 */
2130 if (adapter->failover_pending) {
2131 adapter->state = VNIC_CLOSED;
2132 return 0;
2133 }
2134
2135 rc = __ibmvnic_close(netdev);
2136 ibmvnic_cleanup(netdev);
2137 clean_rx_pools(adapter);
2138 clean_tx_pools(adapter);
2139
2140 return rc;
2141 }
2142
2143 /**
2144 * build_hdr_data - creates L2/L3/L4 header data buffer
2145 * @hdr_field: bitfield determining needed headers
2146 * @skb: socket buffer
2147 * @hdr_len: array of header lengths
2148 * @hdr_data: buffer to write the header to
2149 *
2150 * Reads hdr_field to determine which headers are needed by firmware.
2151 * Builds a buffer containing these headers. Saves individual header
2152 * lengths and total buffer length to be used to build descriptors.
2153 */
build_hdr_data(u8 hdr_field,struct sk_buff * skb,int * hdr_len,u8 * hdr_data)2154 static int build_hdr_data(u8 hdr_field, struct sk_buff *skb,
2155 int *hdr_len, u8 *hdr_data)
2156 {
2157 int len = 0;
2158 u8 *hdr;
2159
2160 if (skb_vlan_tagged(skb) && !skb_vlan_tag_present(skb))
2161 hdr_len[0] = sizeof(struct vlan_ethhdr);
2162 else
2163 hdr_len[0] = sizeof(struct ethhdr);
2164
2165 if (skb->protocol == htons(ETH_P_IP)) {
2166 hdr_len[1] = ip_hdr(skb)->ihl * 4;
2167 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2168 hdr_len[2] = tcp_hdrlen(skb);
2169 else if (ip_hdr(skb)->protocol == IPPROTO_UDP)
2170 hdr_len[2] = sizeof(struct udphdr);
2171 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2172 hdr_len[1] = sizeof(struct ipv6hdr);
2173 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2174 hdr_len[2] = tcp_hdrlen(skb);
2175 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
2176 hdr_len[2] = sizeof(struct udphdr);
2177 } else if (skb->protocol == htons(ETH_P_ARP)) {
2178 hdr_len[1] = arp_hdr_len(skb->dev);
2179 hdr_len[2] = 0;
2180 }
2181
2182 memset(hdr_data, 0, 120);
2183 if ((hdr_field >> 6) & 1) {
2184 hdr = skb_mac_header(skb);
2185 memcpy(hdr_data, hdr, hdr_len[0]);
2186 len += hdr_len[0];
2187 }
2188
2189 if ((hdr_field >> 5) & 1) {
2190 hdr = skb_network_header(skb);
2191 memcpy(hdr_data + len, hdr, hdr_len[1]);
2192 len += hdr_len[1];
2193 }
2194
2195 if ((hdr_field >> 4) & 1) {
2196 hdr = skb_transport_header(skb);
2197 memcpy(hdr_data + len, hdr, hdr_len[2]);
2198 len += hdr_len[2];
2199 }
2200 return len;
2201 }
2202
2203 /**
2204 * create_hdr_descs - create header and header extension descriptors
2205 * @hdr_field: bitfield determining needed headers
2206 * @hdr_data: buffer containing header data
2207 * @len: length of data buffer
2208 * @hdr_len: array of individual header lengths
2209 * @scrq_arr: descriptor array
2210 *
2211 * Creates header and, if needed, header extension descriptors and
2212 * places them in a descriptor array, scrq_arr
2213 */
2214
create_hdr_descs(u8 hdr_field,u8 * hdr_data,int len,int * hdr_len,union sub_crq * scrq_arr)2215 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len,
2216 union sub_crq *scrq_arr)
2217 {
2218 union sub_crq hdr_desc;
2219 int tmp_len = len;
2220 int num_descs = 0;
2221 u8 *data, *cur;
2222 int tmp;
2223
2224 while (tmp_len > 0) {
2225 cur = hdr_data + len - tmp_len;
2226
2227 memset(&hdr_desc, 0, sizeof(hdr_desc));
2228 if (cur != hdr_data) {
2229 data = hdr_desc.hdr_ext.data;
2230 tmp = tmp_len > 29 ? 29 : tmp_len;
2231 hdr_desc.hdr_ext.first = IBMVNIC_CRQ_CMD;
2232 hdr_desc.hdr_ext.type = IBMVNIC_HDR_EXT_DESC;
2233 hdr_desc.hdr_ext.len = tmp;
2234 } else {
2235 data = hdr_desc.hdr.data;
2236 tmp = tmp_len > 24 ? 24 : tmp_len;
2237 hdr_desc.hdr.first = IBMVNIC_CRQ_CMD;
2238 hdr_desc.hdr.type = IBMVNIC_HDR_DESC;
2239 hdr_desc.hdr.len = tmp;
2240 hdr_desc.hdr.l2_len = (u8)hdr_len[0];
2241 hdr_desc.hdr.l3_len = cpu_to_be16((u16)hdr_len[1]);
2242 hdr_desc.hdr.l4_len = (u8)hdr_len[2];
2243 hdr_desc.hdr.flag = hdr_field << 1;
2244 }
2245 memcpy(data, cur, tmp);
2246 tmp_len -= tmp;
2247 *scrq_arr = hdr_desc;
2248 scrq_arr++;
2249 num_descs++;
2250 }
2251
2252 return num_descs;
2253 }
2254
2255 /**
2256 * build_hdr_descs_arr - build a header descriptor array
2257 * @skb: tx socket buffer
2258 * @indir_arr: indirect array
2259 * @num_entries: number of descriptors to be sent
2260 * @hdr_field: bit field determining which headers will be sent
2261 *
2262 * This function will build a TX descriptor array with applicable
2263 * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect.
2264 */
2265
build_hdr_descs_arr(struct sk_buff * skb,union sub_crq * indir_arr,int * num_entries,u8 hdr_field)2266 static void build_hdr_descs_arr(struct sk_buff *skb,
2267 union sub_crq *indir_arr,
2268 int *num_entries, u8 hdr_field)
2269 {
2270 int hdr_len[3] = {0, 0, 0};
2271 u8 hdr_data[140] = {0};
2272 int tot_len;
2273
2274 tot_len = build_hdr_data(hdr_field, skb, hdr_len,
2275 hdr_data);
2276 *num_entries += create_hdr_descs(hdr_field, hdr_data, tot_len, hdr_len,
2277 indir_arr + 1);
2278 }
2279
ibmvnic_xmit_workarounds(struct sk_buff * skb,struct net_device * netdev)2280 static int ibmvnic_xmit_workarounds(struct sk_buff *skb,
2281 struct net_device *netdev)
2282 {
2283 /* For some backing devices, mishandling of small packets
2284 * can result in a loss of connection or TX stall. Device
2285 * architects recommend that no packet should be smaller
2286 * than the minimum MTU value provided to the driver, so
2287 * pad any packets to that length
2288 */
2289 if (skb->len < netdev->min_mtu)
2290 return skb_put_padto(skb, netdev->min_mtu);
2291
2292 return 0;
2293 }
2294
ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter * adapter,struct ibmvnic_sub_crq_queue * tx_scrq)2295 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
2296 struct ibmvnic_sub_crq_queue *tx_scrq)
2297 {
2298 struct ibmvnic_ind_xmit_queue *ind_bufp;
2299 struct ibmvnic_tx_buff *tx_buff;
2300 struct ibmvnic_tx_pool *tx_pool;
2301 union sub_crq tx_scrq_entry;
2302 int queue_num;
2303 int entries;
2304 int index;
2305 int i;
2306
2307 ind_bufp = &tx_scrq->ind_buf;
2308 entries = (u64)ind_bufp->index;
2309 queue_num = tx_scrq->pool_index;
2310
2311 for (i = entries - 1; i >= 0; --i) {
2312 tx_scrq_entry = ind_bufp->indir_arr[i];
2313 if (tx_scrq_entry.v1.type != IBMVNIC_TX_DESC)
2314 continue;
2315 index = be32_to_cpu(tx_scrq_entry.v1.correlator);
2316 if (index & IBMVNIC_TSO_POOL_MASK) {
2317 tx_pool = &adapter->tso_pool[queue_num];
2318 index &= ~IBMVNIC_TSO_POOL_MASK;
2319 } else {
2320 tx_pool = &adapter->tx_pool[queue_num];
2321 }
2322 tx_pool->free_map[tx_pool->consumer_index] = index;
2323 tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
2324 tx_pool->num_buffers - 1 :
2325 tx_pool->consumer_index - 1;
2326 tx_buff = &tx_pool->tx_buff[index];
2327 adapter->netdev->stats.tx_packets--;
2328 adapter->netdev->stats.tx_bytes -= tx_buff->skb->len;
2329 adapter->tx_stats_buffers[queue_num].batched_packets--;
2330 adapter->tx_stats_buffers[queue_num].bytes -=
2331 tx_buff->skb->len;
2332 dev_kfree_skb_any(tx_buff->skb);
2333 tx_buff->skb = NULL;
2334 adapter->netdev->stats.tx_dropped++;
2335 }
2336
2337 ind_bufp->index = 0;
2338
2339 if (atomic_sub_return(entries, &tx_scrq->used) <=
2340 (adapter->req_tx_entries_per_subcrq / 2) &&
2341 __netif_subqueue_stopped(adapter->netdev, queue_num)) {
2342 rcu_read_lock();
2343
2344 if (adapter->tx_queues_active) {
2345 netif_wake_subqueue(adapter->netdev, queue_num);
2346 netdev_dbg(adapter->netdev, "Started queue %d\n",
2347 queue_num);
2348 }
2349
2350 rcu_read_unlock();
2351 }
2352 }
2353
send_subcrq_direct(struct ibmvnic_adapter * adapter,u64 remote_handle,u64 * entry)2354 static int send_subcrq_direct(struct ibmvnic_adapter *adapter,
2355 u64 remote_handle, u64 *entry)
2356 {
2357 unsigned int ua = adapter->vdev->unit_address;
2358 struct device *dev = &adapter->vdev->dev;
2359 int rc;
2360
2361 /* Make sure the hypervisor sees the complete request */
2362 dma_wmb();
2363 rc = plpar_hcall_norets(H_SEND_SUB_CRQ, ua,
2364 cpu_to_be64(remote_handle),
2365 cpu_to_be64(entry[0]), cpu_to_be64(entry[1]),
2366 cpu_to_be64(entry[2]), cpu_to_be64(entry[3]));
2367
2368 if (rc)
2369 print_subcrq_error(dev, rc, __func__);
2370
2371 return rc;
2372 }
2373
ibmvnic_tx_scrq_flush(struct ibmvnic_adapter * adapter,struct ibmvnic_sub_crq_queue * tx_scrq,bool indirect)2374 static int ibmvnic_tx_scrq_flush(struct ibmvnic_adapter *adapter,
2375 struct ibmvnic_sub_crq_queue *tx_scrq,
2376 bool indirect)
2377 {
2378 struct ibmvnic_ind_xmit_queue *ind_bufp;
2379 u64 dma_addr;
2380 u64 entries;
2381 u64 handle;
2382 int rc;
2383
2384 ind_bufp = &tx_scrq->ind_buf;
2385 dma_addr = (u64)ind_bufp->indir_dma;
2386 entries = (u64)ind_bufp->index;
2387 handle = tx_scrq->handle;
2388
2389 if (!entries)
2390 return 0;
2391
2392 if (indirect)
2393 rc = send_subcrq_indirect(adapter, handle, dma_addr, entries);
2394 else
2395 rc = send_subcrq_direct(adapter, handle,
2396 (u64 *)ind_bufp->indir_arr);
2397
2398 if (rc)
2399 ibmvnic_tx_scrq_clean_buffer(adapter, tx_scrq);
2400 else
2401 ind_bufp->index = 0;
2402 return rc;
2403 }
2404
ibmvnic_xmit(struct sk_buff * skb,struct net_device * netdev)2405 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
2406 {
2407 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2408 int queue_num = skb_get_queue_mapping(skb);
2409 u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req;
2410 struct device *dev = &adapter->vdev->dev;
2411 struct ibmvnic_ind_xmit_queue *ind_bufp;
2412 struct ibmvnic_tx_buff *tx_buff = NULL;
2413 struct ibmvnic_sub_crq_queue *tx_scrq;
2414 struct ibmvnic_long_term_buff *ltb;
2415 struct ibmvnic_tx_pool *tx_pool;
2416 unsigned int tx_send_failed = 0;
2417 netdev_tx_t ret = NETDEV_TX_OK;
2418 unsigned int tx_map_failed = 0;
2419 union sub_crq indir_arr[16];
2420 unsigned int tx_dropped = 0;
2421 unsigned int tx_dpackets = 0;
2422 unsigned int tx_bpackets = 0;
2423 unsigned int tx_bytes = 0;
2424 dma_addr_t data_dma_addr;
2425 struct netdev_queue *txq;
2426 unsigned long lpar_rc;
2427 unsigned int skblen;
2428 union sub_crq tx_crq;
2429 unsigned int offset;
2430 int num_entries = 1;
2431 unsigned char *dst;
2432 int bufidx = 0;
2433 u8 proto = 0;
2434
2435 /* If a reset is in progress, drop the packet since
2436 * the scrqs may get torn down. Otherwise use the
2437 * rcu to ensure reset waits for us to complete.
2438 */
2439 rcu_read_lock();
2440 if (!adapter->tx_queues_active) {
2441 dev_kfree_skb_any(skb);
2442
2443 tx_send_failed++;
2444 tx_dropped++;
2445 ret = NETDEV_TX_OK;
2446 goto out;
2447 }
2448
2449 tx_scrq = adapter->tx_scrq[queue_num];
2450 txq = netdev_get_tx_queue(netdev, queue_num);
2451 ind_bufp = &tx_scrq->ind_buf;
2452
2453 if (ibmvnic_xmit_workarounds(skb, netdev)) {
2454 tx_dropped++;
2455 tx_send_failed++;
2456 ret = NETDEV_TX_OK;
2457 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq, true);
2458 if (lpar_rc != H_SUCCESS)
2459 goto tx_err;
2460 goto out;
2461 }
2462
2463 if (skb_is_gso(skb))
2464 tx_pool = &adapter->tso_pool[queue_num];
2465 else
2466 tx_pool = &adapter->tx_pool[queue_num];
2467
2468 bufidx = tx_pool->free_map[tx_pool->consumer_index];
2469
2470 if (bufidx == IBMVNIC_INVALID_MAP) {
2471 dev_kfree_skb_any(skb);
2472 tx_send_failed++;
2473 tx_dropped++;
2474 ret = NETDEV_TX_OK;
2475 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq, true);
2476 if (lpar_rc != H_SUCCESS)
2477 goto tx_err;
2478 goto out;
2479 }
2480
2481 tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP;
2482
2483 map_txpool_buf_to_ltb(tx_pool, bufidx, <b, &offset);
2484
2485 dst = ltb->buff + offset;
2486 memset(dst, 0, tx_pool->buf_size);
2487 data_dma_addr = ltb->addr + offset;
2488
2489 if (skb_shinfo(skb)->nr_frags) {
2490 int cur, i;
2491
2492 /* Copy the head */
2493 skb_copy_from_linear_data(skb, dst, skb_headlen(skb));
2494 cur = skb_headlen(skb);
2495
2496 /* Copy the frags */
2497 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2498 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2499
2500 memcpy(dst + cur, skb_frag_address(frag),
2501 skb_frag_size(frag));
2502 cur += skb_frag_size(frag);
2503 }
2504 } else {
2505 skb_copy_from_linear_data(skb, dst, skb->len);
2506 }
2507
2508 /* post changes to long_term_buff *dst before VIOS accessing it */
2509 dma_wmb();
2510
2511 tx_pool->consumer_index =
2512 (tx_pool->consumer_index + 1) % tx_pool->num_buffers;
2513
2514 tx_buff = &tx_pool->tx_buff[bufidx];
2515
2516 /* Sanity checks on our free map to make sure it points to an index
2517 * that is not being occupied by another skb. If skb memory is
2518 * not freed then we see congestion control kick in and halt tx.
2519 */
2520 if (unlikely(tx_buff->skb)) {
2521 dev_warn_ratelimited(dev, "TX free map points to untracked skb (%s %d idx=%d)\n",
2522 skb_is_gso(skb) ? "tso_pool" : "tx_pool",
2523 queue_num, bufidx);
2524 dev_kfree_skb_any(tx_buff->skb);
2525 }
2526
2527 tx_buff->skb = skb;
2528 tx_buff->index = bufidx;
2529 tx_buff->pool_index = queue_num;
2530 skblen = skb->len;
2531
2532 memset(&tx_crq, 0, sizeof(tx_crq));
2533 tx_crq.v1.first = IBMVNIC_CRQ_CMD;
2534 tx_crq.v1.type = IBMVNIC_TX_DESC;
2535 tx_crq.v1.n_crq_elem = 1;
2536 tx_crq.v1.n_sge = 1;
2537 tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED;
2538
2539 if (skb_is_gso(skb))
2540 tx_crq.v1.correlator =
2541 cpu_to_be32(bufidx | IBMVNIC_TSO_POOL_MASK);
2542 else
2543 tx_crq.v1.correlator = cpu_to_be32(bufidx);
2544 tx_crq.v1.dma_reg = cpu_to_be16(ltb->map_id);
2545 tx_crq.v1.sge_len = cpu_to_be32(skb->len);
2546 tx_crq.v1.ioba = cpu_to_be64(data_dma_addr);
2547
2548 if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) {
2549 tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT;
2550 tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci);
2551 }
2552
2553 if (skb->protocol == htons(ETH_P_IP)) {
2554 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4;
2555 proto = ip_hdr(skb)->protocol;
2556 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2557 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6;
2558 proto = ipv6_hdr(skb)->nexthdr;
2559 }
2560
2561 if (proto == IPPROTO_TCP)
2562 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP;
2563 else if (proto == IPPROTO_UDP)
2564 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP;
2565
2566 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2567 tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD;
2568 hdrs += 2;
2569 }
2570 if (skb_is_gso(skb)) {
2571 tx_crq.v1.flags1 |= IBMVNIC_TX_LSO;
2572 tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size);
2573 hdrs += 2;
2574 } else if (!ind_bufp->index && !netdev_xmit_more()) {
2575 ind_bufp->indir_arr[0] = tx_crq;
2576 ind_bufp->index = 1;
2577 tx_buff->num_entries = 1;
2578 netdev_tx_sent_queue(txq, skb->len);
2579 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq, false);
2580 if (lpar_rc != H_SUCCESS)
2581 goto tx_err;
2582
2583 tx_dpackets++;
2584 goto early_exit;
2585 }
2586
2587 if ((*hdrs >> 7) & 1)
2588 build_hdr_descs_arr(skb, indir_arr, &num_entries, *hdrs);
2589
2590 tx_crq.v1.n_crq_elem = num_entries;
2591 tx_buff->num_entries = num_entries;
2592 /* flush buffer if current entry can not fit */
2593 if (num_entries + ind_bufp->index > IBMVNIC_MAX_IND_DESCS) {
2594 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq, true);
2595 if (lpar_rc != H_SUCCESS)
2596 goto tx_flush_err;
2597 }
2598
2599 indir_arr[0] = tx_crq;
2600 memcpy(&ind_bufp->indir_arr[ind_bufp->index], &indir_arr[0],
2601 num_entries * sizeof(struct ibmvnic_generic_scrq));
2602
2603 ind_bufp->index += num_entries;
2604 if (__netdev_tx_sent_queue(txq, skb->len,
2605 netdev_xmit_more() &&
2606 ind_bufp->index < IBMVNIC_MAX_IND_DESCS)) {
2607 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq, true);
2608 if (lpar_rc != H_SUCCESS)
2609 goto tx_err;
2610 }
2611
2612 tx_bpackets++;
2613
2614 early_exit:
2615 if (atomic_add_return(num_entries, &tx_scrq->used)
2616 >= adapter->req_tx_entries_per_subcrq) {
2617 netdev_dbg(netdev, "Stopping queue %d\n", queue_num);
2618 netif_stop_subqueue(netdev, queue_num);
2619 }
2620
2621 tx_bytes += skblen;
2622 txq_trans_cond_update(txq);
2623 ret = NETDEV_TX_OK;
2624 goto out;
2625
2626 tx_flush_err:
2627 dev_kfree_skb_any(skb);
2628 tx_buff->skb = NULL;
2629 tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
2630 tx_pool->num_buffers - 1 :
2631 tx_pool->consumer_index - 1;
2632 tx_dropped++;
2633 tx_err:
2634 if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER)
2635 dev_err_ratelimited(dev, "tx: send failed\n");
2636
2637 if (lpar_rc == H_CLOSED || adapter->failover_pending) {
2638 /* Disable TX and report carrier off if queue is closed
2639 * or pending failover.
2640 * Firmware guarantees that a signal will be sent to the
2641 * driver, triggering a reset or some other action.
