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
75static const char ibmvnic_driver_name[] = "ibmvnic";
76static const char ibmvnic_driver_string[] = "IBM System i/p Virtual NIC Driver";
77
78MODULE_AUTHOR("Santiago Leon");
79MODULE_DESCRIPTION("IBM System i/p Virtual NIC Driver");
80MODULE_LICENSE("GPL");
81MODULE_VERSION(IBMVNIC_DRIVER_VERSION);
82
83static int ibmvnic_version = IBMVNIC_INITIAL_VERSION;
84static void release_sub_crqs(struct ibmvnic_adapter *, bool);
85static int ibmvnic_reset_crq(struct ibmvnic_adapter *);
86static int ibmvnic_send_crq_init(struct ibmvnic_adapter *);
87static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *);
88static int ibmvnic_send_crq(struct ibmvnic_adapter *, union ibmvnic_crq *);
89static int send_subcrq_indirect(struct ibmvnic_adapter *, u64, u64, u64);
90static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance);
91static int enable_scrq_irq(struct ibmvnic_adapter *,
92 struct ibmvnic_sub_crq_queue *);
93static int disable_scrq_irq(struct ibmvnic_adapter *,
94 struct ibmvnic_sub_crq_queue *);
95static int pending_scrq(struct ibmvnic_adapter *,
96 struct ibmvnic_sub_crq_queue *);
97static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *,
98 struct ibmvnic_sub_crq_queue *);
99static int ibmvnic_poll(struct napi_struct *napi, int data);
100static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter);
101static inline void reinit_init_done(struct ibmvnic_adapter *adapter);
102static void send_query_map(struct ibmvnic_adapter *adapter);
103static int send_request_map(struct ibmvnic_adapter *, dma_addr_t, u32, u8);
104static int send_request_unmap(struct ibmvnic_adapter *, u8);
105static int send_login(struct ibmvnic_adapter *adapter);
106static void send_query_cap(struct ibmvnic_adapter *adapter);
107static int init_sub_crqs(struct ibmvnic_adapter *);
108static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter);
109static int ibmvnic_reset_init(struct ibmvnic_adapter *, bool reset);
110static void release_crq_queue(struct ibmvnic_adapter *);
111static int __ibmvnic_set_mac(struct net_device *, u8 *);
112static int init_crq_queue(struct ibmvnic_adapter *adapter);
113static int send_query_phys_parms(struct ibmvnic_adapter *adapter);
114static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
115 struct ibmvnic_sub_crq_queue *tx_scrq);
116static void free_long_term_buff(struct ibmvnic_adapter *adapter,
117 struct ibmvnic_long_term_buff *ltb);
118static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter);
119static void flush_reset_queue(struct ibmvnic_adapter *adapter);
120
121struct ibmvnic_stat {
122 char name[ETH_GSTRING_LEN];
123 int offset;
124};
125
126#define IBMVNIC_STAT_OFF(stat) (offsetof(struct ibmvnic_adapter, stats) + \
127 offsetof(struct ibmvnic_statistics, stat))
128#define IBMVNIC_GET_STAT(a, off) (*((u64 *)(((unsigned long)(a)) + (off))))
129
130static const struct ibmvnic_stat ibmvnic_stats[] = {
131 {"rx_packets", IBMVNIC_STAT_OFF(rx_packets)},
132 {"rx_bytes", IBMVNIC_STAT_OFF(rx_bytes)},
133 {"tx_packets", IBMVNIC_STAT_OFF(tx_packets)},
134 {"tx_bytes", IBMVNIC_STAT_OFF(tx_bytes)},
135 {"ucast_tx_packets", IBMVNIC_STAT_OFF(ucast_tx_packets)},
136 {"ucast_rx_packets", IBMVNIC_STAT_OFF(ucast_rx_packets)},
137 {"mcast_tx_packets", IBMVNIC_STAT_OFF(mcast_tx_packets)},
138 {"mcast_rx_packets", IBMVNIC_STAT_OFF(mcast_rx_packets)},