2642 */
2643 netif_tx_stop_all_queues(netdev);
2644 netif_carrier_off(netdev);
2645 }
2646 out:
2647 rcu_read_unlock();
2648 netdev->stats.tx_dropped += tx_dropped;
2649 netdev->stats.tx_bytes += tx_bytes;
2650 netdev->stats.tx_packets += tx_bpackets + tx_dpackets;
2651 adapter->tx_send_failed += tx_send_failed;
2652 adapter->tx_map_failed += tx_map_failed;
2653 adapter->tx_stats_buffers[queue_num].batched_packets += tx_bpackets;
2654 adapter->tx_stats_buffers[queue_num].direct_packets += tx_dpackets;
2655 adapter->tx_stats_buffers[queue_num].bytes += tx_bytes;
2656 adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped;
2657
2658 return ret;
2659 }
2660
ibmvnic_set_multi(struct net_device * netdev)2661 static void ibmvnic_set_multi(struct net_device *netdev)
2662 {
2663 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2664 struct netdev_hw_addr *ha;
2665 union ibmvnic_crq crq;
2666
2667 memset(&crq, 0, sizeof(crq));
2668 crq.request_capability.first = IBMVNIC_CRQ_CMD;
2669 crq.request_capability.cmd = REQUEST_CAPABILITY;
2670
2671 if (netdev->flags & IFF_PROMISC) {
2672 if (!adapter->promisc_supported)
2673 return;
2674 } else {
2675 if (netdev->flags & IFF_ALLMULTI) {
2676 /* Accept all multicast */
2677 memset(&crq, 0, sizeof(crq));
2678 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2679 crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2680 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL;
2681 ibmvnic_send_crq(adapter, &crq);
2682 } else if (netdev_mc_empty(netdev)) {
2683 /* Reject all multicast */
2684 memset(&crq, 0, sizeof(crq));
2685 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2686 crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2687 crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL;
2688 ibmvnic_send_crq(adapter, &crq);
2689 } else {
2690 /* Accept one or more multicast(s) */
2691 netdev_for_each_mc_addr(ha, netdev) {
2692 memset(&crq, 0, sizeof(crq));
2693 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2694 crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2695 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC;
2696 ether_addr_copy(&crq.multicast_ctrl.mac_addr[0],
2697 ha->addr);
2698 ibmvnic_send_crq(adapter, &crq);
2699 }
2700 }
2701 }
2702 }
2703
__ibmvnic_set_mac(struct net_device * netdev,u8 * dev_addr)2704 static int __ibmvnic_set_mac(struct net_device *netdev, u8 *dev_addr)
2705 {
2706 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2707 union ibmvnic_crq crq;
2708 int rc;
2709
2710 if (!is_valid_ether_addr(dev_addr)) {
2711 rc = -EADDRNOTAVAIL;
2712 goto err;
2713 }
2714
2715 memset(&crq, 0, sizeof(crq));
2716 crq.change_mac_addr.first = IBMVNIC_CRQ_CMD;
2717 crq.change_mac_addr.cmd = CHANGE_MAC_ADDR;
2718 ether_addr_copy(&crq.change_mac_addr.mac_addr[0], dev_addr);
2719
2720 mutex_lock(&adapter->fw_lock);
2721 adapter->fw_done_rc = 0;
2722 reinit_completion(&adapter->fw_done);
2723
2724 rc = ibmvnic_send_crq(adapter, &crq);
2725 if (rc) {
2726 rc = -EIO;
2727 mutex_unlock(&adapter->fw_lock);
2728 goto err;
2729 }
2730
2731 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
2732 /* netdev->dev_addr is changed in handle_change_mac_rsp function */
2733 if (rc || adapter->fw_done_rc) {
2734 rc = -EIO;
2735 mutex_unlock(&adapter->fw_lock);
2736 goto err;
2737 }
2738 mutex_unlock(&adapter->fw_lock);
2739 return 0;
2740 err:
2741 ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
2742 return rc;
2743 }
2744
ibmvnic_set_mac(struct net_device * netdev,void * p)2745 static int ibmvnic_set_mac(struct net_device *netdev, void *p)
2746 {
2747 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2748 struct sockaddr *addr = p;
2749 int rc;
2750
2751 rc = 0;
2752 if (!is_valid_ether_addr(addr->sa_data))
2753 return -EADDRNOTAVAIL;
2754
2755 ether_addr_copy(adapter->mac_addr, addr->sa_data);
2756 if (adapter->state != VNIC_PROBED)
2757 rc = __ibmvnic_set_mac(netdev, addr->sa_data);
2758
2759 return rc;
2760 }
2761
reset_reason_to_string(enum ibmvnic_reset_reason reason)2762 static const char *reset_reason_to_string(enum ibmvnic_reset_reason reason)
2763 {
2764 switch (reason) {
2765 case VNIC_RESET_FAILOVER:
2766 return "FAILOVER";
2767 case VNIC_RESET_MOBILITY:
2768 return "MOBILITY";
2769 case VNIC_RESET_FATAL:
2770 return "FATAL";
2771 case VNIC_RESET_NON_FATAL:
2772 return "NON_FATAL";
2773 case VNIC_RESET_TIMEOUT:
2774 return "TIMEOUT";
2775 case VNIC_RESET_CHANGE_PARAM:
2776 return "CHANGE_PARAM";
2777 case VNIC_RESET_PASSIVE_INIT:
2778 return "PASSIVE_INIT";
2779 }
2780 return "UNKNOWN";
2781 }
2782
2783 /*
2784 * Initialize the init_done completion and return code values. We
2785 * can get a transport event just after registering the CRQ and the
2786 * tasklet will use this to communicate the transport event. To ensure
2787 * we don't miss the notification/error, initialize these _before_
2788 * regisering the CRQ.
2789 */
reinit_init_done(struct ibmvnic_adapter * adapter)2790 static inline void reinit_init_done(struct ibmvnic_adapter *adapter)
2791 {
2792 reinit_completion(&adapter->init_done);
2793 adapter->init_done_rc = 0;
2794 }
2795
2796 /*
2797 * do_reset returns zero if we are able to keep processing reset events, or
2798 * non-zero if we hit a fatal error and must halt.
2799 */
do_reset(struct ibmvnic_adapter * adapter,struct ibmvnic_rwi * rwi,u32 reset_state)2800 static int do_reset(struct ibmvnic_adapter *adapter,
2801 struct ibmvnic_rwi *rwi, u32 reset_state)
2802 {
2803 struct net_device *netdev = adapter->netdev;
2804 u64 old_num_rx_queues, old_num_tx_queues;
2805 u64 old_num_rx_slots, old_num_tx_slots;
2806 int rc;
2807
2808 netdev_dbg(adapter->netdev,
2809 "[S:%s FOP:%d] Reset reason: %s, reset_state: %s\n",
2810 adapter_state_to_string(adapter->state),
2811 adapter->failover_pending,
2812 reset_reason_to_string(rwi->reset_reason),
2813 adapter_state_to_string(reset_state));
2814
2815 adapter->reset_reason = rwi->reset_reason;
2816 /* requestor of VNIC_RESET_CHANGE_PARAM already has the rtnl lock */
2817 if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
2818 rtnl_lock();
2819
2820 /* Now that we have the rtnl lock, clear any pending failover.
2821 * This will ensure ibmvnic_open() has either completed or will
2822 * block until failover is complete.
2823 */
2824 if (rwi->reset_reason == VNIC_RESET_FAILOVER)
2825 adapter->failover_pending = false;
2826
2827 /* read the state and check (again) after getting rtnl */
2828 reset_state = adapter->state;
2829
2830 if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
2831 rc = -EBUSY;
2832 goto out;
2833 }
2834
2835 netif_carrier_off(netdev);
2836
2837 old_num_rx_queues = adapter->req_rx_queues;
2838 old_num_tx_queues = adapter->req_tx_queues;
2839 old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq;
2840 old_num_tx_slots = adapter->req_tx_entries_per_subcrq;
2841
2842 ibmvnic_cleanup(netdev);
2843
2844 if (reset_state == VNIC_OPEN &&
2845 adapter->reset_reason != VNIC_RESET_MOBILITY &&
2846 adapter->reset_reason != VNIC_RESET_FAILOVER) {
2847 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2848 rc = __ibmvnic_close(netdev);
2849 if (rc)
2850 goto out;
2851 } else {
2852 adapter->state = VNIC_CLOSING;
2853
2854 /* Release the RTNL lock before link state change and
2855 * re-acquire after the link state change to allow
2856 * linkwatch_event to grab the RTNL lock and run during
2857 * a reset.
2858 */
2859 rtnl_unlock();
2860 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
2861 rtnl_lock();
2862 if (rc)
2863 goto out;
2864
2865 if (adapter->state == VNIC_OPEN) {
2866 /* When we dropped rtnl, ibmvnic_open() got
2867 * it and noticed that we are resetting and
2868 * set the adapter state to OPEN. Update our
2869 * new "target" state, and resume the reset
2870 * from VNIC_CLOSING state.
2871 */
2872 netdev_dbg(netdev,
2873 "Open changed state from %s, updating.\n",
2874 adapter_state_to_string(reset_state));
2875 reset_state = VNIC_OPEN;
2876 adapter->state = VNIC_CLOSING;
2877 }
2878
2879 if (adapter->state != VNIC_CLOSING) {
2880 /* If someone else changed the adapter state
2881 * when we dropped the rtnl, fail the reset
2882 */
2883 rc = -EAGAIN;
2884 goto out;
2885 }
2886 adapter->state = VNIC_CLOSED;
2887 }
2888 }
2889
2890 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2891 release_resources(adapter);
2892 release_sub_crqs(adapter, 1);
2893 release_crq_queue(adapter);
2894 }
2895
2896 if (adapter->reset_reason != VNIC_RESET_NON_FATAL) {
2897 /* remove the closed state so when we call open it appears
2898 * we are coming from the probed state.
2899 */
2900 adapter->state = VNIC_PROBED;
2901
2902 reinit_init_done(adapter);
2903
2904 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2905 rc = init_crq_queue(adapter);
2906 } else if (adapter->reset_reason == VNIC_RESET_MOBILITY) {
2907 rc = ibmvnic_reenable_crq_queue(adapter);
2908 release_sub_crqs(adapter, 1);
2909 } else {
2910 rc = ibmvnic_reset_crq(adapter);
2911 if (rc == H_CLOSED || rc == H_SUCCESS) {
2912 rc = vio_enable_interrupts(adapter->vdev);
2913 if (rc)
2914 netdev_err(adapter->netdev,
2915 "Reset failed to enable interrupts. rc=%d\n",
2916 rc);
2917 }
2918 }
2919
2920 if (rc) {
2921 netdev_err(adapter->netdev,
2922 "Reset couldn't initialize crq. rc=%d\n", rc);
2923 goto out;
2924 }
2925
2926 rc = ibmvnic_reset_init(adapter, true);
2927 if (rc)
2928 goto out;
2929
2930 /* If the adapter was in PROBE or DOWN state prior to the reset,
2931 * exit here.
2932 */
2933 if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN) {
2934 rc = 0;
2935 goto out;
2936 }
2937
2938 rc = ibmvnic_login(netdev);
2939 if (rc)
2940 goto out;
2941
2942 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2943 rc = init_resources(adapter);
2944 if (rc)
2945 goto out;
2946 } else if (adapter->req_rx_queues != old_num_rx_queues ||
2947 adapter->req_tx_queues != old_num_tx_queues ||
2948 adapter->req_rx_add_entries_per_subcrq !=
2949 old_num_rx_slots ||
2950 adapter->req_tx_entries_per_subcrq !=
2951 old_num_tx_slots ||
2952 !adapter->rx_pool ||
2953 !adapter->tso_pool ||
2954 !adapter->tx_pool) {
2955 release_napi(adapter);
2956 release_vpd_data(adapter);
2957
2958 rc = init_resources(adapter);
2959 if (rc)
2960 goto out;
2961
2962 } else {
2963 rc = init_tx_pools(netdev);
2964 if (rc) {
2965 netdev_dbg(netdev,
2966 "init tx pools failed (%d)\n",
2967 rc);
2968 goto out;
2969 }
2970
2971 rc = init_rx_pools(netdev);
2972 if (rc) {
2973 netdev_dbg(netdev,
2974 "init rx pools failed (%d)\n",
2975 rc);
2976 goto out;
2977 }
2978 }
2979 ibmvnic_disable_irqs(adapter);
2980 }
2981 adapter->state = VNIC_CLOSED;
2982
2983 if (reset_state == VNIC_CLOSED) {
2984 rc = 0;
2985 goto out;
2986 }
2987
2988 rc = __ibmvnic_open(netdev);
2989 if (rc) {
2990 rc = IBMVNIC_OPEN_FAILED;
2991 goto out;
2992 }
2993
2994 /* refresh device's multicast list */
2995 ibmvnic_set_multi(netdev);
2996
2997 if (adapter->reset_reason == VNIC_RESET_FAILOVER ||
2998 adapter->reset_reason == VNIC_RESET_MOBILITY)
2999 __netdev_notify_peers(netdev);
3000
3001 rc = 0;
3002
3003 out:
3004 /* restore the adapter state if reset failed */
3005 if (rc)
3006 adapter->state = reset_state;
3007 /* requestor of VNIC_RESET_CHANGE_PARAM should still hold the rtnl lock */
3008 if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
3009 rtnl_unlock();
3010
3011 netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Reset done, rc %d\n",
3012 adapter_state_to_string(adapter->state),
3013 adapter->failover_pending, rc);
3014 return rc;
3015 }
3016
do_hard_reset(struct ibmvnic_adapter * adapter,struct ibmvnic_rwi * rwi,u32 reset_state)3017 static int do_hard_reset(struct ibmvnic_adapter *adapter,
3018 struct ibmvnic_rwi *rwi, u32 reset_state)
3019 {
3020 struct net_device *netdev = adapter->netdev;
3021 int rc;
3022
3023 netdev_dbg(adapter->netdev, "Hard resetting driver (%s)\n",
3024 reset_reason_to_string(rwi->reset_reason));
3025
3026 /* read the state and check (again) after getting rtnl */
3027 reset_state = adapter->state;
3028
3029 if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
3030 rc = -EBUSY;
3031 goto out;
3032 }
3033
3034 netif_carrier_off(netdev);
3035 adapter->reset_reason = rwi->reset_reason;
3036
3037 ibmvnic_cleanup(netdev);
3038 release_resources(adapter);
3039 release_sub_crqs(adapter, 0);
3040 release_crq_queue(adapter);
3041
3042 /* remove the closed state so when we call open it appears
3043 * we are coming from the probed state.
3044 */
3045 adapter->state = VNIC_PROBED;
3046
3047 reinit_init_done(adapter);
3048
3049 rc = init_crq_queue(adapter);
3050 if (rc) {
3051 netdev_err(adapter->netdev,
3052 "Couldn't initialize crq. rc=%d\n", rc);
3053 goto out;
3054 }
3055
3056 rc = ibmvnic_reset_init(adapter, false);
3057 if (rc)
3058 goto out;
3059
3060 /* If the adapter was in PROBE or DOWN state prior to the reset,
3061 * exit here.
3062 */
3063 if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN)
3064 goto out;
3065
3066 rc = ibmvnic_login(netdev);
3067 if (rc)
3068 goto out;
3069
3070 rc = init_resources(adapter);
3071 if (rc)
3072 goto out;
3073
3074 ibmvnic_disable_irqs(adapter);
3075 adapter->state = VNIC_CLOSED;
3076
3077 if (reset_state == VNIC_CLOSED)
3078 goto out;
3079
3080 rc = __ibmvnic_open(netdev);
3081 if (rc) {
3082 rc = IBMVNIC_OPEN_FAILED;
3083 goto out;
3084 }
3085
3086 __netdev_notify_peers(netdev);
3087 out:
3088 /* restore adapter state if reset failed */
3089 if (rc)
3090 adapter->state = reset_state;
3091 netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Hard reset done, rc %d\n",
3092 adapter_state_to_string(adapter->state),
3093 adapter->failover_pending, rc);
3094 return rc;
3095 }
3096
get_next_rwi(struct ibmvnic_adapter * adapter)3097 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter)
3098 {
3099 struct ibmvnic_rwi *rwi;
3100 unsigned long flags;
3101
3102 spin_lock_irqsave(&adapter->rwi_lock, flags);
3103
3104 if (!list_empty(&adapter->rwi_list)) {
3105 rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi,
3106 list);
3107 list_del(&rwi->list);
3108 } else {
3109 rwi = NULL;
3110 }
3111
3112 spin_unlock_irqrestore(&adapter->rwi_lock, flags);
3113 return rwi;
3114 }
3115
3116 /**
3117 * do_passive_init - complete probing when partner device is detected.
3118 * @adapter: ibmvnic_adapter struct
3119 *
3120 * If the ibmvnic device does not have a partner device to communicate with at boot
3121 * and that partner device comes online at a later time, this function is called
3122 * to complete the initialization process of ibmvnic device.
3123 * Caller is expected to hold rtnl_lock().
3124 *
3125 * Returns non-zero if sub-CRQs are not initialized properly leaving the device
3126 * in the down state.
3127 * Returns 0 upon success and the device is in PROBED state.
3128 */
3129
do_passive_init(struct ibmvnic_adapter * adapter)3130 static int do_passive_init(struct ibmvnic_adapter *adapter)
3131 {
3132 unsigned long timeout = msecs_to_jiffies(30000);
3133 struct net_device *netdev = adapter->netdev;
3134 struct device *dev = &adapter->vdev->dev;
3135 int rc;
3136
3137 netdev_dbg(netdev, "Partner device found, probing.\n");
3138
3139 adapter->state = VNIC_PROBING;
3140 reinit_completion(&adapter->init_done);
3141 adapter->init_done_rc = 0;
3142 adapter->crq.active = true;
3143
3144 rc = send_crq_init_complete(adapter);
3145 if (rc)
3146 goto out;
3147
3148 rc = send_version_xchg(adapter);
3149 if (rc)
3150 netdev_dbg(adapter->netdev, "send_version_xchg failed, rc=%d\n", rc);
3151
3152 if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
3153 dev_err(dev, "Initialization sequence timed out\n");
3154 rc = -ETIMEDOUT;
3155 goto out;
3156 }
3157
3158 rc = init_sub_crqs(adapter);
3159 if (rc) {
3160 dev_err(dev, "Initialization of sub crqs failed, rc=%d\n", rc);
3161 goto out;
3162 }
3163
3164 rc = init_sub_crq_irqs(adapter);
3165 if (rc) {
3166 dev_err(dev, "Failed to initialize sub crq irqs\n, rc=%d", rc);
3167 goto init_failed;
3168 }
3169
3170 netdev->mtu = adapter->req_mtu - ETH_HLEN;
3171 netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
3172 netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
3173
3174 adapter->state = VNIC_PROBED;
3175 netdev_dbg(netdev, "Probed successfully. Waiting for signal from partner device.\n");
3176
3177 return 0;
3178
3179 init_failed:
3180 release_sub_crqs(adapter, 1);
3181 out:
3182 adapter->state = VNIC_DOWN;
3183 return rc;
3184 }
3185
__ibmvnic_reset(struct work_struct * work)3186 static void __ibmvnic_reset(struct work_struct *work)
3187 {
3188 struct ibmvnic_adapter *adapter;
3189 unsigned int timeout = 5000;
3190 struct ibmvnic_rwi *tmprwi;
3191 bool saved_state = false;
3192 struct ibmvnic_rwi *rwi;
3193 unsigned long flags;
3194 struct device *dev;
3195 bool need_reset;
3196 int num_fails = 0;
3197 u32 reset_state;
3198 int rc = 0;
3199
3200 adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset);
3201 dev = &adapter->vdev->dev;
3202
3203 /* Wait for ibmvnic_probe() to complete. If probe is taking too long
3204 * or if another reset is in progress, defer work for now. If probe
3205 * eventually fails it will flush and terminate our work.
3206 *
3207 * Three possibilities here:
3208 * 1. Adpater being removed - just return
3209 * 2. Timed out on probe or another reset in progress - delay the work
3210 * 3. Completed probe - perform any resets in queue
3211 */
3212 if (adapter->state == VNIC_PROBING &&
3213 !wait_for_completion_timeout(&adapter->probe_done, timeout)) {
3214 dev_err(dev, "Reset thread timed out on probe");
3215 queue_delayed_work(system_long_wq,
3216 &adapter->ibmvnic_delayed_reset,
3217 IBMVNIC_RESET_DELAY);
3218 return;
3219 }
3220
3221 /* adapter is done with probe (i.e state is never VNIC_PROBING now) */
3222 if (adapter->state == VNIC_REMOVING)
3223 return;
3224
3225 /* ->rwi_list is stable now (no one else is removing entries) */
3226
3227 /* ibmvnic_probe() may have purged the reset queue after we were
3228 * scheduled to process a reset so there maybe no resets to process.
3229 * Before setting the ->resetting bit though, we have to make sure
3230 * that there is infact a reset to process. Otherwise we may race
3231 * with ibmvnic_open() and end up leaving the vnic down:
3232 *
3233 * __ibmvnic_reset() ibmvnic_open()
3234 * ----------------- --------------
3235 *
3236 * set ->resetting bit
3237 * find ->resetting bit is set
3238 * set ->state to IBMVNIC_OPEN (i.e
3239 * assume reset will open device)
3240 * return
3241 * find reset queue empty
3242 * return
3243 *
3244 * Neither performed vnic login/open and vnic stays down
3245 *
3246 * If we hold the lock and conditionally set the bit, either we
3247 * or ibmvnic_open() will complete the open.
3248 */
3249 need_reset = false;
3250 spin_lock(&adapter->rwi_lock);
3251 if (!list_empty(&adapter->rwi_list)) {
3252 if (test_and_set_bit_lock(0, &adapter->resetting)) {
3253 queue_delayed_work(system_long_wq,
3254 &adapter->ibmvnic_delayed_reset,
3255 IBMVNIC_RESET_DELAY);
3256 } else {
3257 need_reset = true;
3258 }
3259 }
3260 spin_unlock(&adapter->rwi_lock);
3261
3262 if (!need_reset)
3263 return;
3264
3265 rwi = get_next_rwi(adapter);
3266 while (rwi) {
3267 spin_lock_irqsave(&adapter->state_lock, flags);
3268
3269 if (adapter->state == VNIC_REMOVING ||
3270 adapter->state == VNIC_REMOVED) {
3271 spin_unlock_irqrestore(&adapter->state_lock, flags);
3272 kfree(rwi);
3273 rc = EBUSY;
3274 break;
3275 }
3276
3277 if (!saved_state) {
3278 reset_state = adapter->state;
3279 saved_state = true;
3280 }
3281 spin_unlock_irqrestore(&adapter->state_lock, flags);
3282
3283 if (rwi->reset_reason == VNIC_RESET_PASSIVE_INIT) {
3284 rtnl_lock();
3285 rc = do_passive_init(adapter);
3286 rtnl_unlock();
3287 if (!rc)
3288 netif_carrier_on(adapter->netdev);
3289 } else if (adapter->force_reset_recovery) {
3290 /* Since we are doing a hard reset now, clear the
3291 * failover_pending flag so we don't ignore any
3292 * future MOBILITY or other resets.
3293 */
3294 adapter->failover_pending = false;
3295
3296 /* Transport event occurred during previous reset */
3297 if (adapter->wait_for_reset) {
3298 /* Previous was CHANGE_PARAM; caller locked */
3299 adapter->force_reset_recovery = false;
3300 rc = do_hard_reset(adapter, rwi, reset_state);
3301 } else {
3302 rtnl_lock();
3303 adapter->force_reset_recovery = false;
3304 rc = do_hard_reset(adapter, rwi, reset_state);
3305 rtnl_unlock();
3306 }
3307 if (rc)
3308 num_fails++;
3309 else
3310 num_fails = 0;
3311
3312 /* If auto-priority-failover is enabled we can get
3313 * back to back failovers during resets, resulting
3314 * in at least two failed resets (from high-priority
3315 * backing device to low-priority one and then back)
3316 * If resets continue to fail beyond that, give the
3317 * adapter some time to settle down before retrying.
3318 */
3319 if (num_fails >= 3) {
3320 netdev_dbg(adapter->netdev,
3321 "[S:%s] Hard reset failed %d times, waiting 60 secs\n",
3322 adapter_state_to_string(adapter->state),
3323 num_fails);
3324 set_current_state(TASK_UNINTERRUPTIBLE);
3325 schedule_timeout(60 * HZ);
3326 }
3327 } else {
3328 rc = do_reset(adapter, rwi, reset_state);
3329 }
3330 tmprwi = rwi;
3331 adapter->last_reset_time = jiffies;
3332
3333 if (rc)
3334 netdev_dbg(adapter->netdev, "Reset failed, rc=%d\n", rc);
3335
3336 rwi = get_next_rwi(adapter);
3337
3338 /*
3339 * If there are no resets queued and the previous reset failed,
3340 * the adapter would be in an undefined state. So retry the
3341 * previous reset as a hard reset.
3342 *
3343 * Else, free the previous rwi and, if there is another reset
3344 * queued, process the new reset even if previous reset failed
3345 * (the previous reset could have failed because of a fail
3346 * over for instance, so process the fail over).
3347 */
3348 if (!rwi && rc)
3349 rwi = tmprwi;
3350 else
3351 kfree(tmprwi);
3352
3353 if (rwi && (rwi->reset_reason == VNIC_RESET_FAILOVER ||
3354 rwi->reset_reason == VNIC_RESET_MOBILITY || rc))
3355 adapter->force_reset_recovery = true;
3356 }
3357
3358 if (adapter->wait_for_reset) {
3359 adapter->reset_done_rc = rc;
3360 complete(&adapter->reset_done);
3361 }
3362
3363 clear_bit_unlock(0, &adapter->resetting);
3364
3365 netdev_dbg(adapter->netdev,
3366 "[S:%s FRR:%d WFR:%d] Done processing resets\n",
3367 adapter_state_to_string(adapter->state),
3368 adapter->force_reset_recovery,
3369 adapter->wait_for_reset);
3370 }
3371
__ibmvnic_delayed_reset(struct work_struct * work)3372 static void __ibmvnic_delayed_reset(struct work_struct *work)
3373 {
3374 struct ibmvnic_adapter *adapter;
3375
3376 adapter = container_of(work, struct ibmvnic_adapter,
3377 ibmvnic_delayed_reset.work);
3378 __ibmvnic_reset(&adapter->ibmvnic_reset);
3379 }
3380
flush_reset_queue(struct ibmvnic_adapter * adapter)3381 static void flush_reset_queue(struct ibmvnic_adapter *adapter)
3382 {
3383 struct list_head *entry, *tmp_entry;
3384
3385 if (!list_empty(&adapter->rwi_list)) {
3386 list_for_each_safe(entry, tmp_entry, &adapter->rwi_list) {
3387 list_del(entry);
3388 kfree(list_entry(entry, struct ibmvnic_rwi, list));
3389 }
3390 }
3391 }
3392
ibmvnic_reset(struct ibmvnic_adapter * adapter,enum ibmvnic_reset_reason reason)3393 static int ibmvnic_reset(struct ibmvnic_adapter *adapter,
3394 enum ibmvnic_reset_reason reason)
3395 {
3396 struct net_device *netdev = adapter->netdev;
3397 struct ibmvnic_rwi *rwi, *tmp;
3398 unsigned long flags;
3399 int ret;
3400
3401 spin_lock_irqsave(&adapter->rwi_lock, flags);
3402
3403 /* If failover is pending don't schedule any other reset.
3404 * Instead let the failover complete. If there is already a
3405 * a failover reset scheduled, we will detect and drop the
3406 * duplicate reset when walking the ->rwi_list below.
3407 */
3408 if (adapter->state == VNIC_REMOVING ||
3409 adapter->state == VNIC_REMOVED ||
3410 (adapter->failover_pending && reason != VNIC_RESET_FAILOVER)) {
3411 ret = EBUSY;
3412 netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n");
3413 goto err;
3414 }
3415
3416 list_for_each_entry(tmp, &adapter->rwi_list, list) {
3417 if (tmp->reset_reason == reason) {
3418 netdev_dbg(netdev, "Skipping matching reset, reason=%s\n",
3419 reset_reason_to_string(reason));
3420 ret = EBUSY;
3421 goto err;
3422 }
3423 }
3424
3425 rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC);
3426 if (!rwi) {
3427 ret = ENOMEM;
3428 goto err;
3429 }
3430 /* if we just received a transport event,
3431 * flush reset queue and process this reset
3432 */
3433 if (adapter->force_reset_recovery)
3434 flush_reset_queue(adapter);
3435
3436 rwi->reset_reason = reason;
3437 list_add_tail(&rwi->list, &adapter->rwi_list);
3438 netdev_dbg(adapter->netdev, "Scheduling reset (reason %s)\n",
3439 reset_reason_to_string(reason));
3440 queue_work(system_long_wq, &adapter->ibmvnic_reset);
3441
3442 ret = 0;
3443 err:
3444 /* ibmvnic_close() below can block, so drop the lock first */
3445 spin_unlock_irqrestore(&adapter->rwi_lock, flags);
3446
3447 if (ret == ENOMEM)
3448 ibmvnic_close(netdev);
3449
3450 return -ret;
3451 }
3452
ibmvnic_tx_timeout(struct net_device * dev,unsigned int txqueue)3453 static void ibmvnic_tx_timeout(struct net_device *dev, unsigned int txqueue)
3454 {
3455 struct ibmvnic_adapter *adapter = netdev_priv(dev);
3456
3457 if (test_bit(0, &adapter->resetting)) {
3458 netdev_err(adapter->netdev,
3459 "Adapter is resetting, skip timeout reset\n");
3460 return;
3461 }
3462 /* No queuing up reset until at least 5 seconds (default watchdog val)
3463 * after last reset
3464 */
3465 if (time_before(jiffies, (adapter->last_reset_time + dev->watchdog_timeo))) {
3466 netdev_dbg(dev, "Not yet time to tx timeout.\n");
3467 return;
3468 }
3469 ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT);
3470 }
3471
remove_buff_from_pool(struct ibmvnic_adapter * adapter,struct ibmvnic_rx_buff * rx_buff)3472 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter,
3473 struct ibmvnic_rx_buff *rx_buff)
3474 {
3475 struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index];
3476
3477 rx_buff->skb = NULL;
3478
3479 pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff);
3480 pool->next_alloc = (pool->next_alloc + 1) % pool->size;
3481
3482 atomic_dec(&pool->available);
3483 }
3484
ibmvnic_poll(struct napi_struct * napi,int budget)3485 static int ibmvnic_poll(struct napi_struct *napi, int budget)
3486 {
3487 struct ibmvnic_sub_crq_queue *rx_scrq;
3488 struct ibmvnic_adapter *adapter;
3489 struct net_device *netdev;
3490 int frames_processed;
3491 int scrq_num;
3492
3493 netdev = napi->dev;
3494 adapter = netdev_priv(netdev);
3495 scrq_num = (int)(napi - adapter->napi);
3496 frames_processed = 0;
3497 rx_scrq = adapter->rx_scrq[scrq_num];
3498
3499 restart_poll:
3500 while (frames_processed < budget) {
3501 struct sk_buff *skb;
3502 struct ibmvnic_rx_buff *rx_buff;
3503 union sub_crq *next;
3504 u32 length;
3505 u16 offset;
3506 u8 flags = 0;
3507
3508 if (unlikely(test_bit(0, &adapter->resetting) &&
3509 adapter->reset_reason != VNIC_RESET_NON_FATAL)) {
3510 enable_scrq_irq(adapter, rx_scrq);
3511 napi_complete_done(napi, frames_processed);
3512 return frames_processed;
3513 }
3514
3515 if (!pending_scrq(adapter, rx_scrq))
3516 break;
3517 next = ibmvnic_next_scrq(adapter, rx_scrq);
3518 rx_buff = (struct ibmvnic_rx_buff *)
3519 be64_to_cpu(next->rx_comp.correlator);
3520 /* do error checking */
3521 if (next->rx_comp.rc) {
3522 netdev_dbg(netdev, "rx buffer returned with rc %x\n",
3523 be16_to_cpu(next->rx_comp.rc));
3524 /* free the entry */
3525 next->rx_comp.first = 0;
3526 dev_kfree_skb_any(rx_buff->skb);
3527 remove_buff_from_pool(adapter, rx_buff);
3528 continue;
3529 } else if (!rx_buff->skb) {
3530 /* free the entry */
3531 next->rx_comp.first = 0;
3532 remove_buff_from_pool(adapter, rx_buff);
3533 continue;
3534 }
3535
3536 length = be32_to_cpu(next->rx_comp.len);
3537 offset = be16_to_cpu(next->rx_comp.off_frame_data);
3538 flags = next->rx_comp.flags;
3539 skb = rx_buff->skb;
3540 /* load long_term_buff before copying to skb */
3541 dma_rmb();
3542 skb_copy_to_linear_data(skb, rx_buff->data + offset,
3543 length);
3544
3545 /* VLAN Header has been stripped by the system firmware and
3546 * needs to be inserted by the driver
3547 */
3548 if (adapter->rx_vlan_header_insertion &&
3549 (flags & IBMVNIC_VLAN_STRIPPED))
3550 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
3551 ntohs(next->rx_comp.vlan_tci));
3552
3553 /* free the entry */
3554 next->rx_comp.first = 0;
3555 remove_buff_from_pool(adapter, rx_buff);
3556
3557 skb_put(skb, length);
3558 skb->protocol = eth_type_trans(skb, netdev);
3559 skb_record_rx_queue(skb, scrq_num);
3560
3561 if (flags & IBMVNIC_IP_CHKSUM_GOOD &&
3562 flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) {
3563 skb->ip_summed = CHECKSUM_UNNECESSARY;
3564 }
3565
3566 length = skb->len;
3567 napi_gro_receive(napi, skb); /* send it up */
3568 netdev->stats.rx_packets++;
3569 netdev->stats.rx_bytes += length;
3570 adapter->rx_stats_buffers[scrq_num].packets++;
3571 adapter->rx_stats_buffers[scrq_num].bytes += length;
3572 frames_processed++;
3573 }
3574
3575 if (adapter->state != VNIC_CLOSING &&
3576 ((atomic_read(&adapter->rx_pool[scrq_num].available) <
3577 adapter->req_rx_add_entries_per_subcrq / 2) ||
3578 frames_processed < budget))
3579 replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]);
3580 if (frames_processed < budget) {
3581 if (napi_complete_done(napi, frames_processed)) {
3582 enable_scrq_irq(adapter, rx_scrq);
3583 if (pending_scrq(adapter, rx_scrq)) {
3584 if (napi_reschedule(napi)) {
3585 disable_scrq_irq(adapter, rx_scrq);
3586 goto restart_poll;
3587 }
3588 }
3589 }
3590 }
3591 return frames_processed;
3592 }
3593
wait_for_reset(struct ibmvnic_adapter * adapter)3594 static int wait_for_reset(struct ibmvnic_adapter *adapter)
3595 {
3596 int rc, ret;
3597
3598 adapter->fallback.mtu = adapter->req_mtu;
3599 adapter->fallback.rx_queues = adapter->req_rx_queues;
3600 adapter->fallback.tx_queues = adapter->req_tx_queues;
3601 adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq;
3602 adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq;
3603
3604 reinit_completion(&adapter->reset_done);
3605 adapter->wait_for_reset = true;
3606 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
3607
3608 if (rc) {
3609 ret = rc;
3610 goto out;
3611 }
3612 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 60000);
3613 if (rc) {
3614 ret = -ENODEV;
3615 goto out;
3616 }
3617
3618 ret = 0;
3619 if (adapter->reset_done_rc) {
3620 ret = -EIO;
3621 adapter->desired.mtu = adapter->fallback.mtu;
3622 adapter->desired.rx_queues = adapter->fallback.rx_queues;
3623 adapter->desired.tx_queues = adapter->fallback.tx_queues;
3624 adapter->desired.rx_entries = adapter->fallback.rx_entries;
3625 adapter->desired.tx_entries = adapter->fallback.tx_entries;
3626
3627 reinit_completion(&adapter->reset_done);
3628 adapter->wait_for_reset = true;
3629 rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
3630 if (rc) {
3631 ret = rc;
3632 goto out;
3633 }
3634 rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done,
3635 60000);
3636 if (rc) {
3637 ret = -ENODEV;
3638 goto out;
3639 }
3640 }
3641 out:
3642 adapter->wait_for_reset = false;
3643
3644 return ret;
3645 }
3646
ibmvnic_change_mtu(struct net_device * netdev,int new_mtu)3647 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu)
3648 {
3649 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3650
3651 adapter->desired.mtu = new_mtu + ETH_HLEN;
3652
3653 return wait_for_reset(adapter);
3654 }
3655
ibmvnic_features_check(struct sk_buff * skb,struct net_device * dev,netdev_features_t features)3656 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb,
3657 struct net_device *dev,
3658 netdev_features_t features)
3659 {
3660 /* Some backing hardware adapters can not
3661 * handle packets with a MSS less than 224
3662 * or with only one segment.
3663 */
3664 if (skb_is_gso(skb)) {
3665 if (skb_shinfo(skb)->gso_size < 224 ||
3666 skb_shinfo(skb)->gso_segs == 1)
3667 features &= ~NETIF_F_GSO_MASK;
3668 }
3669
3670 return features;
3671 }
3672
3673 static const struct net_device_ops ibmvnic_netdev_ops = {
3674 .ndo_open = ibmvnic_open,
3675 .ndo_stop = ibmvnic_close,
3676 .ndo_start_xmit = ibmvnic_xmit,
3677 .ndo_set_rx_mode = ibmvnic_set_multi,
3678 .ndo_set_mac_address = ibmvnic_set_mac,
3679 .ndo_validate_addr = eth_validate_addr,
3680 .ndo_tx_timeout = ibmvnic_tx_timeout,
3681 .ndo_change_mtu = ibmvnic_change_mtu,
3682 .ndo_features_check = ibmvnic_features_check,
3683 };
3684
3685 /* ethtool functions */
3686
ibmvnic_get_link_ksettings(struct net_device * netdev,struct ethtool_link_ksettings * cmd)3687 static int ibmvnic_get_link_ksettings(struct net_device *netdev,
3688 struct ethtool_link_ksettings *cmd)
3689 {
3690 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3691 int rc;
3692
3693 rc = send_query_phys_parms(adapter);
3694 if (rc) {
3695 adapter->speed = SPEED_UNKNOWN;
3696 adapter->duplex = DUPLEX_UNKNOWN;
3697 }
3698 cmd->base.speed = adapter->speed;
3699 cmd->base.duplex = adapter->duplex;
3700 cmd->base.port = PORT_FIBRE;
3701 cmd->base.phy_address = 0;
3702 cmd->base.autoneg = AUTONEG_ENABLE;
3703
3704 return 0;
3705 }
3706
ibmvnic_get_drvinfo(struct net_device * netdev,struct ethtool_drvinfo * info)3707 static void ibmvnic_get_drvinfo(struct net_device *netdev,
3708 struct ethtool_drvinfo *info)
3709 {
3710 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3711
3712 strscpy(info->driver, ibmvnic_driver_name, sizeof(info->driver));
3713 strscpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version));
3714 strscpy(info->fw_version, adapter->fw_version,
3715 sizeof(info->fw_version));
3716 }
3717
ibmvnic_get_msglevel(struct net_device * netdev)3718 static u32 ibmvnic_get_msglevel(struct net_device *netdev)
3719 {
3720 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3721
3722 return adapter->msg_enable;
3723 }
3724
ibmvnic_set_msglevel(struct net_device * netdev,u32 data)3725 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data)
3726 {
3727 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3728
3729 adapter->msg_enable = data;
3730 }
3731
ibmvnic_get_link(struct net_device * netdev)3732 static u32 ibmvnic_get_link(struct net_device *netdev)
3733 {
3734 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3735
3736 /* Don't need to send a query because we request a logical link up at
3737 * init and then we wait for link state indications
3738 */
3739 return adapter->logical_link_state;
3740 }
3741
ibmvnic_get_ringparam(struct net_device * netdev,struct ethtool_ringparam * ring,struct kernel_ethtool_ringparam * kernel_ring,struct netlink_ext_ack * extack)3742 static void ibmvnic_get_ringparam(struct net_device *netdev,
3743 struct ethtool_ringparam *ring,
3744 struct kernel_ethtool_ringparam *kernel_ring,
3745 struct netlink_ext_ack *extack)
3746 {
3747 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3748
3749 ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq;
3750 ring->tx_max_pending = adapter->max_tx_entries_per_subcrq;
3751 ring->rx_mini_max_pending = 0;
3752 ring->rx_jumbo_max_pending = 0;
3753 ring->rx_pending = adapter->req_rx_add_entries_per_subcrq;
3754 ring->tx_pending = adapter->req_tx_entries_per_subcrq;
3755 ring->rx_mini_pending = 0;
3756 ring->rx_jumbo_pending = 0;
3757 }
3758
ibmvnic_set_ringparam(struct net_device * netdev,struct ethtool_ringparam * ring,struct kernel_ethtool_ringparam * kernel_ring,struct netlink_ext_ack * extack)3759 static int ibmvnic_set_ringparam(struct net_device *netdev,
3760 struct ethtool_ringparam *ring,
3761 struct kernel_ethtool_ringparam *kernel_ring,
3762 struct netlink_ext_ack *extack)
3763 {
3764 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3765
3766 if (ring->rx_pending > adapter->max_rx_add_entries_per_subcrq ||
3767 ring->tx_pending > adapter->max_tx_entries_per_subcrq) {
3768 netdev_err(netdev, "Invalid request.\n");
3769 netdev_err(netdev, "Max tx buffers = %llu\n",
3770 adapter->max_rx_add_entries_per_subcrq);
3771 netdev_err(netdev, "Max rx buffers = %llu\n",
3772 adapter->max_tx_entries_per_subcrq);
3773 return -EINVAL;
3774 }
3775
3776 adapter->desired.rx_entries = ring->rx_pending;
3777 adapter->desired.tx_entries = ring->tx_pending;
3778
3779 return wait_for_reset(adapter);
3780 }
3781
ibmvnic_get_channels(struct net_device * netdev,struct ethtool_channels * channels)3782 static void ibmvnic_get_channels(struct net_device *netdev,
3783 struct ethtool_channels *channels)
3784 {
3785 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3786
3787 channels->max_rx = adapter->max_rx_queues;
3788 channels->max_tx = adapter->max_tx_queues;
3789 channels->max_other = 0;
3790 channels->max_combined = 0;
3791 channels->rx_count = adapter->req_rx_queues;
3792 channels->tx_count = adapter->req_tx_queues;
3793 channels->other_count = 0;
3794 channels->combined_count = 0;
3795 }
3796
ibmvnic_set_channels(struct net_device * netdev,struct ethtool_channels * channels)3797 static int ibmvnic_set_channels(struct net_device *netdev,
3798 struct ethtool_channels *channels)
3799 {
3800 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3801
3802 adapter->desired.rx_queues = channels->rx_count;
3803 adapter->desired.tx_queues = channels->tx_count;
3804
3805 return wait_for_reset(adapter);
3806 }
3807
ibmvnic_get_strings(struct net_device * dev,u32 stringset,u8 * data)3808 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data)
3809 {
3810 struct ibmvnic_adapter *adapter = netdev_priv(dev);
3811 int i;
3812
3813 if (stringset != ETH_SS_STATS)
3814 return;
3815
3816 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++, data += ETH_GSTRING_LEN)
3817 memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN);
3818
3819 for (i = 0; i < adapter->req_tx_queues; i++) {
3820 snprintf(data, ETH_GSTRING_LEN, "tx%d_batched_packets", i);
3821 data += ETH_GSTRING_LEN;
3822
3823 snprintf(data, ETH_GSTRING_LEN, "tx%d_direct_packets", i);
3824 data += ETH_GSTRING_LEN;
3825
3826 snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i);
3827 data += ETH_GSTRING_LEN;
3828
3829 snprintf(data, ETH_GSTRING_LEN, "tx%d_dropped_packets", i);
3830 data += ETH_GSTRING_LEN;
3831 }
3832
3833 for (i = 0; i < adapter->req_rx_queues; i++) {
3834 snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i);
3835 data += ETH_GSTRING_LEN;
3836
3837 snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i);
3838 data += ETH_GSTRING_LEN;
3839
3840 snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i);
3841 data += ETH_GSTRING_LEN;
3842 }
3843 }
3844
ibmvnic_get_sset_count(struct net_device * dev,int sset)3845 static int ibmvnic_get_sset_count(struct net_device *dev, int sset)
3846 {
3847 struct ibmvnic_adapter *adapter = netdev_priv(dev);
3848
3849 switch (sset) {
3850 case ETH_SS_STATS:
3851 return ARRAY_SIZE(ibmvnic_stats) +
3852 adapter->req_tx_queues * NUM_TX_STATS +
3853 adapter->req_rx_queues * NUM_RX_STATS;
3854 default:
3855 return -EOPNOTSUPP;
3856 }
3857 }
3858
ibmvnic_get_ethtool_stats(struct net_device * dev,struct ethtool_stats * stats,u64 * data)3859 static void ibmvnic_get_ethtool_stats(struct net_device *dev,
3860 struct ethtool_stats *stats, u64 *data)
3861 {
3862 struct ibmvnic_adapter *adapter = netdev_priv(dev);
3863 union ibmvnic_crq crq;
3864 int i, j;
3865 int rc;
3866
3867 memset(&crq, 0, sizeof(crq));
3868 crq.request_statistics.first = IBMVNIC_CRQ_CMD;