139 {"bcast_tx_packets", IBMVNIC_STAT_OFF(bcast_tx_packets)},
140 {"bcast_rx_packets", IBMVNIC_STAT_OFF(bcast_rx_packets)},
141 {"align_errors", IBMVNIC_STAT_OFF(align_errors)},
142 {"fcs_errors", IBMVNIC_STAT_OFF(fcs_errors)},
143 {"single_collision_frames", IBMVNIC_STAT_OFF(single_collision_frames)},
144 {"multi_collision_frames", IBMVNIC_STAT_OFF(multi_collision_frames)},
145 {"sqe_test_errors", IBMVNIC_STAT_OFF(sqe_test_errors)},
146 {"deferred_tx", IBMVNIC_STAT_OFF(deferred_tx)},
147 {"late_collisions", IBMVNIC_STAT_OFF(late_collisions)},
148 {"excess_collisions", IBMVNIC_STAT_OFF(excess_collisions)},
149 {"internal_mac_tx_errors", IBMVNIC_STAT_OFF(internal_mac_tx_errors)},
150 {"carrier_sense", IBMVNIC_STAT_OFF(carrier_sense)},
151 {"too_long_frames", IBMVNIC_STAT_OFF(too_long_frames)},
152 {"internal_mac_rx_errors", IBMVNIC_STAT_OFF(internal_mac_rx_errors)},
153};
154
155static int send_crq_init_complete(struct ibmvnic_adapter *adapter)
156{
157 union ibmvnic_crq crq;
158
159 memset(&crq, 0, sizeof(crq));
160 crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
161 crq.generic.cmd = IBMVNIC_CRQ_INIT_COMPLETE;
162
163 return ibmvnic_send_crq(adapter, &crq);
164}
165
166static int send_version_xchg(struct ibmvnic_adapter *adapter)
167{
168 union ibmvnic_crq crq;
169
170 memset(&crq, 0, sizeof(crq));
171 crq.version_exchange.first = IBMVNIC_CRQ_CMD;
172 crq.version_exchange.cmd = VERSION_EXCHANGE;
173 crq.version_exchange.version = cpu_to_be16(ibmvnic_version);
174
175 return ibmvnic_send_crq(adapter, &crq);
176}
177
178static void ibmvnic_clean_queue_affinity(struct ibmvnic_adapter *adapter,
179 struct ibmvnic_sub_crq_queue *queue)
180{
181 if (!(queue && queue->irq))
182 return;
183
184 cpumask_clear(dstp: queue->affinity_mask);
185
186 if (irq_set_affinity_and_hint(irq: queue->irq, NULL))
187 netdev_warn(dev: adapter->netdev,
188 format: "%s: Clear affinity failed, queue addr = %p, IRQ = %d\n",
189 __func__, queue, queue->irq);
190}
191
192static void ibmvnic_clean_affinity(struct ibmvnic_adapter *adapter)
193{
194 struct ibmvnic_sub_crq_queue **rxqs;
195 struct ibmvnic_sub_crq_queue **txqs;
196 int num_rxqs, num_txqs;
197 int i;
198
199 rxqs = adapter->rx_scrq;
200 txqs = adapter->tx_scrq;
201 num_txqs = adapter->num_active_tx_scrqs;
202 num_rxqs = adapter->num_active_rx_scrqs;
203
204 netdev_dbg(adapter->netdev, "%s: Cleaning irq affinity hints", __func__);
205 if (txqs) {
206 for (i = 0; i < num_txqs; i++)
207 ibmvnic_clean_queue_affinity(adapter, queue: txqs[i]);
208 }
209 if (rxqs) {
210 for (i = 0; i < num_rxqs; i++)
211 ibmvnic_clean_queue_affinity(adapter, queue: rxqs[i]);
212 }
213}
214
215static int ibmvnic_set_queue_affinity(struct ibmvnic_sub_crq_queue *queue,
216 unsigned int *cpu, int *stragglers,
217 int stride)
218{
219 cpumask_var_t mask;
220 int i;
221 int rc = 0;
222
223 if (!(queue && queue->irq))
224 return rc;
225
226 /* cpumask_var_t is either a pointer or array, allocation works here */
227 if (!zalloc_cpumask_var(mask: &mask, GFP_KERNEL))
228 return -ENOMEM;
229
230 /* while we have extra cpu give one extra to this irq */
231 if (*stragglers) {
232 stride++;
233 (*stragglers)--;
234 }
235 /* atomic write is safer than writing bit by bit directly */
236 for (i = 0; i < stride; i++) {
237 cpumask_set_cpu(cpu: *cpu, dstp: mask);
238 *cpu = cpumask_next_wrap(n: *cpu, cpu_online_mask,