3869 crq.request_statistics.cmd = REQUEST_STATISTICS;
3870 crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token);
3871 crq.request_statistics.len =
3872 cpu_to_be32(sizeof(struct ibmvnic_statistics));
3873
3874 /* Wait for data to be written */
3875 reinit_completion(&adapter->stats_done);
3876 rc = ibmvnic_send_crq(adapter, &crq);
3877 if (rc)
3878 return;
3879 rc = ibmvnic_wait_for_completion(adapter, &adapter->stats_done, 10000);
3880 if (rc)
3881 return;
3882
3883 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++)
3884 data[i] = be64_to_cpu(IBMVNIC_GET_STAT
3885 (adapter, ibmvnic_stats[i].offset));
3886
3887 for (j = 0; j < adapter->req_tx_queues; j++) {
3888 data[i] = adapter->tx_stats_buffers[j].batched_packets;
3889 i++;
3890 data[i] = adapter->tx_stats_buffers[j].direct_packets;
3891 i++;
3892 data[i] = adapter->tx_stats_buffers[j].bytes;
3893 i++;
3894 data[i] = adapter->tx_stats_buffers[j].dropped_packets;
3895 i++;
3896 }
3897
3898 for (j = 0; j < adapter->req_rx_queues; j++) {
3899 data[i] = adapter->rx_stats_buffers[j].packets;
3900 i++;
3901 data[i] = adapter->rx_stats_buffers[j].bytes;
3902 i++;
3903 data[i] = adapter->rx_stats_buffers[j].interrupts;
3904 i++;
3905 }
3906 }
3907
3908 static const struct ethtool_ops ibmvnic_ethtool_ops = {
3909 .get_drvinfo = ibmvnic_get_drvinfo,
3910 .get_msglevel = ibmvnic_get_msglevel,
3911 .set_msglevel = ibmvnic_set_msglevel,
3912 .get_link = ibmvnic_get_link,
3913 .get_ringparam = ibmvnic_get_ringparam,
3914 .set_ringparam = ibmvnic_set_ringparam,
3915 .get_channels = ibmvnic_get_channels,
3916 .set_channels = ibmvnic_set_channels,
3917 .get_strings = ibmvnic_get_strings,
3918 .get_sset_count = ibmvnic_get_sset_count,
3919 .get_ethtool_stats = ibmvnic_get_ethtool_stats,
3920 .get_link_ksettings = ibmvnic_get_link_ksettings,
3921 };
3922
3923 /* Routines for managing CRQs/sCRQs */
3924
reset_one_sub_crq_queue(struct ibmvnic_adapter * adapter,struct ibmvnic_sub_crq_queue * scrq)3925 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter,
3926 struct ibmvnic_sub_crq_queue *scrq)
3927 {
3928 int rc;
3929
3930 if (!scrq) {
3931 netdev_dbg(adapter->netdev, "Invalid scrq reset.\n");
3932 return -EINVAL;
3933 }
3934
3935 if (scrq->irq) {
3936 free_irq(scrq->irq, scrq);
3937 irq_dispose_mapping(scrq->irq);
3938 scrq->irq = 0;
3939 }
3940
3941 if (scrq->msgs) {
3942 memset(scrq->msgs, 0, 4 * PAGE_SIZE);
3943 atomic_set(&scrq->used, 0);
3944 scrq->cur = 0;
3945 scrq->ind_buf.index = 0;
3946 } else {
3947 netdev_dbg(adapter->netdev, "Invalid scrq reset\n");
3948 return -EINVAL;
3949 }
3950
3951 rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
3952 4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
3953 return rc;
3954 }
3955
reset_sub_crq_queues(struct ibmvnic_adapter * adapter)3956 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter)
3957 {
3958 int i, rc;
3959
3960 if (!adapter->tx_scrq || !adapter->rx_scrq)
3961 return -EINVAL;
3962
3963 ibmvnic_clean_affinity(adapter);
3964
3965 for (i = 0; i < adapter->req_tx_queues; i++) {
3966 netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i);
3967 rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]);
3968 if (rc)
3969 return rc;
3970 }
3971
3972 for (i = 0; i < adapter->req_rx_queues; i++) {
3973 netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i);
3974 rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]);
3975 if (rc)
3976 return rc;
3977 }
3978
3979 return rc;
3980 }
3981
release_sub_crq_queue(struct ibmvnic_adapter * adapter,struct ibmvnic_sub_crq_queue * scrq,bool do_h_free)3982 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter,
3983 struct ibmvnic_sub_crq_queue *scrq,
3984 bool do_h_free)
3985 {
3986 struct device *dev = &adapter->vdev->dev;
3987 long rc;
3988
3989 netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n");
3990
3991 if (do_h_free) {
3992 /* Close the sub-crqs */
3993 do {
3994 rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3995 adapter->vdev->unit_address,
3996 scrq->crq_num);
3997 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
3998
3999 if (rc) {
4000 netdev_err(adapter->netdev,
4001 "Failed to release sub-CRQ %16lx, rc = %ld\n",
4002 scrq->crq_num, rc);
4003 }
4004 }
4005
4006 dma_free_coherent(dev,
4007 IBMVNIC_IND_ARR_SZ,
4008 scrq->ind_buf.indir_arr,
4009 scrq->ind_buf.indir_dma);
4010
4011 dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
4012 DMA_BIDIRECTIONAL);
4013 free_pages((unsigned long)scrq->msgs, 2);
4014 free_cpumask_var(scrq->affinity_mask);
4015 kfree(scrq);
4016 }
4017
init_sub_crq_queue(struct ibmvnic_adapter * adapter)4018 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter
4019 *adapter)
4020 {
4021 struct device *dev = &adapter->vdev->dev;
4022 struct ibmvnic_sub_crq_queue *scrq;
4023 int rc;
4024
4025 scrq = kzalloc(sizeof(*scrq), GFP_KERNEL);
4026 if (!scrq)
4027 return NULL;
4028
4029 scrq->msgs =
4030 (union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2);
4031 if (!scrq->msgs) {
4032 dev_warn(dev, "Couldn't allocate crq queue messages page\n");
4033 goto zero_page_failed;
4034 }
4035 if (!zalloc_cpumask_var(&scrq->affinity_mask, GFP_KERNEL))
4036 goto cpumask_alloc_failed;
4037
4038 scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE,
4039 DMA_BIDIRECTIONAL);
4040 if (dma_mapping_error(dev, scrq->msg_token)) {
4041 dev_warn(dev, "Couldn't map crq queue messages page\n");
4042 goto map_failed;
4043 }
4044
4045 rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
4046 4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
4047
4048 if (rc == H_RESOURCE)
4049 rc = ibmvnic_reset_crq(adapter);
4050
4051 if (rc == H_CLOSED) {
4052 dev_warn(dev, "Partner adapter not ready, waiting.\n");
4053 } else if (rc) {
4054 dev_warn(dev, "Error %d registering sub-crq\n", rc);
4055 goto reg_failed;
4056 }
4057
4058 scrq->adapter = adapter;
4059 scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs);
4060 scrq->ind_buf.index = 0;
4061
4062 scrq->ind_buf.indir_arr =
4063 dma_alloc_coherent(dev,
4064 IBMVNIC_IND_ARR_SZ,
4065 &scrq->ind_buf.indir_dma,
4066 GFP_KERNEL);
4067
4068 if (!scrq->ind_buf.indir_arr)
4069 goto indir_failed;
4070
4071 spin_lock_init(&scrq->lock);
4072
4073 netdev_dbg(adapter->netdev,
4074 "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n",
4075 scrq->crq_num, scrq->hw_irq, scrq->irq);
4076
4077 return scrq;
4078
4079 indir_failed:
4080 do {
4081 rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
4082 adapter->vdev->unit_address,
4083 scrq->crq_num);
4084 } while (rc == H_BUSY || rc == H_IS_LONG_BUSY(rc));
4085 reg_failed:
4086 dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
4087 DMA_BIDIRECTIONAL);
4088 map_failed:
4089 free_cpumask_var(scrq->affinity_mask);
4090 cpumask_alloc_failed:
4091 free_pages((unsigned long)scrq->msgs, 2);
4092 zero_page_failed:
4093 kfree(scrq);
4094
4095 return NULL;
4096 }
4097
release_sub_crqs(struct ibmvnic_adapter * adapter,bool do_h_free)4098 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free)
4099 {
4100 int i;
4101
4102 ibmvnic_clean_affinity(adapter);
4103 if (adapter->tx_scrq) {
4104 for (i = 0; i < adapter->num_active_tx_scrqs; i++) {
4105 if (!adapter->tx_scrq[i])
4106 continue;
4107
4108 netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n",
4109 i);
4110 ibmvnic_tx_scrq_clean_buffer(adapter, adapter->tx_scrq[i]);
4111 if (adapter->tx_scrq[i]->irq) {
4112 free_irq(adapter->tx_scrq[i]->irq,
4113 adapter->tx_scrq[i]);
4114 irq_dispose_mapping(adapter->tx_scrq[i]->irq);
4115 adapter->tx_scrq[i]->irq = 0;
4116 }
4117
4118 release_sub_crq_queue(adapter, adapter->tx_scrq[i],
4119 do_h_free);
4120 }
4121
4122 kfree(adapter->tx_scrq);
4123 adapter->tx_scrq = NULL;
4124 adapter->num_active_tx_scrqs = 0;
4125 }
4126
4127 /* Clean any remaining outstanding SKBs
4128 * we freed the irq so we won't be hearing
4129 * from them
4130 */
4131 clean_tx_pools(adapter);
4132
4133 if (adapter->rx_scrq) {
4134 for (i = 0; i < adapter->num_active_rx_scrqs; i++) {
4135 if (!adapter->rx_scrq[i])
4136 continue;
4137
4138 netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n",
4139 i);
4140 if (adapter->rx_scrq[i]->irq) {
4141 free_irq(adapter->rx_scrq[i]->irq,
4142 adapter->rx_scrq[i]);
4143 irq_dispose_mapping(adapter->rx_scrq[i]->irq);
4144 adapter->rx_scrq[i]->irq = 0;
4145 }
4146
4147 release_sub_crq_queue(adapter, adapter->rx_scrq[i],
4148 do_h_free);
4149 }
4150
4151 kfree(adapter->rx_scrq);
4152 adapter->rx_scrq = NULL;
4153 adapter->num_active_rx_scrqs = 0;
4154 }
4155 }
4156
disable_scrq_irq(struct ibmvnic_adapter * adapter,struct ibmvnic_sub_crq_queue * scrq)4157 static int disable_scrq_irq(struct ibmvnic_adapter *adapter,
4158 struct ibmvnic_sub_crq_queue *scrq)
4159 {
4160 struct device *dev = &adapter->vdev->dev;
4161 unsigned long rc;
4162
4163 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
4164 H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
4165 if (rc)
4166 dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n",
4167 scrq->hw_irq, rc);
4168 return rc;
4169 }
4170
4171 /* We can not use the IRQ chip EOI handler because that has the
4172 * unintended effect of changing the interrupt priority.
4173 */
ibmvnic_xics_eoi(struct device * dev,struct ibmvnic_sub_crq_queue * scrq)4174 static void ibmvnic_xics_eoi(struct device *dev, struct ibmvnic_sub_crq_queue *scrq)
4175 {
4176 u64 val = 0xff000000 | scrq->hw_irq;
4177 unsigned long rc;
4178
4179 rc = plpar_hcall_norets(H_EOI, val);
4180 if (rc)
4181 dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n", val, rc);
4182 }
4183
4184 /* Due to a firmware bug, the hypervisor can send an interrupt to a
4185 * transmit or receive queue just prior to a partition migration.
4186 * Force an EOI after migration.
4187 */
ibmvnic_clear_pending_interrupt(struct device * dev,struct ibmvnic_sub_crq_queue * scrq)4188 static void ibmvnic_clear_pending_interrupt(struct device *dev,
4189 struct ibmvnic_sub_crq_queue *scrq)
4190 {
4191 if (!xive_enabled())
4192 ibmvnic_xics_eoi(dev, scrq);
4193 }
4194
enable_scrq_irq(struct ibmvnic_adapter * adapter,struct ibmvnic_sub_crq_queue * scrq)4195 static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
4196 struct ibmvnic_sub_crq_queue *scrq)
4197 {
4198 struct device *dev = &adapter->vdev->dev;
4199 unsigned long rc;
4200
4201 if (scrq->hw_irq > 0x100000000ULL) {
4202 dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq);
4203 return 1;
4204 }
4205
4206 if (test_bit(0, &adapter->resetting) &&
4207 adapter->reset_reason == VNIC_RESET_MOBILITY) {
4208 ibmvnic_clear_pending_interrupt(dev, scrq);
4209 }
4210
4211 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
4212 H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
4213 if (rc)
4214 dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n",
4215 scrq->hw_irq, rc);
4216 return rc;
4217 }
4218
ibmvnic_complete_tx(struct ibmvnic_adapter * adapter,struct ibmvnic_sub_crq_queue * scrq)4219 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter,
4220 struct ibmvnic_sub_crq_queue *scrq)
4221 {
4222 struct device *dev = &adapter->vdev->dev;
4223 struct ibmvnic_tx_pool *tx_pool;
4224 struct ibmvnic_tx_buff *txbuff;
4225 struct netdev_queue *txq;
4226 union sub_crq *next;
4227 int index;
4228 int i;
4229
4230 restart_loop:
4231 while (pending_scrq(adapter, scrq)) {
4232 unsigned int pool = scrq->pool_index;
4233 int num_entries = 0;
4234 int total_bytes = 0;
4235 int num_packets = 0;
4236
4237 next = ibmvnic_next_scrq(adapter, scrq);
4238 for (i = 0; i < next->tx_comp.num_comps; i++) {
4239 index = be32_to_cpu(next->tx_comp.correlators[i]);
4240 if (index & IBMVNIC_TSO_POOL_MASK) {
4241 tx_pool = &adapter->tso_pool[pool];
4242 index &= ~IBMVNIC_TSO_POOL_MASK;
4243 } else {
4244 tx_pool = &adapter->tx_pool[pool];
4245 }
4246
4247 txbuff = &tx_pool->tx_buff[index];
4248 num_packets++;
4249 num_entries += txbuff->num_entries;
4250 if (txbuff->skb) {
4251 total_bytes += txbuff->skb->len;
4252 if (next->tx_comp.rcs[i]) {
4253 dev_err(dev, "tx error %x\n",
4254 next->tx_comp.rcs[i]);
4255 dev_kfree_skb_irq(txbuff->skb);
4256 } else {
4257 dev_consume_skb_irq(txbuff->skb);
4258 }
4259 txbuff->skb = NULL;
4260 } else {
4261 netdev_warn(adapter->netdev,
4262 "TX completion received with NULL socket buffer\n");
4263 }
4264 tx_pool->free_map[tx_pool->producer_index] = index;
4265 tx_pool->producer_index =
4266 (tx_pool->producer_index + 1) %
4267 tx_pool->num_buffers;
4268 }
4269 /* remove tx_comp scrq*/
4270 next->tx_comp.first = 0;
4271
4272 txq = netdev_get_tx_queue(adapter->netdev, scrq->pool_index);
4273 netdev_tx_completed_queue(txq, num_packets, total_bytes);
4274
4275 if (atomic_sub_return(num_entries, &scrq->used) <=
4276 (adapter->req_tx_entries_per_subcrq / 2) &&
4277 __netif_subqueue_stopped(adapter->netdev,
4278 scrq->pool_index)) {
4279 rcu_read_lock();
4280 if (adapter->tx_queues_active) {
4281 netif_wake_subqueue(adapter->netdev,
4282 scrq->pool_index);
4283 netdev_dbg(adapter->netdev,
4284 "Started queue %d\n",
4285 scrq->pool_index);
4286 }
4287 rcu_read_unlock();
4288 }
4289 }
4290
4291 enable_scrq_irq(adapter, scrq);
4292
4293 if (pending_scrq(adapter, scrq)) {
4294 disable_scrq_irq(adapter, scrq);
4295 goto restart_loop;
4296 }
4297
4298 return 0;
4299 }
4300
ibmvnic_interrupt_tx(int irq,void * instance)4301 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance)
4302 {
4303 struct ibmvnic_sub_crq_queue *scrq = instance;
4304 struct ibmvnic_adapter *adapter = scrq->adapter;
4305
4306 disable_scrq_irq(adapter, scrq);
4307 ibmvnic_complete_tx(adapter, scrq);
4308
4309 return IRQ_HANDLED;
4310 }
4311
ibmvnic_interrupt_rx(int irq,void * instance)4312 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance)
4313 {
4314 struct ibmvnic_sub_crq_queue *scrq = instance;
4315 struct ibmvnic_adapter *adapter = scrq->adapter;
4316
4317 /* When booting a kdump kernel we can hit pending interrupts
4318 * prior to completing driver initialization.
4319 */
4320 if (unlikely(adapter->state != VNIC_OPEN))
4321 return IRQ_NONE;
4322
4323 adapter->rx_stats_buffers[scrq->scrq_num].interrupts++;
4324
4325 if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) {
4326 disable_scrq_irq(adapter, scrq);
4327 __napi_schedule(&adapter->napi[scrq->scrq_num]);
4328 }
4329
4330 return IRQ_HANDLED;
4331 }
4332
init_sub_crq_irqs(struct ibmvnic_adapter * adapter)4333 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter)
4334 {
4335 struct device *dev = &adapter->vdev->dev;
4336 struct ibmvnic_sub_crq_queue *scrq;
4337 int i = 0, j = 0;
4338 int rc = 0;
4339
4340 for (i = 0; i < adapter->req_tx_queues; i++) {
4341 netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n",
4342 i);
4343 scrq = adapter->tx_scrq[i];
4344 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
4345
4346 if (!scrq->irq) {
4347 rc = -EINVAL;
4348 dev_err(dev, "Error mapping irq\n");
4349 goto req_tx_irq_failed;
4350 }
4351
4352 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-tx%d",
4353 adapter->vdev->unit_address, i);
4354 rc = request_irq(scrq->irq, ibmvnic_interrupt_tx,
4355 0, scrq->name, scrq);
4356
4357 if (rc) {
4358 dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n",
4359 scrq->irq, rc);
4360 irq_dispose_mapping(scrq->irq);
4361 goto req_tx_irq_failed;
4362 }
4363 }
4364
4365 for (i = 0; i < adapter->req_rx_queues; i++) {
4366 netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n",
4367 i);
4368 scrq = adapter->rx_scrq[i];
4369 scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
4370 if (!scrq->irq) {
4371 rc = -EINVAL;
4372 dev_err(dev, "Error mapping irq\n");
4373 goto req_rx_irq_failed;
4374 }
4375 snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-rx%d",
4376 adapter->vdev->unit_address, i);
4377 rc = request_irq(scrq->irq, ibmvnic_interrupt_rx,
4378 0, scrq->name, scrq);
4379 if (rc) {
4380 dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n",
4381 scrq->irq, rc);
4382 irq_dispose_mapping(scrq->irq);
4383 goto req_rx_irq_failed;
4384 }
4385 }
4386
4387 cpus_read_lock();
4388 ibmvnic_set_affinity(adapter);
4389 cpus_read_unlock();
4390
4391 return rc;
4392
4393 req_rx_irq_failed:
4394 for (j = 0; j < i; j++) {
4395 free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]);
4396 irq_dispose_mapping(adapter->rx_scrq[j]->irq);
4397 }
4398 i = adapter->req_tx_queues;
4399 req_tx_irq_failed:
4400 for (j = 0; j < i; j++) {
4401 free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]);
4402 irq_dispose_mapping(adapter->tx_scrq[j]->irq);
4403 }
4404 release_sub_crqs(adapter, 1);
4405 return rc;
4406 }
4407
init_sub_crqs(struct ibmvnic_adapter * adapter)4408 static int init_sub_crqs(struct ibmvnic_adapter *adapter)
4409 {
4410 struct device *dev = &adapter->vdev->dev;
4411 struct ibmvnic_sub_crq_queue **allqueues;
4412 int registered_queues = 0;
4413 int total_queues;
4414 int more = 0;
4415 int i;
4416
4417 total_queues = adapter->req_tx_queues + adapter->req_rx_queues;
4418
4419 allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL);
4420 if (!allqueues)
4421 return -ENOMEM;
4422
4423 for (i = 0; i < total_queues; i++) {
4424 allqueues[i] = init_sub_crq_queue(adapter);
4425 if (!allqueues[i]) {
4426 dev_warn(dev, "Couldn't allocate all sub-crqs\n");
4427 break;
4428 }
4429 registered_queues++;
4430 }
4431
4432 /* Make sure we were able to register the minimum number of queues */
4433 if (registered_queues <
4434 adapter->min_tx_queues + adapter->min_rx_queues) {
4435 dev_err(dev, "Fatal: Couldn't init min number of sub-crqs\n");
4436 goto tx_failed;
4437 }
4438
4439 /* Distribute the failed allocated queues*/
4440 for (i = 0; i < total_queues - registered_queues + more ; i++) {
4441 netdev_dbg(adapter->netdev, "Reducing number of queues\n");
4442 switch (i % 3) {
4443 case 0:
4444 if (adapter->req_rx_queues > adapter->min_rx_queues)
4445 adapter->req_rx_queues--;
4446 else
4447 more++;
4448 break;
4449 case 1:
4450 if (adapter->req_tx_queues > adapter->min_tx_queues)
4451 adapter->req_tx_queues--;
4452 else
4453 more++;
4454 break;
4455 }
4456 }
4457
4458 adapter->tx_scrq = kcalloc(adapter->req_tx_queues,
4459 sizeof(*adapter->tx_scrq), GFP_KERNEL);
4460 if (!adapter->tx_scrq)
4461 goto tx_failed;
4462
4463 for (i = 0; i < adapter->req_tx_queues; i++) {
4464 adapter->tx_scrq[i] = allqueues[i];
4465 adapter->tx_scrq[i]->pool_index = i;
4466 adapter->num_active_tx_scrqs++;
4467 }
4468
4469 adapter->rx_scrq = kcalloc(adapter->req_rx_queues,
4470 sizeof(*adapter->rx_scrq), GFP_KERNEL);
4471 if (!adapter->rx_scrq)
4472 goto rx_failed;
4473
4474 for (i = 0; i < adapter->req_rx_queues; i++) {
4475 adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues];
4476 adapter->rx_scrq[i]->scrq_num = i;
4477 adapter->num_active_rx_scrqs++;
4478 }
4479
4480 kfree(allqueues);
4481 return 0;
4482
4483 rx_failed:
4484 kfree(adapter->tx_scrq);
4485 adapter->tx_scrq = NULL;
4486 tx_failed:
4487 for (i = 0; i < registered_queues; i++)
4488 release_sub_crq_queue(adapter, allqueues[i], 1);
4489 kfree(allqueues);
4490 return -ENOMEM;
4491 }
4492
send_request_cap(struct ibmvnic_adapter * adapter,int retry)4493 static void send_request_cap(struct ibmvnic_adapter *adapter, int retry)
4494 {
4495 struct device *dev = &adapter->vdev->dev;
4496 union ibmvnic_crq crq;
4497 int max_entries;
4498 int cap_reqs;
4499
4500 /* We send out 6 or 7 REQUEST_CAPABILITY CRQs below (depending on
4501 * the PROMISC flag). Initialize this count upfront. When the tasklet
4502 * receives a response to all of these, it will send the next protocol
4503 * message (QUERY_IP_OFFLOAD).