239 start: nr_cpu_ids, wrap: false);
240 }
241 /* set queue affinity mask */
242 cpumask_copy(dstp: queue->affinity_mask, srcp: mask);
243 rc = irq_set_affinity_and_hint(irq: queue->irq, m: queue->affinity_mask);
244 free_cpumask_var(mask);
245
246 return rc;
247}
248
249/* assumes cpu read lock is held */
250static void ibmvnic_set_affinity(struct ibmvnic_adapter *adapter)
251{
252 struct ibmvnic_sub_crq_queue **rxqs = adapter->rx_scrq;
253 struct ibmvnic_sub_crq_queue **txqs = adapter->tx_scrq;
254 struct ibmvnic_sub_crq_queue *queue;
255 int num_rxqs = adapter->num_active_rx_scrqs, i_rxqs = 0;
256 int num_txqs = adapter->num_active_tx_scrqs, i_txqs = 0;
257 int total_queues, stride, stragglers, i;
258 unsigned int num_cpu, cpu;
259 bool is_rx_queue;
260 int rc = 0;
261
262 netdev_dbg(adapter->netdev, "%s: Setting irq affinity hints", __func__);
263 if (!(adapter->rx_scrq && adapter->tx_scrq)) {
264 netdev_warn(dev: adapter->netdev,
265 format: "%s: Set affinity failed, queues not allocated\n",
266 __func__);
267 return;
268 }
269
270 total_queues = num_rxqs + num_txqs;
271 num_cpu = num_online_cpus();
272 /* number of cpu's assigned per irq */
273 stride = max_t(int, num_cpu / total_queues, 1);
274 /* number of leftover cpu's */
275 stragglers = num_cpu >= total_queues ? num_cpu % total_queues : 0;
276 /* next available cpu to assign irq to */
277 cpu = cpumask_next(n: -1, cpu_online_mask);
278
279 for (i = 0; i < total_queues; i++) {
280 is_rx_queue = false;
281 /* balance core load by alternating rx and tx assignments
282 * ex: TX0 -> RX0 -> TX1 -> RX1 etc.
283 */
284 if ((i % 2 == 1 && i_rxqs < num_rxqs) || i_txqs == num_txqs) {
285 queue = rxqs[i_rxqs++];
286 is_rx_queue = true;
287 } else {
288 queue = txqs[i_txqs++];
289 }
290
291 rc = ibmvnic_set_queue_affinity(queue, cpu: &cpu, stragglers: &stragglers,
292 stride);
293 if (rc)
294 goto out;
295
296 if (!queue || is_rx_queue)
297 continue;
298
299 rc = __netif_set_xps_queue(dev: adapter->netdev,
300 cpumask_bits(queue->affinity_mask),
301 index: i_txqs - 1, type: XPS_CPUS);
302 if (rc)
303 netdev_warn(dev: adapter->netdev, format: "%s: Set XPS on queue %d failed, rc = %d.\n",
304 __func__, i_txqs - 1, rc);
305 }
306
307out:
308 if (rc) {
309 netdev_warn(dev: adapter->netdev,
310 format: "%s: Set affinity failed, queue addr = %p, IRQ = %d, rc = %d.\n",
311 __func__, queue, queue->irq, rc);
312 ibmvnic_clean_affinity(adapter);
313 }
314}
315
316static int ibmvnic_cpu_online(unsigned int cpu, struct hlist_node *node)
317{
318 struct ibmvnic_adapter *adapter;
319
320 adapter = hlist_entry_safe(node, struct ibmvnic_adapter, node);
321 ibmvnic_set_affinity(adapter);
322 return 0;
323}
324
325static int ibmvnic_cpu_dead(unsigned int cpu, struct hlist_node *node)
326{
327 struct ibmvnic_adapter *adapter;
328
329 adapter = hlist_entry_safe(node, struct ibmvnic_adapter, node_dead);
330 ibmvnic_set_affinity(adapter);
331 return 0;
332}
333
334static int ibmvnic_cpu_down_prep(unsigned int cpu, struct hlist_node *node)
335{
336 struct ibmvnic_adapter *adapter;
337
338 adapter = hlist_entry_safe(node, struct ibmvnic_adapter, node);
339 ibmvnic_clean_affinity(adapter);
340 return 0;
341}
342
343static enum cpuhp_state ibmvnic_online;
344
345static int ibmvnic_cpu_notif_add(struct ibmvnic_adapter *adapter)
346{
347 int ret;
348