4504 */
4505 if (!(adapter->netdev->flags & IFF_PROMISC) ||
4506 adapter->promisc_supported)
4507 cap_reqs = 7;
4508 else
4509 cap_reqs = 6;
4510
4511 if (!retry) {
4512 /* Sub-CRQ entries are 32 byte long */
4513 int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4);
4514
4515 atomic_set(&adapter->running_cap_crqs, cap_reqs);
4516
4517 if (adapter->min_tx_entries_per_subcrq > entries_page ||
4518 adapter->min_rx_add_entries_per_subcrq > entries_page) {
4519 dev_err(dev, "Fatal, invalid entries per sub-crq\n");
4520 return;
4521 }
4522
4523 if (adapter->desired.mtu)
4524 adapter->req_mtu = adapter->desired.mtu;
4525 else
4526 adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN;
4527
4528 if (!adapter->desired.tx_entries)
4529 adapter->desired.tx_entries =
4530 adapter->max_tx_entries_per_subcrq;
4531 if (!adapter->desired.rx_entries)
4532 adapter->desired.rx_entries =
4533 adapter->max_rx_add_entries_per_subcrq;
4534
4535 max_entries = IBMVNIC_LTB_SET_SIZE /
4536 (adapter->req_mtu + IBMVNIC_BUFFER_HLEN);
4537
4538 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
4539 adapter->desired.tx_entries > IBMVNIC_LTB_SET_SIZE) {
4540 adapter->desired.tx_entries = max_entries;
4541 }
4542
4543 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
4544 adapter->desired.rx_entries > IBMVNIC_LTB_SET_SIZE) {
4545 adapter->desired.rx_entries = max_entries;
4546 }
4547
4548 if (adapter->desired.tx_entries)
4549 adapter->req_tx_entries_per_subcrq =
4550 adapter->desired.tx_entries;
4551 else
4552 adapter->req_tx_entries_per_subcrq =
4553 adapter->max_tx_entries_per_subcrq;
4554
4555 if (adapter->desired.rx_entries)
4556 adapter->req_rx_add_entries_per_subcrq =
4557 adapter->desired.rx_entries;
4558 else
4559 adapter->req_rx_add_entries_per_subcrq =
4560 adapter->max_rx_add_entries_per_subcrq;
4561
4562 if (adapter->desired.tx_queues)
4563 adapter->req_tx_queues =
4564 adapter->desired.tx_queues;
4565 else
4566 adapter->req_tx_queues =
4567 adapter->opt_tx_comp_sub_queues;
4568
4569 if (adapter->desired.rx_queues)
4570 adapter->req_rx_queues =
4571 adapter->desired.rx_queues;
4572 else
4573 adapter->req_rx_queues =
4574 adapter->opt_rx_comp_queues;
4575
4576 adapter->req_rx_add_queues = adapter->max_rx_add_queues;
4577 } else {
4578 atomic_add(cap_reqs, &adapter->running_cap_crqs);
4579 }
4580 memset(&crq, 0, sizeof(crq));
4581 crq.request_capability.first = IBMVNIC_CRQ_CMD;
4582 crq.request_capability.cmd = REQUEST_CAPABILITY;
4583
4584 crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES);
4585 crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues);
4586 cap_reqs--;
4587 ibmvnic_send_crq(adapter, &crq);
4588
4589 crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES);
4590 crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues);
4591 cap_reqs--;
4592 ibmvnic_send_crq(adapter, &crq);
4593
4594 crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES);
4595 crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues);
4596 cap_reqs--;
4597 ibmvnic_send_crq(adapter, &crq);
4598
4599 crq.request_capability.capability =
4600 cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ);
4601 crq.request_capability.number =
4602 cpu_to_be64(adapter->req_tx_entries_per_subcrq);
4603 cap_reqs--;
4604 ibmvnic_send_crq(adapter, &crq);
4605
4606 crq.request_capability.capability =
4607 cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ);
4608 crq.request_capability.number =
4609 cpu_to_be64(adapter->req_rx_add_entries_per_subcrq);
4610 cap_reqs--;
4611 ibmvnic_send_crq(adapter, &crq);
4612
4613 crq.request_capability.capability = cpu_to_be16(REQ_MTU);
4614 crq.request_capability.number = cpu_to_be64(adapter->req_mtu);
4615 cap_reqs--;
4616 ibmvnic_send_crq(adapter, &crq);
4617
4618 if (adapter->netdev->flags & IFF_PROMISC) {
4619 if (adapter->promisc_supported) {
4620 crq.request_capability.capability =
4621 cpu_to_be16(PROMISC_REQUESTED);
4622 crq.request_capability.number = cpu_to_be64(1);
4623 cap_reqs--;
4624 ibmvnic_send_crq(adapter, &crq);
4625 }
4626 } else {
4627 crq.request_capability.capability =
4628 cpu_to_be16(PROMISC_REQUESTED);
4629 crq.request_capability.number = cpu_to_be64(0);
4630 cap_reqs--;
4631 ibmvnic_send_crq(adapter, &crq);
4632 }
4633
4634 /* Keep at end to catch any discrepancy between expected and actual
4635 * CRQs sent.
4636 */
4637 WARN_ON(cap_reqs != 0);
4638 }
4639
pending_scrq(struct ibmvnic_adapter * adapter,struct ibmvnic_sub_crq_queue * scrq)4640 static int pending_scrq(struct ibmvnic_adapter *adapter,
4641 struct ibmvnic_sub_crq_queue *scrq)
4642 {
4643 union sub_crq *entry = &scrq->msgs[scrq->cur];
4644 int rc;
4645
4646 rc = !!(entry->generic.first & IBMVNIC_CRQ_CMD_RSP);
4647
4648 /* Ensure that the SCRQ valid flag is loaded prior to loading the
4649 * contents of the SCRQ descriptor
4650 */
4651 dma_rmb();
4652
4653 return rc;
4654 }
4655
ibmvnic_next_scrq(struct ibmvnic_adapter * adapter,struct ibmvnic_sub_crq_queue * scrq)4656 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter,
4657 struct ibmvnic_sub_crq_queue *scrq)
4658 {
4659 union sub_crq *entry;
4660 unsigned long flags;
4661
4662 spin_lock_irqsave(&scrq->lock, flags);
4663 entry = &scrq->msgs[scrq->cur];
4664 if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) {
4665 if (++scrq->cur == scrq->size)
4666 scrq->cur = 0;
4667 } else {
4668 entry = NULL;
4669 }
4670 spin_unlock_irqrestore(&scrq->lock, flags);
4671
4672 /* Ensure that the SCRQ valid flag is loaded prior to loading the
4673 * contents of the SCRQ descriptor
4674 */
4675 dma_rmb();
4676
4677 return entry;
4678 }
4679
ibmvnic_next_crq(struct ibmvnic_adapter * adapter)4680 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter)
4681 {
4682 struct ibmvnic_crq_queue *queue = &adapter->crq;
4683 union ibmvnic_crq *crq;
4684
4685 crq = &queue->msgs[queue->cur];
4686 if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) {
4687 if (++queue->cur == queue->size)
4688 queue->cur = 0;
4689 } else {
4690 crq = NULL;
4691 }
4692
4693 return crq;
4694 }
4695
print_subcrq_error(struct device * dev,int rc,const char * func)4696 static void print_subcrq_error(struct device *dev, int rc, const char *func)
4697 {
4698 switch (rc) {
4699 case H_PARAMETER:
4700 dev_warn_ratelimited(dev,
4701 "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n",
4702 func, rc);
4703 break;
4704 case H_CLOSED:
4705 dev_warn_ratelimited(dev,
4706 "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n",
4707 func, rc);
4708 break;
4709 default:
4710 dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc);
4711 break;
4712 }
4713 }
4714
send_subcrq_indirect(struct ibmvnic_adapter * adapter,u64 remote_handle,u64 ioba,u64 num_entries)4715 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter,
4716 u64 remote_handle, u64 ioba, u64 num_entries)
4717 {
4718 unsigned int ua = adapter->vdev->unit_address;
4719 struct device *dev = &adapter->vdev->dev;
4720 int rc;
4721
4722 /* Make sure the hypervisor sees the complete request */
4723 dma_wmb();
4724 rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua,
4725 cpu_to_be64(remote_handle),
4726 ioba, num_entries);
4727
4728 if (rc)
4729 print_subcrq_error(dev, rc, __func__);
4730
4731 return rc;
4732 }
4733
ibmvnic_send_crq(struct ibmvnic_adapter * adapter,union ibmvnic_crq * crq)4734 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter,
4735 union ibmvnic_crq *crq)
4736 {
4737 unsigned int ua = adapter->vdev->unit_address;
4738 struct device *dev = &adapter->vdev->dev;
4739 u64 *u64_crq = (u64 *)crq;
4740 int rc;
4741
4742 netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n",
4743 (unsigned long)cpu_to_be64(u64_crq[0]),
4744 (unsigned long)cpu_to_be64(u64_crq[1]));
4745
4746 if (!adapter->crq.active &&
4747 crq->generic.first != IBMVNIC_CRQ_INIT_CMD) {
4748 dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n");
4749 return -EINVAL;
4750 }
4751
4752 /* Make sure the hypervisor sees the complete request */
4753 dma_wmb();
4754
4755 rc = plpar_hcall_norets(H_SEND_CRQ, ua,
4756 cpu_to_be64(u64_crq[0]),
4757 cpu_to_be64(u64_crq[1]));
4758
4759 if (rc) {
4760 if (rc == H_CLOSED) {
4761 dev_warn(dev, "CRQ Queue closed\n");
4762 /* do not reset, report the fail, wait for passive init from server */
4763 }
4764
4765 dev_warn(dev, "Send error (rc=%d)\n", rc);
4766 }
4767
4768 return rc;
4769 }
4770
ibmvnic_send_crq_init(struct ibmvnic_adapter * adapter)4771 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter)
4772 {
4773 struct device *dev = &adapter->vdev->dev;
4774 union ibmvnic_crq crq;
4775 int retries = 100;
4776 int rc;
4777
4778 memset(&crq, 0, sizeof(crq));
4779 crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
4780 crq.generic.cmd = IBMVNIC_CRQ_INIT;
4781 netdev_dbg(adapter->netdev, "Sending CRQ init\n");
4782
4783 do {
4784 rc = ibmvnic_send_crq(adapter, &crq);
4785 if (rc != H_CLOSED)
4786 break;
4787 retries--;
4788 msleep(50);
4789
4790 } while (retries > 0);
4791
4792 if (rc) {
4793 dev_err(dev, "Failed to send init request, rc = %d\n", rc);
4794 return rc;
4795 }
4796
4797 return 0;
4798 }
4799
4800 struct vnic_login_client_data {
4801 u8 type;
4802 __be16 len;
4803 char name[];
4804 } __packed;
4805
vnic_client_data_len(struct ibmvnic_adapter * adapter)4806 static int vnic_client_data_len(struct ibmvnic_adapter *adapter)
4807 {
4808 int len;
4809
4810 /* Calculate the amount of buffer space needed for the
4811 * vnic client data in the login buffer. There are four entries,
4812 * OS name, LPAR name, device name, and a null last entry.
4813 */
4814 len = 4 * sizeof(struct vnic_login_client_data);
4815 len += 6; /* "Linux" plus NULL */
4816 len += strlen(utsname()->nodename) + 1;
4817 len += strlen(adapter->netdev->name) + 1;
4818
4819 return len;
4820 }
4821
vnic_add_client_data(struct ibmvnic_adapter * adapter,struct vnic_login_client_data * vlcd)4822 static void vnic_add_client_data(struct ibmvnic_adapter *adapter,
4823 struct vnic_login_client_data *vlcd)
4824 {
4825 const char *os_name = "Linux";
4826 int len;
4827
4828 /* Type 1 - LPAR OS */
4829 vlcd->type = 1;
4830 len = strlen(os_name) + 1;
4831 vlcd->len = cpu_to_be16(len);
4832 strscpy(vlcd->name, os_name, len);
4833 vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
4834
4835 /* Type 2 - LPAR name */
4836 vlcd->type = 2;
4837 len = strlen(utsname()->nodename) + 1;
4838 vlcd->len = cpu_to_be16(len);
4839 strscpy(vlcd->name, utsname()->nodename, len);
4840 vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
4841
4842 /* Type 3 - device name */
4843 vlcd->type = 3;
4844 len = strlen(adapter->netdev->name) + 1;
4845 vlcd->len = cpu_to_be16(len);
4846 strscpy(vlcd->name, adapter->netdev->name, len);
4847 }
4848
send_login(struct ibmvnic_adapter * adapter)4849 static int send_login(struct ibmvnic_adapter *adapter)
4850 {
4851 struct ibmvnic_login_rsp_buffer *login_rsp_buffer;
4852 struct ibmvnic_login_buffer *login_buffer;
4853 struct device *dev = &adapter->vdev->dev;
4854 struct vnic_login_client_data *vlcd;
4855 dma_addr_t rsp_buffer_token;
4856 dma_addr_t buffer_token;
4857 size_t rsp_buffer_size;
4858 union ibmvnic_crq crq;
4859 int client_data_len;
4860 size_t buffer_size;
4861 __be64 *tx_list_p;
4862 __be64 *rx_list_p;
4863 int rc;
4864 int i;
4865
4866 if (!adapter->tx_scrq || !adapter->rx_scrq) {
4867 netdev_err(adapter->netdev,
4868 "RX or TX queues are not allocated, device login failed\n");
4869 return -ENOMEM;
4870 }
4871
4872 release_login_buffer(adapter);
4873 release_login_rsp_buffer(adapter);
4874
4875 client_data_len = vnic_client_data_len(adapter);
4876
4877 buffer_size =
4878 sizeof(struct ibmvnic_login_buffer) +
4879 sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) +
4880 client_data_len;
4881
4882 login_buffer = kzalloc(buffer_size, GFP_ATOMIC);
4883 if (!login_buffer)
4884 goto buf_alloc_failed;
4885
4886 buffer_token = dma_map_single(dev, login_buffer, buffer_size,
4887 DMA_TO_DEVICE);
4888 if (dma_mapping_error(dev, buffer_token)) {
4889 dev_err(dev, "Couldn't map login buffer\n");
4890 goto buf_map_failed;
4891 }
4892
4893 rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) +
4894 sizeof(u64) * adapter->req_tx_queues +
4895 sizeof(u64) * adapter->req_rx_queues +
4896 sizeof(u64) * adapter->req_rx_queues +
4897 sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS;
4898
4899 login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC);
4900 if (!login_rsp_buffer)
4901 goto buf_rsp_alloc_failed;
4902
4903 rsp_buffer_token = dma_map_single(dev, login_rsp_buffer,
4904 rsp_buffer_size, DMA_FROM_DEVICE);
4905 if (dma_mapping_error(dev, rsp_buffer_token)) {
4906 dev_err(dev, "Couldn't map login rsp buffer\n");
4907 goto buf_rsp_map_failed;
4908 }
4909
4910 adapter->login_buf = login_buffer;
4911 adapter->login_buf_token = buffer_token;
4912 adapter->login_buf_sz = buffer_size;
4913 adapter->login_rsp_buf = login_rsp_buffer;
4914 adapter->login_rsp_buf_token = rsp_buffer_token;
4915 adapter->login_rsp_buf_sz = rsp_buffer_size;
4916
4917 login_buffer->len = cpu_to_be32(buffer_size);
4918 login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB);
4919 login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues);
4920 login_buffer->off_txcomp_subcrqs =
4921 cpu_to_be32(sizeof(struct ibmvnic_login_buffer));
4922 login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues);
4923 login_buffer->off_rxcomp_subcrqs =
4924 cpu_to_be32(sizeof(struct ibmvnic_login_buffer) +
4925 sizeof(u64) * adapter->req_tx_queues);
4926 login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token);
4927 login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size);
4928
4929 tx_list_p = (__be64 *)((char *)login_buffer +
4930 sizeof(struct ibmvnic_login_buffer));
4931 rx_list_p = (__be64 *)((char *)login_buffer +
4932 sizeof(struct ibmvnic_login_buffer) +
4933 sizeof(u64) * adapter->req_tx_queues);
4934
4935 for (i = 0; i < adapter->req_tx_queues; i++) {
4936 if (adapter->tx_scrq[i]) {
4937 tx_list_p[i] =
4938 cpu_to_be64(adapter->tx_scrq[i]->crq_num);
4939 }
4940 }
4941
4942 for (i = 0; i < adapter->req_rx_queues; i++) {
4943 if (adapter->rx_scrq[i]) {
4944 rx_list_p[i] =
4945 cpu_to_be64(adapter->rx_scrq[i]->crq_num);
4946 }
4947 }
4948
4949 /* Insert vNIC login client data */
4950 vlcd = (struct vnic_login_client_data *)
4951 ((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues));
4952 login_buffer->client_data_offset =
4953 cpu_to_be32((char *)vlcd - (char *)login_buffer);
4954 login_buffer->client_data_len = cpu_to_be32(client_data_len);
4955
4956 vnic_add_client_data(adapter, vlcd);
4957
4958 netdev_dbg(adapter->netdev, "Login Buffer:\n");
4959 for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) {
4960 netdev_dbg(adapter->netdev, "%016lx\n",
4961 ((unsigned long *)(adapter->login_buf))[i]);
4962 }
4963
4964 memset(&crq, 0, sizeof(crq));
4965 crq.login.first = IBMVNIC_CRQ_CMD;
4966 crq.login.cmd = LOGIN;
4967 crq.login.ioba = cpu_to_be32(buffer_token);
4968 crq.login.len = cpu_to_be32(buffer_size);
4969
4970 adapter->login_pending = true;
4971 rc = ibmvnic_send_crq(adapter, &crq);
4972 if (rc) {
4973 adapter->login_pending = false;
4974 netdev_err(adapter->netdev, "Failed to send login, rc=%d\n", rc);
4975 goto buf_send_failed;
4976 }
4977
4978 return 0;
4979
4980 buf_send_failed:
4981 dma_unmap_single(dev, rsp_buffer_token, rsp_buffer_size,
4982 DMA_FROM_DEVICE);
4983 buf_rsp_map_failed:
4984 kfree(login_rsp_buffer);
4985 adapter->login_rsp_buf = NULL;
4986 buf_rsp_alloc_failed:
4987 dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE);
4988 buf_map_failed:
4989 kfree(login_buffer);
4990 adapter->login_buf = NULL;
4991 buf_alloc_failed:
4992 return -ENOMEM;
4993 }
4994
send_request_map(struct ibmvnic_adapter * adapter,dma_addr_t addr,u32 len,u8 map_id)4995 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr,
4996 u32 len, u8 map_id)
4997 {
4998 union ibmvnic_crq crq;
4999
5000 memset(&crq, 0, sizeof(crq));
5001 crq.request_map.first = IBMVNIC_CRQ_CMD;
5002 crq.request_map.cmd = REQUEST_MAP;
5003 crq.request_map.map_id = map_id;
5004 crq.request_map.ioba = cpu_to_be32(addr);
5005 crq.request_map.len = cpu_to_be32(len);
5006 return ibmvnic_send_crq(adapter, &crq);
5007 }
5008
send_request_unmap(struct ibmvnic_adapter * adapter,u8 map_id)5009 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id)
5010 {
5011 union ibmvnic_crq crq;
5012
5013 memset(&crq, 0, sizeof(crq));
5014 crq.request_unmap.first = IBMVNIC_CRQ_CMD;
5015 crq.request_unmap.cmd = REQUEST_UNMAP;
5016 crq.request_unmap.map_id = map_id;
5017 return ibmvnic_send_crq(adapter, &crq);
5018 }
5019
send_query_map(struct ibmvnic_adapter * adapter)5020 static void send_query_map(struct ibmvnic_adapter *adapter)
5021 {
5022 union ibmvnic_crq crq;
5023
5024 memset(&crq, 0, sizeof(crq));
5025 crq.query_map.first = IBMVNIC_CRQ_CMD;
5026 crq.query_map.cmd = QUERY_MAP;
5027 ibmvnic_send_crq(adapter, &crq);
5028 }
5029
5030 /* Send a series of CRQs requesting various capabilities of the VNIC server */
send_query_cap(struct ibmvnic_adapter * adapter)5031 static void send_query_cap(struct ibmvnic_adapter *adapter)
5032 {
5033 union ibmvnic_crq crq;
5034 int cap_reqs;
5035
5036 /* We send out 25 QUERY_CAPABILITY CRQs below. Initialize this count
5037 * upfront. When the tasklet receives a response to all of these, it
5038 * can send out the next protocol messaage (REQUEST_CAPABILITY).