349 ret = cpuhp_state_add_instance_nocalls(state: ibmvnic_online, node: &adapter->node);
350 if (ret)
351 return ret;
352 ret = cpuhp_state_add_instance_nocalls(state: CPUHP_IBMVNIC_DEAD,
353 node: &adapter->node_dead);
354 if (!ret)
355 return ret;
356 cpuhp_state_remove_instance_nocalls(state: ibmvnic_online, node: &adapter->node);
357 return ret;
358}
359
360static void ibmvnic_cpu_notif_remove(struct ibmvnic_adapter *adapter)
361{
362 cpuhp_state_remove_instance_nocalls(state: ibmvnic_online, node: &adapter->node);
363 cpuhp_state_remove_instance_nocalls(state: CPUHP_IBMVNIC_DEAD,
364 node: &adapter->node_dead);
365}
366
367static long h_reg_sub_crq(unsigned long unit_address, unsigned long token,
368 unsigned long length, unsigned long *number,
369 unsigned long *irq)
370{
371 unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
372 long rc;
373
374 rc = plpar_hcall(H_REG_SUB_CRQ, retbuf, unit_address, token, length);
375 *number = retbuf[0];
376 *irq = retbuf[1];
377
378 return rc;
379}
380
381/**
382 * ibmvnic_wait_for_completion - Check device state and wait for completion
383 * @adapter: private device data
384 * @comp_done: completion structure to wait for
385 * @timeout: time to wait in milliseconds
386 *
387 * Wait for a completion signal or until the timeout limit is reached
388 * while checking that the device is still active.
389 */
390static int ibmvnic_wait_for_completion(struct ibmvnic_adapter *adapter,
391 struct completion *comp_done,
392 unsigned long timeout)
393{
394 struct net_device *netdev;
395 unsigned long div_timeout;
396 u8 retry;
397
398 netdev = adapter->netdev;
399 retry = 5;
400 div_timeout = msecs_to_jiffies(m: timeout / retry);
401 while (true) {
402 if (!adapter->crq.active) {
403 netdev_err(dev: netdev, format: "Device down!\n");
404 return -ENODEV;
405 }
406 if (!retry--)
407 break;
408 if (wait_for_completion_timeout(x: comp_done, timeout: div_timeout))
409 return 0;
410 }
411 netdev_err(dev: netdev, format: "Operation timed out.\n");
412 return -ETIMEDOUT;
413}
414
415/**
416 * reuse_ltb() - Check if a long term buffer can be reused
417 * @ltb: The long term buffer to be checked
418 * @size: The size of the long term buffer.
419 *
420 * An LTB can be reused unless its size has changed.
421 *
422 * Return: Return true if the LTB can be reused, false otherwise.
423 */
424static bool reuse_ltb(struct ibmvnic_long_term_buff *ltb, int size)
425{
426 return (ltb->buff && ltb->size == size);
427}
428
429/**
430 * alloc_long_term_buff() - Allocate a long term buffer (LTB)
431 *
432 * @adapter: ibmvnic adapter associated to the LTB
433 * @ltb: container object for the LTB
434 * @size: size of the LTB
435 *
436 * Allocate an LTB of the specified size and notify VIOS.
437 *
438 * If the given @ltb already has the correct size, reuse it. Otherwise if
439 * its non-NULL, free it. Then allocate a new one of the correct size.
440 * Notify the VIOS either way since we may now be working with a new VIOS.
441 *
442 * Allocating larger chunks of memory during resets, specially LPM or under
443 * low memory situations can cause resets to fail/timeout and for LPAR to
444 * lose connectivity. So hold onto the LTB even if we fail to communicate
445 * with the VIOS and reuse it on next open. Free LTB when adapter is closed.
446 *
447 * Return: 0 if we were able to allocate the LTB and notify the VIOS and
448 * a negative value otherwise.
449 */
450static int alloc_long_term_buff(struct ibmvnic_adapter *adapter,
451 struct ibmvnic_long_term_buff *ltb, int size)
452{
453 struct device *dev = &adapter->vdev->dev;
454 u64 prev = 0;
455 int rc;
456
457 if (!reuse_ltb(ltb, size)) {
458 dev_dbg(dev,
459 "LTB size changed from 0x%llx to 0x%x, reallocating\n",
460 ltb->size, size);
461 prev = ltb->size;
462 free_long_term_buff(adapter, ltb);
463 }
464
465 if (ltb->buff) {
466 dev_dbg(dev, "Reusing LTB [map %d, size 0x%llx]\n",
467 ltb->map_id, ltb->size);
468 } else {
469 ltb->buff = dma_alloc_coherent(dev, size, dma_handle: &ltb->addr,
470 GFP_KERNEL);
471 if (!ltb->buff) {
472 dev_err(dev, "Couldn't alloc long term buffer\n");
473 return -ENOMEM;
474 }
475 ltb->size = size;
476
477 ltb->map_id = find_first_zero_bit(addr: adapter->map_ids,
478 MAX_MAP_ID);
479 bitmap_set(map: adapter->map_ids, start: ltb->map_id, nbits: 1);
480
481 dev_dbg(dev,
482 "Allocated new LTB [map %d, size 0x%llx was 0x%llx]\n",
483 ltb->map_id, ltb->size, prev);
484 }
485
486 /* Ensure ltb is zeroed - specially when reusing it. */
487 memset(ltb->buff, 0, ltb->size);
488
489 mutex_lock(&adapter->fw_lock);
490 adapter->fw_done_rc = 0;
491 reinit_completion(x: &adapter->fw_done);
492
493 rc = send_request_map(adapter, ltb->addr, ltb->size, ltb->map_id);
494 if (rc) {
495 dev_err(dev, "send_request_map failed, rc = %d\n", rc);
496 goto out;
497 }
498
499 rc = ibmvnic_wait_for_completion(adapter, comp_done: &adapter->fw_done, timeout: 10000);
500 if (rc) {
501 dev_err(dev, "LTB map request aborted or timed out, rc = %d\n",
502 rc);
503 goto out;
504 }
505
506 if (adapter->fw_done_rc) {
507 dev_err(dev, "Couldn't map LTB, rc = %d\n",
508 adapter->fw_done_rc);
509 rc = -EIO;
510 goto out;
511 }
512 rc = 0;
513out:
514 /* don't free LTB on communication error - see function header */
515 mutex_unlock(lock: &adapter->fw_lock);
516 return rc;
517}
518
519static void free_long_term_buff(struct ibmvnic_adapter *adapter,
520 struct ibmvnic_long_term_buff *ltb)
521{
522 struct device *dev = &adapter->vdev->dev;
523
524 if (!ltb->buff)
525 return;
526
527 /* VIOS automatically unmaps the long term buffer at remote
528 * end for the following resets:
529 * FAILOVER, MOBILITY, TIMEOUT.
530 */
531 if (adapter->reset_reason != VNIC_RESET_FAILOVER &&
532 adapter->reset_reason != VNIC_RESET_MOBILITY &&
533 adapter->reset_reason != VNIC_RESET_TIMEOUT)
534 send_request_unmap(adapter, ltb->map_id);
535
536 dma_free_coherent(dev, size: ltb->size, cpu_addr: ltb->buff, dma_handle: ltb->addr);
537
538 ltb->buff = NULL;
539 /* mark this map_id free */
540 bitmap_clear(map: adapter->map_ids, start: ltb->map_id, nbits: 1);
541 ltb->map_id = 0;
542}
543
544/**
545 * free_ltb_set - free the given set of long term buffers (LTBS)
546 * @adapter: The ibmvnic adapter containing this ltb set
547 * @ltb_set: The ltb_set to be freed
548 *
549 * Free the set of LTBs in the given set.
550 */
551
552static void free_ltb_set(struct ibmvnic_adapter *adapter,
553 struct ibmvnic_ltb_set *ltb_set)
554{
555 int i;
556
557 for (i = 0; i < ltb_set->num_ltbs; i++)
558 free_long_term_buff(adapter, ltb: &ltb_set->ltbs[i]);
559
560 kfree(objp: ltb_set->ltbs);
561 ltb_set->ltbs = NULL;
562 ltb_set->num_ltbs = 0;
563}
564
565/**
566 * alloc_ltb_set() - Allocate a set of long term buffers (LTBs)
567 *
568 * @adapter: ibmvnic adapter associated to the LTB
569 * @ltb_set: container object for the set of LTBs
570 * @num_buffs: Number of buffers in the LTB
571 * @buff_size: Size of each buffer in the LTB
572 *
573 * Allocate a set of LTBs to accommodate @num_buffs buffers of @buff_size
574 * each. We currently cap size each LTB to IBMVNIC_ONE_LTB_SIZE. If the
575 * new set of LTBs have fewer LTBs than the old set, free the excess LTBs.