5039 */
5040 cap_reqs = 25;
5041
5042 atomic_set(&adapter->running_cap_crqs, cap_reqs);
5043
5044 memset(&crq, 0, sizeof(crq));
5045 crq.query_capability.first = IBMVNIC_CRQ_CMD;
5046 crq.query_capability.cmd = QUERY_CAPABILITY;
5047
5048 crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES);
5049 ibmvnic_send_crq(adapter, &crq);
5050 cap_reqs--;
5051
5052 crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES);
5053 ibmvnic_send_crq(adapter, &crq);
5054 cap_reqs--;
5055
5056 crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES);
5057 ibmvnic_send_crq(adapter, &crq);
5058 cap_reqs--;
5059
5060 crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES);
5061 ibmvnic_send_crq(adapter, &crq);
5062 cap_reqs--;
5063
5064 crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES);
5065 ibmvnic_send_crq(adapter, &crq);
5066 cap_reqs--;
5067
5068 crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES);
5069 ibmvnic_send_crq(adapter, &crq);
5070 cap_reqs--;
5071
5072 crq.query_capability.capability =
5073 cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ);
5074 ibmvnic_send_crq(adapter, &crq);
5075 cap_reqs--;
5076
5077 crq.query_capability.capability =
5078 cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ);
5079 ibmvnic_send_crq(adapter, &crq);
5080 cap_reqs--;
5081
5082 crq.query_capability.capability =
5083 cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ);
5084 ibmvnic_send_crq(adapter, &crq);
5085 cap_reqs--;
5086
5087 crq.query_capability.capability =
5088 cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ);
5089 ibmvnic_send_crq(adapter, &crq);
5090 cap_reqs--;
5091
5092 crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD);
5093 ibmvnic_send_crq(adapter, &crq);
5094 cap_reqs--;
5095
5096 crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED);
5097 ibmvnic_send_crq(adapter, &crq);
5098 cap_reqs--;
5099
5100 crq.query_capability.capability = cpu_to_be16(MIN_MTU);
5101 ibmvnic_send_crq(adapter, &crq);
5102 cap_reqs--;
5103
5104 crq.query_capability.capability = cpu_to_be16(MAX_MTU);
5105 ibmvnic_send_crq(adapter, &crq);
5106 cap_reqs--;
5107
5108 crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS);
5109 ibmvnic_send_crq(adapter, &crq);
5110 cap_reqs--;
5111
5112 crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION);
5113 ibmvnic_send_crq(adapter, &crq);
5114 cap_reqs--;
5115
5116 crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION);
5117 ibmvnic_send_crq(adapter, &crq);
5118 cap_reqs--;
5119
5120 crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES);
5121 ibmvnic_send_crq(adapter, &crq);
5122 cap_reqs--;
5123
5124 crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED);
5125 ibmvnic_send_crq(adapter, &crq);
5126 cap_reqs--;
5127
5128 crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES);
5129 ibmvnic_send_crq(adapter, &crq);
5130 cap_reqs--;
5131
5132 crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES);
5133 ibmvnic_send_crq(adapter, &crq);
5134 cap_reqs--;
5135
5136 crq.query_capability.capability =
5137 cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q);
5138 ibmvnic_send_crq(adapter, &crq);
5139 cap_reqs--;
5140
5141 crq.query_capability.capability =
5142 cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ);
5143 ibmvnic_send_crq(adapter, &crq);
5144 cap_reqs--;
5145
5146 crq.query_capability.capability =
5147 cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ);
5148 ibmvnic_send_crq(adapter, &crq);
5149 cap_reqs--;
5150
5151 crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ);
5152
5153 ibmvnic_send_crq(adapter, &crq);
5154 cap_reqs--;
5155
5156 /* Keep at end to catch any discrepancy between expected and actual
5157 * CRQs sent.
5158 */
5159 WARN_ON(cap_reqs != 0);
5160 }
5161
send_query_ip_offload(struct ibmvnic_adapter * adapter)5162 static void send_query_ip_offload(struct ibmvnic_adapter *adapter)
5163 {
5164 int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer);
5165 struct device *dev = &adapter->vdev->dev;
5166 union ibmvnic_crq crq;
5167
5168 adapter->ip_offload_tok =
5169 dma_map_single(dev,
5170 &adapter->ip_offload_buf,
5171 buf_sz,
5172 DMA_FROM_DEVICE);
5173
5174 if (dma_mapping_error(dev, adapter->ip_offload_tok)) {
5175 if (!firmware_has_feature(FW_FEATURE_CMO))
5176 dev_err(dev, "Couldn't map offload buffer\n");
5177 return;
5178 }
5179
5180 memset(&crq, 0, sizeof(crq));
5181 crq.query_ip_offload.first = IBMVNIC_CRQ_CMD;
5182 crq.query_ip_offload.cmd = QUERY_IP_OFFLOAD;
5183 crq.query_ip_offload.len = cpu_to_be32(buf_sz);
5184 crq.query_ip_offload.ioba =
5185 cpu_to_be32(adapter->ip_offload_tok);
5186
5187 ibmvnic_send_crq(adapter, &crq);
5188 }
5189
send_control_ip_offload(struct ibmvnic_adapter * adapter)5190 static void send_control_ip_offload(struct ibmvnic_adapter *adapter)
5191 {
5192 struct ibmvnic_control_ip_offload_buffer *ctrl_buf = &adapter->ip_offload_ctrl;
5193 struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
5194 struct device *dev = &adapter->vdev->dev;
5195 netdev_features_t old_hw_features = 0;
5196 union ibmvnic_crq crq;
5197
5198 adapter->ip_offload_ctrl_tok =
5199 dma_map_single(dev,
5200 ctrl_buf,
5201 sizeof(adapter->ip_offload_ctrl),
5202 DMA_TO_DEVICE);
5203
5204 if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) {
5205 dev_err(dev, "Couldn't map ip offload control buffer\n");
5206 return;
5207 }
5208
5209 ctrl_buf->len = cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
5210 ctrl_buf->version = cpu_to_be32(INITIAL_VERSION_IOB);
5211 ctrl_buf->ipv4_chksum = buf->ipv4_chksum;
5212 ctrl_buf->ipv6_chksum = buf->ipv6_chksum;
5213 ctrl_buf->tcp_ipv4_chksum = buf->tcp_ipv4_chksum;
5214 ctrl_buf->udp_ipv4_chksum = buf->udp_ipv4_chksum;
5215 ctrl_buf->tcp_ipv6_chksum = buf->tcp_ipv6_chksum;
5216 ctrl_buf->udp_ipv6_chksum = buf->udp_ipv6_chksum;
5217 ctrl_buf->large_tx_ipv4 = buf->large_tx_ipv4;
5218 ctrl_buf->large_tx_ipv6 = buf->large_tx_ipv6;
5219
5220 /* large_rx disabled for now, additional features needed */
5221 ctrl_buf->large_rx_ipv4 = 0;
5222 ctrl_buf->large_rx_ipv6 = 0;
5223
5224 if (adapter->state != VNIC_PROBING) {
5225 old_hw_features = adapter->netdev->hw_features;
5226 adapter->netdev->hw_features = 0;
5227 }
5228
5229 adapter->netdev->hw_features = NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
5230
5231 if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum)
5232 adapter->netdev->hw_features |= NETIF_F_IP_CSUM;
5233
5234 if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum)
5235 adapter->netdev->hw_features |= NETIF_F_IPV6_CSUM;
5236
5237 if ((adapter->netdev->features &
5238 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
5239 adapter->netdev->hw_features |= NETIF_F_RXCSUM;
5240
5241 if (buf->large_tx_ipv4)
5242 adapter->netdev->hw_features |= NETIF_F_TSO;
5243 if (buf->large_tx_ipv6)
5244 adapter->netdev->hw_features |= NETIF_F_TSO6;
5245
5246 if (adapter->state == VNIC_PROBING) {
5247 adapter->netdev->features |= adapter->netdev->hw_features;
5248 } else if (old_hw_features != adapter->netdev->hw_features) {
5249 netdev_features_t tmp = 0;
5250
5251 /* disable features no longer supported */
5252 adapter->netdev->features &= adapter->netdev->hw_features;
5253 /* turn on features now supported if previously enabled */
5254 tmp = (old_hw_features ^ adapter->netdev->hw_features) &
5255 adapter->netdev->hw_features;
5256 adapter->netdev->features |=
5257 tmp & adapter->netdev->wanted_features;
5258 }
5259
5260 memset(&crq, 0, sizeof(crq));
5261 crq.control_ip_offload.first = IBMVNIC_CRQ_CMD;
5262 crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD;
5263 crq.control_ip_offload.len =
5264 cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
5265 crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok);
5266 ibmvnic_send_crq(adapter, &crq);
5267 }
5268
handle_vpd_size_rsp(union ibmvnic_crq * crq,struct ibmvnic_adapter * adapter)5269 static void handle_vpd_size_rsp(union ibmvnic_crq *crq,
5270 struct ibmvnic_adapter *adapter)
5271 {
5272 struct device *dev = &adapter->vdev->dev;
5273
5274 if (crq->get_vpd_size_rsp.rc.code) {
5275 dev_err(dev, "Error retrieving VPD size, rc=%x\n",
5276 crq->get_vpd_size_rsp.rc.code);
5277 complete(&adapter->fw_done);
5278 return;
5279 }
5280
5281 adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len);
5282 complete(&adapter->fw_done);
5283 }
5284
handle_vpd_rsp(union ibmvnic_crq * crq,struct ibmvnic_adapter * adapter)5285 static void handle_vpd_rsp(union ibmvnic_crq *crq,
5286 struct ibmvnic_adapter *adapter)
5287 {
5288 struct device *dev = &adapter->vdev->dev;
5289 unsigned char *substr = NULL;
5290 u8 fw_level_len = 0;
5291
5292 memset(adapter->fw_version, 0, 32);
5293
5294 dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len,
5295 DMA_FROM_DEVICE);
5296
5297 if (crq->get_vpd_rsp.rc.code) {
5298 dev_err(dev, "Error retrieving VPD from device, rc=%x\n",
5299 crq->get_vpd_rsp.rc.code);
5300 goto complete;
5301 }
5302
5303 /* get the position of the firmware version info
5304 * located after the ASCII 'RM' substring in the buffer
5305 */
5306 substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len);
5307 if (!substr) {
5308 dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n");
5309 goto complete;
5310 }
5311
5312 /* get length of firmware level ASCII substring */
5313 if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) {
5314 fw_level_len = *(substr + 2);
5315 } else {
5316 dev_info(dev, "Length of FW substr extrapolated VDP buff\n");
5317 goto complete;
5318 }
5319
5320 /* copy firmware version string from vpd into adapter */
5321 if ((substr + 3 + fw_level_len) <
5322 (adapter->vpd->buff + adapter->vpd->len)) {
5323 strncpy((char *)adapter->fw_version, substr + 3, fw_level_len);
5324 } else {
5325 dev_info(dev, "FW substr extrapolated VPD buff\n");
5326 }
5327
5328 complete:
5329 if (adapter->fw_version[0] == '\0')
5330 strscpy((char *)adapter->fw_version, "N/A", sizeof(adapter->fw_version));
5331 complete(&adapter->fw_done);
5332 }
5333
handle_query_ip_offload_rsp(struct ibmvnic_adapter * adapter)5334 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter)
5335 {
5336 struct device *dev = &adapter->vdev->dev;
5337 struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
5338 int i;
5339
5340 dma_unmap_single(dev, adapter->ip_offload_tok,
5341 sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE);
5342
5343 netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n");
5344 for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++)
5345 netdev_dbg(adapter->netdev, "%016lx\n",
5346 ((unsigned long *)(buf))[i]);
5347
5348 netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum);
5349 netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum);
5350 netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n",
5351 buf->tcp_ipv4_chksum);
5352 netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n",
5353 buf->tcp_ipv6_chksum);
5354 netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n",
5355 buf->udp_ipv4_chksum);
5356 netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n",
5357 buf->udp_ipv6_chksum);
5358 netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n",
5359 buf->large_tx_ipv4);
5360 netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n",
5361 buf->large_tx_ipv6);
5362 netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n",
5363 buf->large_rx_ipv4);
5364 netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n",
5365 buf->large_rx_ipv6);
5366 netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n",
5367 buf->max_ipv4_header_size);
5368 netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n",
5369 buf->max_ipv6_header_size);
5370 netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n",
5371 buf->max_tcp_header_size);
5372 netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n",
5373 buf->max_udp_header_size);
5374 netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n",
5375 buf->max_large_tx_size);
5376 netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n",
5377 buf->max_large_rx_size);
5378 netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n",
5379 buf->ipv6_extension_header);
5380 netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n",
5381 buf->tcp_pseudosum_req);
5382 netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n",
5383 buf->num_ipv6_ext_headers);
5384 netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n",
5385 buf->off_ipv6_ext_headers);
5386
5387 send_control_ip_offload(adapter);
5388 }
5389
ibmvnic_fw_err_cause(u16 cause)5390 static const char *ibmvnic_fw_err_cause(u16 cause)
5391 {
5392 switch (cause) {
5393 case ADAPTER_PROBLEM:
5394 return "adapter problem";
5395 case BUS_PROBLEM:
5396 return "bus problem";
5397 case FW_PROBLEM:
5398 return "firmware problem";
5399 case DD_PROBLEM:
5400 return "device driver problem";
5401 case EEH_RECOVERY:
5402 return "EEH recovery";
5403 case FW_UPDATED:
5404 return "firmware updated";
5405 case LOW_MEMORY:
5406 return "low Memory";
5407 default:
5408 return "unknown";
5409 }
5410 }
5411
handle_error_indication(union ibmvnic_crq * crq,struct ibmvnic_adapter * adapter)5412 static void handle_error_indication(union ibmvnic_crq *crq,
5413 struct ibmvnic_adapter *adapter)
5414 {
5415 struct device *dev = &adapter->vdev->dev;
5416 u16 cause;
5417
5418 cause = be16_to_cpu(crq->error_indication.error_cause);
5419
5420 dev_warn_ratelimited(dev,
5421 "Firmware reports %serror, cause: %s. Starting recovery...\n",
5422 crq->error_indication.flags
5423 & IBMVNIC_FATAL_ERROR ? "FATAL " : "",
5424 ibmvnic_fw_err_cause(cause));
5425
5426 if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR)
5427 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
5428 else
5429 ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL);
5430 }
5431
handle_change_mac_rsp(union ibmvnic_crq * crq,struct ibmvnic_adapter * adapter)5432 static int handle_change_mac_rsp(union ibmvnic_crq *crq,
5433 struct ibmvnic_adapter *adapter)
5434 {
5435 struct net_device *netdev = adapter->netdev;
5436 struct device *dev = &adapter->vdev->dev;
5437 long rc;
5438
5439 rc = crq->change_mac_addr_rsp.rc.code;
5440 if (rc) {
5441 dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc);
5442 goto out;
5443 }
5444 /* crq->change_mac_addr.mac_addr is the requested one
5445 * crq->change_mac_addr_rsp.mac_addr is the returned valid one.
5446 */
5447 eth_hw_addr_set(netdev, &crq->change_mac_addr_rsp.mac_addr[0]);
5448 ether_addr_copy(adapter->mac_addr,
5449 &crq->change_mac_addr_rsp.mac_addr[0]);
5450 out:
5451 complete(&adapter->fw_done);
5452 return rc;
5453 }
5454
handle_request_cap_rsp(union ibmvnic_crq * crq,struct ibmvnic_adapter * adapter)5455 static void handle_request_cap_rsp(union ibmvnic_crq *crq,
5456 struct ibmvnic_adapter *adapter)
5457 {
5458 struct device *dev = &adapter->vdev->dev;
5459 u64 *req_value;
5460 char *name;
5461
5462 atomic_dec(&adapter->running_cap_crqs);
5463 netdev_dbg(adapter->netdev, "Outstanding request-caps: %d\n",
5464 atomic_read(&adapter->running_cap_crqs));
5465 switch (be16_to_cpu(crq->request_capability_rsp.capability)) {
5466 case REQ_TX_QUEUES:
5467 req_value = &adapter->req_tx_queues;
5468 name = "tx";
5469 break;
5470 case REQ_RX_QUEUES:
5471 req_value = &adapter->req_rx_queues;
5472 name = "rx";
5473 break;
5474 case REQ_RX_ADD_QUEUES:
5475 req_value = &adapter->req_rx_add_queues;
5476 name = "rx_add";
5477 break;
5478 case REQ_TX_ENTRIES_PER_SUBCRQ:
5479 req_value = &adapter->req_tx_entries_per_subcrq;
5480 name = "tx_entries_per_subcrq";
5481 break;
5482 case REQ_RX_ADD_ENTRIES_PER_SUBCRQ:
5483 req_value = &adapter->req_rx_add_entries_per_subcrq;
5484 name = "rx_add_entries_per_subcrq";
5485 break;
5486 case REQ_MTU:
5487 req_value = &adapter->req_mtu;
5488 name = "mtu";
5489 break;
5490 case PROMISC_REQUESTED:
5491 req_value = &adapter->promisc;
5492 name = "promisc";
5493 break;
5494 default:
5495 dev_err(dev, "Got invalid cap request rsp %d\n",
5496 crq->request_capability.capability);
5497 return;
5498 }
5499
5500 switch (crq->request_capability_rsp.rc.code) {
5501 case SUCCESS:
5502 break;
5503 case PARTIALSUCCESS:
5504 dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n",
5505 *req_value,
5506 (long)be64_to_cpu(crq->request_capability_rsp.number),
5507 name);
5508
5509 if (be16_to_cpu(crq->request_capability_rsp.capability) ==
5510 REQ_MTU) {
5511 pr_err("mtu of %llu is not supported. Reverting.\n",
5512 *req_value);
5513 *req_value = adapter->fallback.mtu;
5514 } else {
5515 *req_value =
5516 be64_to_cpu(crq->request_capability_rsp.number);
5517 }
5518
5519 send_request_cap(adapter, 1);
5520 return;
5521 default:
5522 dev_err(dev, "Error %d in request cap rsp\n",
5523 crq->request_capability_rsp.rc.code);
5524 return;
5525 }
5526
5527 /* Done receiving requested capabilities, query IP offload support */
5528 if (atomic_read(&adapter->running_cap_crqs) == 0)
5529 send_query_ip_offload(adapter);
5530 }
5531
handle_login_rsp(union ibmvnic_crq * login_rsp_crq,struct ibmvnic_adapter * adapter)5532 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq,
5533 struct ibmvnic_adapter *adapter)
5534 {
5535 struct device *dev = &adapter->vdev->dev;
5536 struct net_device *netdev = adapter->netdev;
5537 struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf;
5538 struct ibmvnic_login_buffer *login = adapter->login_buf;
5539 u64 *tx_handle_array;
5540 u64 *rx_handle_array;
5541 int num_tx_pools;
5542 int num_rx_pools;
5543 u64 *size_array;
5544 u32 rsp_len;
5545 int i;
5546
5547 /* CHECK: Test/set of login_pending does not need to be atomic
5548 * because only ibmvnic_tasklet tests/clears this.
5549 */
5550 if (!adapter->login_pending) {
5551 netdev_warn(netdev, "Ignoring unexpected login response\n");
5552 return 0;
5553 }
5554 adapter->login_pending = false;
5555
5556 /* If the number of queues requested can't be allocated by the
5557 * server, the login response will return with code 1. We will need
5558 * to resend the login buffer with fewer queues requested.
5559 */
5560 if (login_rsp_crq->generic.rc.code) {
5561 adapter->init_done_rc = login_rsp_crq->generic.rc.code;
5562 complete(&adapter->init_done);
5563 return 0;
5564 }
5565
5566 if (adapter->failover_pending) {
5567 adapter->init_done_rc = -EAGAIN;
5568 netdev_dbg(netdev, "Failover pending, ignoring login response\n");
5569 complete(&adapter->init_done);
5570 /* login response buffer will be released on reset */
5571 return 0;
5572 }
5573
5574 netdev->mtu = adapter->req_mtu - ETH_HLEN;
5575
5576 netdev_dbg(adapter->netdev, "Login Response Buffer:\n");
5577 for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) {
5578 netdev_dbg(adapter->netdev, "%016lx\n",
5579 ((unsigned long *)(adapter->login_rsp_buf))[i]);
5580 }
5581
5582 /* Sanity checks */
5583 if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs ||
5584 (be32_to_cpu(login->num_rxcomp_subcrqs) *
5585 adapter->req_rx_add_queues !=
5586 be32_to_cpu(login_rsp->num_rxadd_subcrqs))) {
5587 dev_err(dev, "FATAL: Inconsistent login and login rsp\n");
5588 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
5589 return -EIO;
5590 }
5591
5592 rsp_len = be32_to_cpu(login_rsp->len);
5593 if (be32_to_cpu(login->login_rsp_len) < rsp_len ||
5594 rsp_len <= be32_to_cpu(login_rsp->off_txsubm_subcrqs) ||
5595 rsp_len <= be32_to_cpu(login_rsp->off_rxadd_subcrqs) ||
5596 rsp_len <= be32_to_cpu(login_rsp->off_rxadd_buff_size) ||
5597 rsp_len <= be32_to_cpu(login_rsp->off_supp_tx_desc)) {
5598 /* This can happen if a login request times out and there are
5599 * 2 outstanding login requests sent, the LOGIN_RSP crq
5600 * could have been for the older login request. So we are
5601 * parsing the newer response buffer which may be incomplete
5602 */
5603 dev_err(dev, "FATAL: Login rsp offsets/lengths invalid\n");
5604 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
5605 return -EIO;
5606 }
5607
5608 size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
5609 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size));
5610 /* variable buffer sizes are not supported, so just read the
5611 * first entry.