576 * If new set needs more than in old set, allocate the remaining ones.
577 * Try and reuse as many LTBs as possible and avoid reallocation.
578 *
579 * Any changes to this allocation strategy must be reflected in
580 * map_rxpool_buff_to_ltb() and map_txpool_buff_to_ltb().
581 */
582static int alloc_ltb_set(struct ibmvnic_adapter *adapter,
583 struct ibmvnic_ltb_set *ltb_set, int num_buffs,
584 int buff_size)
585{
586 struct device *dev = &adapter->vdev->dev;
587 struct ibmvnic_ltb_set old_set;
588 struct ibmvnic_ltb_set new_set;
589 int rem_size;
590 int tot_size; /* size of all ltbs */
591 int ltb_size; /* size of one ltb */
592 int nltbs;
593 int rc;
594 int n;
595 int i;
596
597 dev_dbg(dev, "%s() num_buffs %d, buff_size %d\n", __func__, num_buffs,
598 buff_size);
599
600 ltb_size = rounddown(IBMVNIC_ONE_LTB_SIZE, buff_size);
601 tot_size = num_buffs * buff_size;
602
603 if (ltb_size > tot_size)
604 ltb_size = tot_size;
605
606 nltbs = tot_size / ltb_size;
607 if (tot_size % ltb_size)
608 nltbs++;
609
610 old_set = *ltb_set;
611
612 if (old_set.num_ltbs == nltbs) {
613 new_set = old_set;
614 } else {
615 int tmp = nltbs * sizeof(struct ibmvnic_long_term_buff);
616
617 new_set.ltbs = kzalloc(size: tmp, GFP_KERNEL);
618 if (!new_set.ltbs)
619 return -ENOMEM;
620
621 new_set.num_ltbs = nltbs;
622
623 /* Free any excess ltbs in old set */
624 for (i = new_set.num_ltbs; i < old_set.num_ltbs; i++)
625 free_long_term_buff(adapter, ltb: &old_set.ltbs[i]);
626
627 /* Copy remaining ltbs to new set. All LTBs except the
628 * last one are of the same size. alloc_long_term_buff()
629 * will realloc if the size changes.
630 */
631 n = min(old_set.num_ltbs, new_set.num_ltbs);
632 for (i = 0; i < n; i++)
633 new_set.ltbs[i] = old_set.ltbs[i];
634
635 /* Any additional ltbs in new set will have NULL ltbs for
636 * now and will be allocated in alloc_long_term_buff().
637 */
638
639 /* We no longer need the old_set so free it. Note that we
640 * may have reused some ltbs from old set and freed excess
641 * ltbs above. So we only need to free the container now
642 * not the LTBs themselves. (i.e. dont free_ltb_set()!)
643 */
644 kfree(objp: old_set.ltbs);
645 old_set.ltbs = NULL;
646 old_set.num_ltbs = 0;
647
648 /* Install the new set. If allocations fail below, we will
649 * retry later and know what size LTBs we need.
650 */
651 *ltb_set = new_set;
652 }
653
654 i = 0;
655 rem_size = tot_size;
656 while (rem_size) {
657 if (ltb_size > rem_size)
658 ltb_size = rem_size;
659
660 rem_size -= ltb_size;
661
662 rc = alloc_long_term_buff(adapter, ltb: &new_set.ltbs[i], size: ltb_size);
663 if (rc)
664 goto out;
665 i++;
666 }
667
668 WARN_ON(i != new_set.num_ltbs);
669
670 return 0;
671out:
672 /* We may have allocated one/more LTBs before failing and we
673 * want to try and reuse on next reset. So don't free ltb set.
674 */
675 return rc;
676}
677
678/**
679 * map_rxpool_buf_to_ltb - Map given rxpool buffer to offset in an LTB.