5612 */
5613 adapter->cur_rx_buf_sz = be64_to_cpu(size_array[0]);
5614
5615 num_tx_pools = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs);
5616 num_rx_pools = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
5617
5618 tx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
5619 be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs));
5620 rx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
5621 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_subcrqs));
5622
5623 for (i = 0; i < num_tx_pools; i++)
5624 adapter->tx_scrq[i]->handle = tx_handle_array[i];
5625
5626 for (i = 0; i < num_rx_pools; i++)
5627 adapter->rx_scrq[i]->handle = rx_handle_array[i];
5628
5629 adapter->num_active_tx_scrqs = num_tx_pools;
5630 adapter->num_active_rx_scrqs = num_rx_pools;
5631 release_login_rsp_buffer(adapter);
5632 release_login_buffer(adapter);
5633 complete(&adapter->init_done);
5634
5635 return 0;
5636 }
5637
handle_request_unmap_rsp(union ibmvnic_crq * crq,struct ibmvnic_adapter * adapter)5638 static void handle_request_unmap_rsp(union ibmvnic_crq *crq,
5639 struct ibmvnic_adapter *adapter)
5640 {
5641 struct device *dev = &adapter->vdev->dev;
5642 long rc;
5643
5644 rc = crq->request_unmap_rsp.rc.code;
5645 if (rc)
5646 dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc);
5647 }
5648
handle_query_map_rsp(union ibmvnic_crq * crq,struct ibmvnic_adapter * adapter)5649 static void handle_query_map_rsp(union ibmvnic_crq *crq,
5650 struct ibmvnic_adapter *adapter)
5651 {
5652 struct net_device *netdev = adapter->netdev;
5653 struct device *dev = &adapter->vdev->dev;
5654 long rc;
5655
5656 rc = crq->query_map_rsp.rc.code;
5657 if (rc) {
5658 dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc);
5659 return;
5660 }
5661 netdev_dbg(netdev, "page_size = %d\ntot_pages = %u\nfree_pages = %u\n",
5662 crq->query_map_rsp.page_size,
5663 __be32_to_cpu(crq->query_map_rsp.tot_pages),
5664 __be32_to_cpu(crq->query_map_rsp.free_pages));
5665 }
5666
handle_query_cap_rsp(union ibmvnic_crq * crq,struct ibmvnic_adapter * adapter)5667 static void handle_query_cap_rsp(union ibmvnic_crq *crq,
5668 struct ibmvnic_adapter *adapter)
5669 {
5670 struct net_device *netdev = adapter->netdev;
5671 struct device *dev = &adapter->vdev->dev;
5672 long rc;
5673
5674 atomic_dec(&adapter->running_cap_crqs);
5675 netdev_dbg(netdev, "Outstanding queries: %d\n",
5676 atomic_read(&adapter->running_cap_crqs));
5677 rc = crq->query_capability.rc.code;
5678 if (rc) {
5679 dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc);
5680 goto out;
5681 }
5682
5683 switch (be16_to_cpu(crq->query_capability.capability)) {
5684 case MIN_TX_QUEUES:
5685 adapter->min_tx_queues =
5686 be64_to_cpu(crq->query_capability.number);
5687 netdev_dbg(netdev, "min_tx_queues = %lld\n",
5688 adapter->min_tx_queues);
5689 break;
5690 case MIN_RX_QUEUES:
5691 adapter->min_rx_queues =
5692 be64_to_cpu(crq->query_capability.number);
5693 netdev_dbg(netdev, "min_rx_queues = %lld\n",
5694 adapter->min_rx_queues);
5695 break;
5696 case MIN_RX_ADD_QUEUES:
5697 adapter->min_rx_add_queues =
5698 be64_to_cpu(crq->query_capability.number);
5699 netdev_dbg(netdev, "min_rx_add_queues = %lld\n",
5700 adapter->min_rx_add_queues);
5701 break;
5702 case MAX_TX_QUEUES:
5703 adapter->max_tx_queues =
5704 be64_to_cpu(crq->query_capability.number);
5705 netdev_dbg(netdev, "max_tx_queues = %lld\n",
5706 adapter->max_tx_queues);
5707 break;
5708 case MAX_RX_QUEUES:
5709 adapter->max_rx_queues =
5710 be64_to_cpu(crq->query_capability.number);
5711 netdev_dbg(netdev, "max_rx_queues = %lld\n",
5712 adapter->max_rx_queues);
5713 break;
5714 case MAX_RX_ADD_QUEUES:
5715 adapter->max_rx_add_queues =
5716 be64_to_cpu(crq->query_capability.number);
5717 netdev_dbg(netdev, "max_rx_add_queues = %lld\n",
5718 adapter->max_rx_add_queues);
5719 break;
5720 case MIN_TX_ENTRIES_PER_SUBCRQ:
5721 adapter->min_tx_entries_per_subcrq =
5722 be64_to_cpu(crq->query_capability.number);
5723 netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n",
5724 adapter->min_tx_entries_per_subcrq);
5725 break;
5726 case MIN_RX_ADD_ENTRIES_PER_SUBCRQ:
5727 adapter->min_rx_add_entries_per_subcrq =
5728 be64_to_cpu(crq->query_capability.number);
5729 netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n",
5730 adapter->min_rx_add_entries_per_subcrq);
5731 break;
5732 case MAX_TX_ENTRIES_PER_SUBCRQ:
5733 adapter->max_tx_entries_per_subcrq =
5734 be64_to_cpu(crq->query_capability.number);
5735 netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n",
5736 adapter->max_tx_entries_per_subcrq);
5737 break;
5738 case MAX_RX_ADD_ENTRIES_PER_SUBCRQ:
5739 adapter->max_rx_add_entries_per_subcrq =
5740 be64_to_cpu(crq->query_capability.number);
5741 netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n",
5742 adapter->max_rx_add_entries_per_subcrq);
5743 break;
5744 case TCP_IP_OFFLOAD:
5745 adapter->tcp_ip_offload =
5746 be64_to_cpu(crq->query_capability.number);
5747 netdev_dbg(netdev, "tcp_ip_offload = %lld\n",
5748 adapter->tcp_ip_offload);
5749 break;
5750 case PROMISC_SUPPORTED:
5751 adapter->promisc_supported =
5752 be64_to_cpu(crq->query_capability.number);
5753 netdev_dbg(netdev, "promisc_supported = %lld\n",
5754 adapter->promisc_supported);
5755 break;
5756 case MIN_MTU:
5757 adapter->min_mtu = be64_to_cpu(crq->query_capability.number);
5758 netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
5759 netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu);
5760 break;
5761 case MAX_MTU:
5762 adapter->max_mtu = be64_to_cpu(crq->query_capability.number);
5763 netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
5764 netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu);
5765 break;
5766 case MAX_MULTICAST_FILTERS:
5767 adapter->max_multicast_filters =
5768 be64_to_cpu(crq->query_capability.number);
5769 netdev_dbg(netdev, "max_multicast_filters = %lld\n",
5770 adapter->max_multicast_filters);
5771 break;
5772 case VLAN_HEADER_INSERTION:
5773 adapter->vlan_header_insertion =
5774 be64_to_cpu(crq->query_capability.number);
5775 if (adapter->vlan_header_insertion)
5776 netdev->features |= NETIF_F_HW_VLAN_STAG_TX;
5777 netdev_dbg(netdev, "vlan_header_insertion = %lld\n",
5778 adapter->vlan_header_insertion);
5779 break;
5780 case RX_VLAN_HEADER_INSERTION:
5781 adapter->rx_vlan_header_insertion =
5782 be64_to_cpu(crq->query_capability.number);
5783 netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n",
5784 adapter->rx_vlan_header_insertion);
5785 break;
5786 case MAX_TX_SG_ENTRIES:
5787 adapter->max_tx_sg_entries =
5788 be64_to_cpu(crq->query_capability.number);
5789 netdev_dbg(netdev, "max_tx_sg_entries = %lld\n",
5790 adapter->max_tx_sg_entries);
5791 break;
5792 case RX_SG_SUPPORTED:
5793 adapter->rx_sg_supported =
5794 be64_to_cpu(crq->query_capability.number);
5795 netdev_dbg(netdev, "rx_sg_supported = %lld\n",
5796 adapter->rx_sg_supported);
5797 break;
5798 case OPT_TX_COMP_SUB_QUEUES:
5799 adapter->opt_tx_comp_sub_queues =
5800 be64_to_cpu(crq->query_capability.number);
5801 netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n",
5802 adapter->opt_tx_comp_sub_queues);
5803 break;
5804 case OPT_RX_COMP_QUEUES:
5805 adapter->opt_rx_comp_queues =
5806 be64_to_cpu(crq->query_capability.number);
5807 netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n",
5808 adapter->opt_rx_comp_queues);
5809 break;
5810 case OPT_RX_BUFADD_Q_PER_RX_COMP_Q:
5811 adapter->opt_rx_bufadd_q_per_rx_comp_q =
5812 be64_to_cpu(crq->query_capability.number);
5813 netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n",
5814 adapter->opt_rx_bufadd_q_per_rx_comp_q);
5815 break;
5816 case OPT_TX_ENTRIES_PER_SUBCRQ:
5817 adapter->opt_tx_entries_per_subcrq =
5818 be64_to_cpu(crq->query_capability.number);
5819 netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n",
5820 adapter->opt_tx_entries_per_subcrq);
5821 break;
5822 case OPT_RXBA_ENTRIES_PER_SUBCRQ:
5823 adapter->opt_rxba_entries_per_subcrq =
5824 be64_to_cpu(crq->query_capability.number);
5825 netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n",
5826 adapter->opt_rxba_entries_per_subcrq);
5827 break;
5828 case TX_RX_DESC_REQ:
5829 adapter->tx_rx_desc_req = crq->query_capability.number;
5830 netdev_dbg(netdev, "tx_rx_desc_req = %llx\n",
5831 adapter->tx_rx_desc_req);
5832 break;
5833
5834 default:
5835 netdev_err(netdev, "Got invalid cap rsp %d\n",
5836 crq->query_capability.capability);
5837 }
5838
5839 out:
5840 if (atomic_read(&adapter->running_cap_crqs) == 0)
5841 send_request_cap(adapter, 0);
5842 }
5843
send_query_phys_parms(struct ibmvnic_adapter * adapter)5844 static int send_query_phys_parms(struct ibmvnic_adapter *adapter)
5845 {
5846 union ibmvnic_crq crq;
5847 int rc;
5848
5849 memset(&crq, 0, sizeof(crq));
5850 crq.query_phys_parms.first = IBMVNIC_CRQ_CMD;
5851 crq.query_phys_parms.cmd = QUERY_PHYS_PARMS;
5852
5853 mutex_lock(&adapter->fw_lock);
5854 adapter->fw_done_rc = 0;
5855 reinit_completion(&adapter->fw_done);
5856
5857 rc = ibmvnic_send_crq(adapter, &crq);
5858 if (rc) {
5859 mutex_unlock(&adapter->fw_lock);
5860 return rc;
5861 }
5862
5863 rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
5864 if (rc) {
5865 mutex_unlock(&adapter->fw_lock);
5866 return rc;
5867 }
5868
5869 mutex_unlock(&adapter->fw_lock);
5870 return adapter->fw_done_rc ? -EIO : 0;
5871 }
5872
handle_query_phys_parms_rsp(union ibmvnic_crq * crq,struct ibmvnic_adapter * adapter)5873 static int handle_query_phys_parms_rsp(union ibmvnic_crq *crq,
5874 struct ibmvnic_adapter *adapter)
5875 {
5876 struct net_device *netdev = adapter->netdev;
5877 int rc;
5878 __be32 rspeed = cpu_to_be32(crq->query_phys_parms_rsp.speed);
5879
5880 rc = crq->query_phys_parms_rsp.rc.code;
5881 if (rc) {
5882 netdev_err(netdev, "Error %d in QUERY_PHYS_PARMS\n", rc);
5883 return rc;
5884 }
5885 switch (rspeed) {
5886 case IBMVNIC_10MBPS:
5887 adapter->speed = SPEED_10;
5888 break;
5889 case IBMVNIC_100MBPS:
5890 adapter->speed = SPEED_100;
5891 break;
5892 case IBMVNIC_1GBPS:
5893 adapter->speed = SPEED_1000;
5894 break;
5895 case IBMVNIC_10GBPS:
5896 adapter->speed = SPEED_10000;
5897 break;
5898 case IBMVNIC_25GBPS:
5899 adapter->speed = SPEED_25000;
5900 break;
5901 case IBMVNIC_40GBPS:
5902 adapter->speed = SPEED_40000;
5903 break;
5904 case IBMVNIC_50GBPS:
5905 adapter->speed = SPEED_50000;
5906 break;
5907 case IBMVNIC_100GBPS:
5908 adapter->speed = SPEED_100000;
5909 break;
5910 case IBMVNIC_200GBPS:
5911 adapter->speed = SPEED_200000;
5912 break;
5913 default:
5914 if (netif_carrier_ok(netdev))
5915 netdev_warn(netdev, "Unknown speed 0x%08x\n", rspeed);
5916 adapter->speed = SPEED_UNKNOWN;
5917 }
5918 if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_FULL_DUPLEX)
5919 adapter->duplex = DUPLEX_FULL;
5920 else if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_HALF_DUPLEX)
5921 adapter->duplex = DUPLEX_HALF;
5922 else
5923 adapter->duplex = DUPLEX_UNKNOWN;
5924
5925 return rc;
5926 }
5927
ibmvnic_handle_crq(union ibmvnic_crq * crq,struct ibmvnic_adapter * adapter)5928 static void ibmvnic_handle_crq(union ibmvnic_crq *crq,
5929 struct ibmvnic_adapter *adapter)
5930 {
5931 struct ibmvnic_generic_crq *gen_crq = &crq->generic;
5932 struct net_device *netdev = adapter->netdev;
5933 struct device *dev = &adapter->vdev->dev;
5934 u64 *u64_crq = (u64 *)crq;
5935 long rc;
5936
5937 netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n",
5938 (unsigned long)cpu_to_be64(u64_crq[0]),
5939 (unsigned long)cpu_to_be64(u64_crq[1]));
5940 switch (gen_crq->first) {
5941 case IBMVNIC_CRQ_INIT_RSP:
5942 switch (gen_crq->cmd) {
5943 case IBMVNIC_CRQ_INIT:
5944 dev_info(dev, "Partner initialized\n");
5945 adapter->from_passive_init = true;
5946 /* Discard any stale login responses from prev reset.
5947 * CHECK: should we clear even on INIT_COMPLETE?
5948 */
5949 adapter->login_pending = false;
5950
5951 if (adapter->state == VNIC_DOWN)
5952 rc = ibmvnic_reset(adapter, VNIC_RESET_PASSIVE_INIT);
5953 else
5954 rc = ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
5955
5956 if (rc && rc != -EBUSY) {
5957 /* We were unable to schedule the failover
5958 * reset either because the adapter was still
5959 * probing (eg: during kexec) or we could not
5960 * allocate memory. Clear the failover_pending
5961 * flag since no one else will. We ignore
5962 * EBUSY because it means either FAILOVER reset
5963 * is already scheduled or the adapter is
5964 * being removed.
5965 */
5966 netdev_err(netdev,
5967 "Error %ld scheduling failover reset\n",
5968 rc);
5969 adapter->failover_pending = false;
5970 }
5971
5972 if (!completion_done(&adapter->init_done)) {
5973 if (!adapter->init_done_rc)
5974 adapter->init_done_rc = -EAGAIN;
5975 complete(&adapter->init_done);
5976 }
5977
5978 break;
5979 case IBMVNIC_CRQ_INIT_COMPLETE:
5980 dev_info(dev, "Partner initialization complete\n");
5981 adapter->crq.active = true;
5982 send_version_xchg(adapter);
5983 break;
5984 default:
5985 dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd);
5986 }
5987 return;
5988 case IBMVNIC_CRQ_XPORT_EVENT:
5989 netif_carrier_off(netdev);
5990 adapter->crq.active = false;
5991 /* terminate any thread waiting for a response
5992 * from the device
5993 */
5994 if (!completion_done(&adapter->fw_done)) {
5995 adapter->fw_done_rc = -EIO;
5996 complete(&adapter->fw_done);
5997 }
5998
5999 /* if we got here during crq-init, retry crq-init */
6000 if (!completion_done(&adapter->init_done)) {
6001 adapter->init_done_rc = -EAGAIN;
6002 complete(&adapter->init_done);
6003 }
6004
6005 if (!completion_done(&adapter->stats_done))
6006 complete(&adapter->stats_done);
6007 if (test_bit(0, &adapter->resetting))
6008 adapter->force_reset_recovery = true;
6009 if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) {
6010 dev_info(dev, "Migrated, re-enabling adapter\n");
6011 ibmvnic_reset(adapter, VNIC_RESET_MOBILITY);
6012 } else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) {
6013 dev_info(dev, "Backing device failover detected\n");
6014 adapter->failover_pending = true;
6015 } else {
6016 /* The adapter lost the connection */
6017 dev_err(dev, "Virtual Adapter failed (rc=%d)\n",
6018 gen_crq->cmd);
6019 ibmvnic_reset(adapter, VNIC_RESET_FATAL);
6020 }
6021 return;
6022 case IBMVNIC_CRQ_CMD_RSP:
6023 break;
6024 default:
6025 dev_err(dev, "Got an invalid msg type 0x%02x\n",
6026 gen_crq->first);
6027 return;
6028 }
6029
6030 switch (gen_crq->cmd) {
6031 case VERSION_EXCHANGE_RSP:
6032 rc = crq->version_exchange_rsp.rc.code;
6033 if (rc) {
6034 dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc);
6035 break;
6036 }
6037 ibmvnic_version =
6038 be16_to_cpu(crq->version_exchange_rsp.version);
6039 dev_info(dev, "Partner protocol version is %d\n",
6040 ibmvnic_version);
6041 send_query_cap(adapter);
6042 break;
6043 case QUERY_CAPABILITY_RSP:
6044 handle_query_cap_rsp(crq, adapter);
6045 break;
6046 case QUERY_MAP_RSP:
6047 handle_query_map_rsp(crq, adapter);
6048 break;
6049 case REQUEST_MAP_RSP:
6050 adapter->fw_done_rc = crq->request_map_rsp.rc.code;
6051 complete(&adapter->fw_done);
6052 break;
6053 case REQUEST_UNMAP_RSP:
6054 handle_request_unmap_rsp(crq, adapter);
6055 break;
6056 case REQUEST_CAPABILITY_RSP:
6057 handle_request_cap_rsp(crq, adapter);
6058 break;
6059 case LOGIN_RSP:
6060 netdev_dbg(netdev, "Got Login Response\n");
6061 handle_login_rsp(crq, adapter);
6062 break;
6063 case LOGICAL_LINK_STATE_RSP:
6064 netdev_dbg(netdev,
6065 "Got Logical Link State Response, state: %d rc: %d\n",
6066 crq->logical_link_state_rsp.link_state,
6067 crq->logical_link_state_rsp.rc.code);
6068 adapter->logical_link_state =
6069 crq->logical_link_state_rsp.link_state;
6070 adapter->init_done_rc = crq->logical_link_state_rsp.rc.code;
6071 complete(&adapter->init_done);
6072 break;
6073 case LINK_STATE_INDICATION:
6074 netdev_dbg(netdev, "Got Logical Link State Indication\n");
6075 adapter->phys_link_state =
6076 crq->link_state_indication.phys_link_state;
6077 adapter->logical_link_state =
6078 crq->link_state_indication.logical_link_state;
6079 if (adapter->phys_link_state && adapter->logical_link_state)
6080 netif_carrier_on(netdev);
6081 else
6082 netif_carrier_off(netdev);
6083 break;
6084 case CHANGE_MAC_ADDR_RSP:
6085 netdev_dbg(netdev, "Got MAC address change Response\n");
6086 adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter);
6087 break;
6088 case ERROR_INDICATION:
6089 netdev_dbg(netdev, "Got Error Indication\n");
6090 handle_error_indication(crq, adapter);
6091 break;
6092 case REQUEST_STATISTICS_RSP:
6093 netdev_dbg(netdev, "Got Statistics Response\n");
6094 complete(&adapter->stats_done);
6095 break;
6096 case QUERY_IP_OFFLOAD_RSP:
6097 netdev_dbg(netdev, "Got Query IP offload Response\n");
6098 handle_query_ip_offload_rsp(adapter);
6099 break;
6100 case MULTICAST_CTRL_RSP:
6101 netdev_dbg(netdev, "Got multicast control Response\n");
6102 break;
6103 case CONTROL_IP_OFFLOAD_RSP:
6104 netdev_dbg(netdev, "Got Control IP offload Response\n");
6105 dma_unmap_single(dev, adapter->ip_offload_ctrl_tok,
6106 sizeof(adapter->ip_offload_ctrl),
6107 DMA_TO_DEVICE);
6108 complete(&adapter->init_done);
6109 break;
6110 case COLLECT_FW_TRACE_RSP:
6111 netdev_dbg(netdev, "Got Collect firmware trace Response\n");
6112 complete(&adapter->fw_done);
6113 break;
6114 case GET_VPD_SIZE_RSP:
6115 handle_vpd_size_rsp(crq, adapter);
6116 break;
6117 case GET_VPD_RSP:
6118 handle_vpd_rsp(crq, adapter);
6119 break;
6120 case QUERY_PHYS_PARMS_RSP:
6121 adapter->fw_done_rc = handle_query_phys_parms_rsp(crq, adapter);
6122 complete(&adapter->fw_done);
6123 break;
6124 default:
6125 netdev_err(netdev, "Got an invalid cmd type 0x%02x\n",
6126 gen_crq->cmd);
6127 }
6128 }
6129
ibmvnic_interrupt(int irq,void * instance)6130 static irqreturn_t ibmvnic_interrupt(int irq, void *instance)
6131 {
6132 struct ibmvnic_adapter *adapter = instance;
6133
6134 tasklet_schedule(&adapter->tasklet);
6135 return IRQ_HANDLED;
6136 }
6137
ibmvnic_tasklet(struct tasklet_struct * t)6138 static void ibmvnic_tasklet(struct tasklet_struct *t)
6139 {
6140 struct ibmvnic_adapter *adapter = from_tasklet(adapter, t, tasklet);
6141 struct ibmvnic_crq_queue *queue = &adapter->crq;
6142 union ibmvnic_crq *crq;
6143 unsigned long flags;
6144
6145 spin_lock_irqsave(&queue->lock, flags);
6146
6147 /* Pull all the valid messages off the CRQ */
6148 while ((crq = ibmvnic_next_crq(adapter)) != NULL) {
6149 /* This barrier makes sure ibmvnic_next_crq()'s
6150 * crq->generic.first & IBMVNIC_CRQ_CMD_RSP is loaded
6151 * before ibmvnic_handle_crq()'s
6152 * switch(gen_crq->first) and switch(gen_crq->cmd).
6153 */
6154 dma_rmb();
6155 ibmvnic_handle_crq(crq, adapter);
6156 crq->generic.first = 0;
6157 }
6158
6159 spin_unlock_irqrestore(&queue->lock, flags);
6160 }
6161
ibmvnic_reenable_crq_queue(struct ibmvnic_adapter * adapter)6162 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter)
6163 {
6164 struct vio_dev *vdev = adapter->vdev;
6165 int rc;
6166
6167 do {
6168 rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
6169 } while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc));
6170
6171 if (rc)
6172 dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc);
6173
6174 return rc;
6175 }
6176
ibmvnic_reset_crq(struct ibmvnic_adapter * adapter)6177 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter)
6178 {
6179 struct ibmvnic_crq_queue *crq = &adapter->crq;
6180 struct device *dev = &adapter->vdev->dev;
6181 struct vio_dev *vdev = adapter->vdev;
6182 int rc;
6183
6184 /* Close the CRQ */
6185 do {
6186 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
6187 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
6188
6189 /* Clean out the queue */
6190 if (!crq->msgs)
6191 return -EINVAL;
6192
6193 memset(crq->msgs, 0, PAGE_SIZE);
6194 crq->cur = 0;
6195 crq->active = false;
6196
6197 /* And re-open it again */
6198 rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
6199 crq->msg_token, PAGE_SIZE);
6200
6201 if (rc == H_CLOSED)
6202 /* Adapter is good, but other end is not ready */
6203 dev_warn(dev, "Partner adapter not ready\n");
6204 else if (rc != 0)
6205 dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc);
6206
6207 return rc;
6208 }
6209
release_crq_queue(struct ibmvnic_adapter * adapter)6210 static void release_crq_queue(struct ibmvnic_adapter *adapter)
6211 {
6212 struct ibmvnic_crq_queue *crq = &adapter->crq;
6213 struct vio_dev *vdev = adapter->vdev;
6214 long rc;
6215
6216 if (!crq->msgs)
6217 return;
6218
6219 netdev_dbg(adapter->netdev, "Releasing CRQ\n");
6220 free_irq(vdev->irq, adapter);
6221 tasklet_kill(&adapter->tasklet);
6222 do {
6223 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
6224 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
6225
6226 dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE,
6227 DMA_BIDIRECTIONAL);
6228 free_page((unsigned long)crq->msgs);
6229 crq->msgs = NULL;
6230 crq->active = false;
6231 }
6232
init_crq_queue(struct ibmvnic_adapter * adapter)6233 static int init_crq_queue(struct ibmvnic_adapter *adapter)
6234 {
6235 struct ibmvnic_crq_queue *crq = &adapter->crq;
6236 struct device *dev = &adapter->vdev->dev;
6237 struct vio_dev *vdev = adapter->vdev;
6238 int rc, retrc = -ENOMEM;
6239
6240 if (crq->msgs)
6241 return 0;
6242
6243 crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL);
6244 /* Should we allocate more than one page? */
6245
6246 if (!crq->msgs)
6247 return -ENOMEM;
6248
6249 crq->size = PAGE_SIZE / sizeof(*crq->msgs);
6250 crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE,
6251 DMA_BIDIRECTIONAL);
6252 if (dma_mapping_error(dev, crq->msg_token))
6253 goto map_failed;
6254
6255 rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
6256 crq->msg_token, PAGE_SIZE);
6257
6258 if (rc == H_RESOURCE)
6259 /* maybe kexecing and resource is busy. try a reset */
6260 rc = ibmvnic_reset_crq(adapter);
6261 retrc = rc;
6262
6263 if (rc == H_CLOSED) {
6264 dev_warn(dev, "Partner adapter not ready\n");
6265 } else if (rc) {
6266 dev_warn(dev, "Error %d opening adapter\n", rc);
6267 goto reg_crq_failed;
6268 }
6269
6270 retrc = 0;
6271
6272 tasklet_setup(&adapter->tasklet, (void *)ibmvnic_tasklet);
6273
6274 netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq);
6275 snprintf(crq->name, sizeof(crq->name), "ibmvnic-%x",
6276 adapter->vdev->unit_address);
6277 rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, crq->name, adapter);
6278 if (rc) {
6279 dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n",
6280 vdev->irq, rc);
6281 goto req_irq_failed;
6282 }
6283
6284 rc = vio_enable_interrupts(vdev);
6285 if (rc) {
6286 dev_err(dev, "Error %d enabling interrupts\n", rc);
6287 goto req_irq_failed;
6288 }
6289
6290 crq->cur = 0;
6291 spin_lock_init(&crq->lock);
6292
6293 /* process any CRQs that were queued before we enabled interrupts */
6294 tasklet_schedule(&adapter->tasklet);
6295
6296 return retrc;
6297
6298 req_irq_failed:
6299 tasklet_kill(&adapter->tasklet);
6300 do {
6301 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
6302 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
6303 reg_crq_failed:
6304 dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL);
6305 map_failed:
6306 free_page((unsigned long)crq->msgs);
6307 crq->msgs = NULL;
6308 return retrc;
6309 }
6310
ibmvnic_reset_init(struct ibmvnic_adapter * adapter,bool reset)6311 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset)
6312 {
6313 struct device *dev = &adapter->vdev->dev;
6314 unsigned long timeout = msecs_to_jiffies(20000);
6315 u64 old_num_rx_queues = adapter->req_rx_queues;
6316 u64 old_num_tx_queues = adapter->req_tx_queues;
6317 int rc;
6318
6319 adapter->from_passive_init = false;
6320
6321 rc = ibmvnic_send_crq_init(adapter);
6322 if (rc) {
6323 dev_err(dev, "Send crq init failed with error %d\n", rc);
6324 return rc;
6325 }
6326
6327 if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
6328 dev_err(dev, "Initialization sequence timed out\n");
6329 return -ETIMEDOUT;
6330 }
6331
6332 if (adapter->init_done_rc) {
6333 release_crq_queue(adapter);
6334 dev_err(dev, "CRQ-init failed, %d\n", adapter->init_done_rc);
6335 return adapter->init_done_rc;
6336 }
6337
6338 if (adapter->from_passive_init) {
6339 adapter->state = VNIC_OPEN;
6340 adapter->from_passive_init = false;
6341 dev_err(dev, "CRQ-init failed, passive-init\n");
6342 return -EINVAL;
6343 }
6344
6345 if (reset &&
6346 test_bit(0, &adapter->resetting) && !adapter->wait_for_reset &&
6347 adapter->reset_reason != VNIC_RESET_MOBILITY) {
6348 if (adapter->req_rx_queues != old_num_rx_queues ||
6349 adapter->req_tx_queues != old_num_tx_queues) {
6350 release_sub_crqs(adapter, 0);
6351 rc = init_sub_crqs(adapter);
6352 } else {
6353 /* no need to reinitialize completely, but we do
6354 * need to clean up transmits that were in flight
6355 * when we processed the reset. Failure to do so
6356 * will confound the upper layer, usually TCP, by
6357 * creating the illusion of transmits that are
6358 * awaiting completion.
6359 */
6360 clean_tx_pools(adapter);
6361
6362 rc = reset_sub_crq_queues(adapter);
6363 }
6364 } else {
6365 rc = init_sub_crqs(adapter);
6366 }
6367
6368 if (rc) {
6369 dev_err(dev, "Initialization of sub crqs failed\n");
6370 release_crq_queue(adapter);
6371 return rc;
6372 }
6373
6374 rc = init_sub_crq_irqs(adapter);
6375 if (rc) {
6376 dev_err(dev, "Failed to initialize sub crq irqs\n");
6377 release_crq_queue(adapter);
6378 }
6379
6380 return rc;
6381 }
6382
6383 static struct device_attribute dev_attr_failover;
6384
ibmvnic_probe(struct vio_dev * dev,const struct vio_device_id * id)6385 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
6386 {
6387 struct ibmvnic_adapter *adapter;
6388 struct net_device *netdev;
6389 unsigned char *mac_addr_p;
6390 unsigned long flags;
6391 bool init_success;
6392 int rc;
6393
6394 dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n",
6395 dev->unit_address);
6396
6397 mac_addr_p = (unsigned char *)vio_get_attribute(dev,
6398 VETH_MAC_ADDR, NULL);
6399 if (!mac_addr_p) {
6400 dev_err(&dev->dev,
6401 "(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n",
6402 __FILE__, __LINE__);
6403 return 0;
6404 }
6405
6406 netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter),
6407 IBMVNIC_MAX_QUEUES);
6408 if (!netdev)
6409 return -ENOMEM;
6410
6411 adapter = netdev_priv(netdev);
6412 adapter->state = VNIC_PROBING;
6413 dev_set_drvdata(&dev->dev, netdev);
6414 adapter->vdev = dev;
6415 adapter->netdev = netdev;
6416 adapter->login_pending = false;
6417 memset(&adapter->map_ids, 0, sizeof(adapter->map_ids));
6418 /* map_ids start at 1, so ensure map_id 0 is always "in-use" */
6419 bitmap_set(adapter->map_ids, 0, 1);
6420
6421 ether_addr_copy(adapter->mac_addr, mac_addr_p);
6422 eth_hw_addr_set(netdev, adapter->mac_addr);
6423 netdev->irq = dev->irq;
6424 netdev->netdev_ops = &ibmvnic_netdev_ops;
6425 netdev->ethtool_ops = &ibmvnic_ethtool_ops;
6426 SET_NETDEV_DEV(netdev, &dev->dev);
6427
6428 INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset);
6429 INIT_DELAYED_WORK(&adapter->ibmvnic_delayed_reset,
6430 __ibmvnic_delayed_reset);
6431 INIT_LIST_HEAD(&adapter->rwi_list);
6432 spin_lock_init(&adapter->rwi_lock);
6433 spin_lock_init(&adapter->state_lock);
6434 mutex_init(&adapter->fw_lock);
6435 init_completion(&adapter->probe_done);
6436 init_completion(&adapter->init_done);
6437 init_completion(&adapter->fw_done);
6438 init_completion(&adapter->reset_done);
6439 init_completion(&adapter->stats_done);
6440 clear_bit(0, &adapter->resetting);
6441 adapter->prev_rx_buf_sz = 0;
6442 adapter->prev_mtu = 0;
6443
6444 init_success = false;
6445 do {
6446 reinit_init_done(adapter);
6447
6448 /* clear any failovers we got in the previous pass
6449 * since we are reinitializing the CRQ
6450 */
6451 adapter->failover_pending = false;
6452
6453 /* If we had already initialized CRQ, we may have one or
6454 * more resets queued already. Discard those and release
6455 * the CRQ before initializing the CRQ again.
6456 */
6457 release_crq_queue(adapter);
6458
6459 /* Since we are still in PROBING state, __ibmvnic_reset()
6460 * will not access the ->rwi_list and since we released CRQ,
6461 * we won't get _new_ transport events. But there maybe an
6462 * ongoing ibmvnic_reset() call. So serialize access to
6463 * rwi_list. If we win the race, ibvmnic_reset() could add
6464 * a reset after we purged but thats ok - we just may end
6465 * up with an extra reset (i.e similar to having two or more
6466 * resets in the queue at once).
6467 * CHECK.
6468 */
6469 spin_lock_irqsave(&adapter->rwi_lock, flags);
6470 flush_reset_queue(adapter);
6471 spin_unlock_irqrestore(&adapter->rwi_lock, flags);
6472
6473 rc = init_crq_queue(adapter);
6474 if (rc) {
6475 dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n",
6476 rc);
6477 goto ibmvnic_init_fail;
6478 }
6479
6480 rc = ibmvnic_reset_init(adapter, false);
6481 } while (rc == -EAGAIN);
6482
6483 /* We are ignoring the error from ibmvnic_reset_init() assuming that the
6484 * partner is not ready. CRQ is not active. When the partner becomes
6485 * ready, we will do the passive init reset.
6486 */
6487
6488 if (!rc)
6489 init_success = true;
6490
6491 rc = init_stats_buffers(adapter);
6492 if (rc)
6493 goto ibmvnic_init_fail;
6494
6495 rc = init_stats_token(adapter);
6496 if (rc)
6497 goto ibmvnic_stats_fail;
6498
6499 rc = device_create_file(&dev->dev, &dev_attr_failover);
6500 if (rc)
6501 goto ibmvnic_dev_file_err;
6502
6503 netif_carrier_off(netdev);
6504
6505 if (init_success) {
6506 adapter->state = VNIC_PROBED;
6507 netdev->mtu = adapter->req_mtu - ETH_HLEN;
6508 netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
6509 netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
6510 } else {
6511 adapter->state = VNIC_DOWN;
6512 }
6513
6514 adapter->wait_for_reset = false;
6515 adapter->last_reset_time = jiffies;
6516
6517 rc = register_netdev(netdev);
6518 if (rc) {
6519 dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc);
6520 goto ibmvnic_register_fail;
6521 }
6522 dev_info(&dev->dev, "ibmvnic registered\n");
6523
6524 rc = ibmvnic_cpu_notif_add(adapter);
6525 if (rc) {
6526 netdev_err(netdev, "Registering cpu notifier failed\n");
6527 goto cpu_notif_add_failed;
6528 }
6529
6530 complete(&adapter->probe_done);
6531
6532 return 0;
6533
6534 cpu_notif_add_failed:
6535 unregister_netdev(netdev);
6536
6537 ibmvnic_register_fail:
6538 device_remove_file(&dev->dev, &dev_attr_failover);
6539
6540 ibmvnic_dev_file_err:
6541 release_stats_token(adapter);
6542
6543 ibmvnic_stats_fail:
6544 release_stats_buffers(adapter);
6545
6546 ibmvnic_init_fail:
6547 release_sub_crqs(adapter, 1);
6548 release_crq_queue(adapter);
6549
6550 /* cleanup worker thread after releasing CRQ so we don't get
6551 * transport events (i.e new work items for the worker thread).
6552 */
6553 adapter->state = VNIC_REMOVING;
6554 complete(&adapter->probe_done);
6555 flush_work(&adapter->ibmvnic_reset);
6556 flush_delayed_work(&adapter->ibmvnic_delayed_reset);
6557
6558 flush_reset_queue(adapter);
6559
6560 mutex_destroy(&adapter->fw_lock);
6561 free_netdev(netdev);
6562
6563 return rc;
6564 }
6565
ibmvnic_remove(struct vio_dev * dev)6566 static void ibmvnic_remove(struct vio_dev *dev)
6567 {
6568 struct net_device *netdev = dev_get_drvdata(&dev->dev);
6569 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
6570 unsigned long flags;
6571
6572 spin_lock_irqsave(&adapter->state_lock, flags);
6573
6574 /* If ibmvnic_reset() is scheduling a reset, wait for it to
6575 * finish. Then, set the state to REMOVING to prevent it from
6576 * scheduling any more work and to have reset functions ignore
6577 * any resets that have already been scheduled. Drop the lock
6578 * after setting state, so __ibmvnic_reset() which is called
6579 * from the flush_work() below, can make progress.
6580 */
6581 spin_lock(&adapter->rwi_lock);
6582 adapter->state = VNIC_REMOVING;
6583 spin_unlock(&adapter->rwi_lock);
6584
6585 spin_unlock_irqrestore(&adapter->state_lock, flags);
6586
6587 ibmvnic_cpu_notif_remove(adapter);
6588
6589 flush_work(&adapter->ibmvnic_reset);
6590 flush_delayed_work(&adapter->ibmvnic_delayed_reset);
6591
6592 rtnl_lock();
6593 unregister_netdevice(netdev);
6594
6595 release_resources(adapter);
6596 release_rx_pools(adapter);
6597 release_tx_pools(adapter);
6598 release_sub_crqs(adapter, 1);
6599 release_crq_queue(adapter);
6600
6601 release_stats_token(adapter);
6602 release_stats_buffers(adapter);
6603
6604 adapter->state = VNIC_REMOVED;
6605
6606 rtnl_unlock();
6607 mutex_destroy(&adapter->fw_lock);
6608 device_remove_file(&dev->dev, &dev_attr_failover);
6609 free_netdev(netdev);
6610 dev_set_drvdata(&dev->dev, NULL);
6611 }
6612
failover_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)6613 static ssize_t failover_store(struct device *dev, struct device_attribute *attr,
6614 const char *buf, size_t count)
6615 {
6616 struct net_device *netdev = dev_get_drvdata(dev);
6617 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
6618 unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
6619 __be64 session_token;
6620 long rc;
6621
6622 if (!sysfs_streq(buf, "1"))
6623 return -EINVAL;
6624
6625 rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address,
6626 H_GET_SESSION_TOKEN, 0, 0, 0);
6627 if (rc) {
6628 netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n",
6629 rc);
6630 goto last_resort;
6631 }
6632
6633 session_token = (__be64)retbuf[0];
6634 netdev_dbg(netdev, "Initiating client failover, session id %llx\n",
6635 be64_to_cpu(session_token));
6636 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
6637 H_SESSION_ERR_DETECTED, session_token, 0, 0);
6638 if (rc) {
6639 netdev_err(netdev,
6640 "H_VIOCTL initiated failover failed, rc %ld\n",
6641 rc);
6642 goto last_resort;
6643 }
6644
6645 return count;
6646
6647 last_resort:
6648 netdev_dbg(netdev, "Trying to send CRQ_CMD, the last resort\n");
6649 ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
6650
6651 return count;
6652 }
6653 static DEVICE_ATTR_WO(failover);
6654
ibmvnic_get_desired_dma(struct vio_dev * vdev)6655 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev)
6656 {
6657 struct net_device *netdev = dev_get_drvdata(&vdev->dev);
6658 struct ibmvnic_adapter *adapter;
6659 struct iommu_table *tbl;
6660 unsigned long ret = 0;
6661 int i;
6662
6663 tbl = get_iommu_table_base(&vdev->dev);
6664
6665 /* netdev inits at probe time along with the structures we need below*/
6666 if (!netdev)
6667 return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl);
6668
6669 adapter = netdev_priv(netdev);
6670
6671 ret += PAGE_SIZE; /* the crq message queue */
6672 ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl);
6673
6674 for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++)
6675 ret += 4 * PAGE_SIZE; /* the scrq message queue */
6676
6677 for (i = 0; i < adapter->num_active_rx_pools; i++)
6678 ret += adapter->rx_pool[i].size *
6679 IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl);
6680
6681 return ret;
6682 }
6683
ibmvnic_resume(struct device * dev)6684 static int ibmvnic_resume(struct device *dev)
6685 {
6686 struct net_device *netdev = dev_get_drvdata(dev);
6687 struct ibmvnic_adapter *adapter = netdev_priv(netdev);
6688
6689 if (adapter->state != VNIC_OPEN)
6690 return 0;
6691
6692 tasklet_schedule(&adapter->tasklet);
6693
6694 return 0;
6695 }
6696
6697 static const struct vio_device_id ibmvnic_device_table[] = {
6698 {"network", "IBM,vnic"},
6699 {"", "" }
6700 };
6701 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table);
6702
6703 static const struct dev_pm_ops ibmvnic_pm_ops = {
6704 .resume = ibmvnic_resume
6705 };
6706
6707 static struct vio_driver ibmvnic_driver = {
6708 .id_table = ibmvnic_device_table,
6709 .probe = ibmvnic_probe,
6710 .remove = ibmvnic_remove,
6711 .get_desired_dma = ibmvnic_get_desired_dma,
6712 .name = ibmvnic_driver_name,
6713 .pm = &ibmvnic_pm_ops,
6714 };
6715
6716 /* module functions */
ibmvnic_module_init(void)6717 static int __init ibmvnic_module_init(void)
6718 {
6719 int ret;
6720
6721 ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "net/ibmvnic:online",
6722 ibmvnic_cpu_online,
6723 ibmvnic_cpu_down_prep);
6724 if (ret < 0)
6725 goto out;
6726 ibmvnic_online = ret;
6727 ret = cpuhp_setup_state_multi(CPUHP_IBMVNIC_DEAD, "net/ibmvnic:dead",
6728 NULL, ibmvnic_cpu_dead);
6729 if (ret)
6730 goto err_dead;
6731
6732 ret = vio_register_driver(&ibmvnic_driver);
6733 if (ret)
6734 goto err_vio_register;
6735
6736 pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string,
6737 IBMVNIC_DRIVER_VERSION);
6738
6739 return 0;
6740 err_vio_register:
6741 cpuhp_remove_multi_state(CPUHP_IBMVNIC_DEAD);
6742 err_dead:
6743 cpuhp_remove_multi_state(ibmvnic_online);
6744 out:
6745 return ret;
6746 }
6747
ibmvnic_module_exit(void)6748 static void __exit ibmvnic_module_exit(void)
6749 {
6750 vio_unregister_driver(&ibmvnic_driver);
6751 cpuhp_remove_multi_state(CPUHP_IBMVNIC_DEAD);
6752 cpuhp_remove_multi_state(ibmvnic_online);
6753 }
6754
6755 module_init(ibmvnic_module_init);
6756 module_exit(ibmvnic_module_exit);
6757