680 * @rxpool: The receive buffer pool containing buffer
681 * @bufidx: Index of buffer in rxpool
682 * @ltbp: (Output) pointer to the long term buffer containing the buffer
683 * @offset: (Output) offset of buffer in the LTB from @ltbp
684 *
685 * Map the given buffer identified by [rxpool, bufidx] to an LTB in the
686 * pool and its corresponding offset. Assume for now that each LTB is of
687 * different size but could possibly be optimized based on the allocation
688 * strategy in alloc_ltb_set().
689 */
690static void map_rxpool_buf_to_ltb(struct ibmvnic_rx_pool *rxpool,
691 unsigned int bufidx,
692 struct ibmvnic_long_term_buff **ltbp,
693 unsigned int *offset)
694{
695 struct ibmvnic_long_term_buff *ltb;
696 int nbufs; /* # of buffers in one ltb */
697 int i;
698
699 WARN_ON(bufidx >= rxpool->size);
700
701 for (i = 0; i < rxpool->ltb_set.num_ltbs; i++) {
702 ltb = &rxpool->ltb_set.ltbs[i];
703 nbufs = ltb->size / rxpool->buff_size;
704 if (bufidx < nbufs)
705 break;
706 bufidx -= nbufs;
707 }
708
709 *ltbp = ltb;
710 *offset = bufidx * rxpool->buff_size;
711}
712
713/**
714 * map_txpool_buf_to_ltb - Map given txpool buffer to offset in an LTB.
715 * @txpool: The transmit buffer pool containing buffer
716 * @bufidx: Index of buffer in txpool
717 * @ltbp: (Output) pointer to the long term buffer (LTB) containing the buffer
718 * @offset: (Output) offset of buffer in the LTB from @ltbp
719 *
720 * Map the given buffer identified by [txpool, bufidx] to an LTB in the
721 * pool and its corresponding offset.
722 */
723static void map_txpool_buf_to_ltb(struct ibmvnic_tx_pool *txpool,
724 unsigned int bufidx,
725 struct ibmvnic_long_term_buff **ltbp,
726 unsigned int *offset)
727{
728 struct ibmvnic_long_term_buff *ltb;
729 int nbufs; /* # of buffers in one ltb */
730 int i;
731
732 WARN_ON_ONCE(bufidx >= txpool->num_buffers);
733
734 for (i = 0; i < txpool->ltb_set.num_ltbs; i++) {
735 ltb = &txpool->ltb_set.ltbs[i];
736 nbufs = ltb->size / txpool->buf_size;
737 if (bufidx < nbufs)
738 break;
739 bufidx -= nbufs;
740 }
741
742 *ltbp = ltb;
743 *offset = bufidx * txpool->buf_size;
744}
745
746static void deactivate_rx_pools(struct ibmvnic_adapter *adapter)
747{
748 int i;
749
750 for (i = 0; i < adapter->num_active_rx_pools; i++)
751 adapter->rx_pool[i].active = 0;
752}
753
754static void replenish_rx_pool(struct ibmvnic_adapter *adapter,
755 struct ibmvnic_rx_pool *pool)
756{
757 int count = pool->size - atomic_read(v: &pool->available);
758 u64 handle = adapter->rx_scrq[pool->index]->handle;
759 struct device *dev = &adapter->vdev->dev;
760 struct ibmvnic_ind_xmit_queue *ind_bufp;
761 struct ibmvnic_sub_crq_queue *rx_scrq;
762 struct ibmvnic_long_term_buff *ltb;
763 union sub_crq *sub_crq;
764 int buffers_added = 0;
765 unsigned long lpar_rc;
766 struct sk_buff *skb;
767 unsigned int offset;
768 dma_addr_t dma_addr;
769 unsigned char *dst;
770 int shift = 0;
771 int bufidx;
772 int i;
773
774 if (!pool->active)
775 return;
776
777 rx_scrq = adapter->rx_scrq[pool->index];
778 ind_bufp = &rx_scrq->ind_buf;
779
780 /* netdev_skb_alloc() could have failed after we saved a few skbs
781 * in the indir_buf and we would not have sent them to VIOS yet.
782 * To account for them, start the loop at ind_bufp->index rather
783 * than 0. If we pushed all the skbs to VIOS, ind_bufp->index will
784 * be 0.
785 */
786 for (i = ind_bufp->index; i < count; ++i) {
787 bufidx = pool->free_map[pool->next_free];
788
789 /* We maybe reusing the skb from earlier resets. Allocate
790 * only if necessary. But since the LTB may have changed
791 * during reset (see init_rx_pools()), update LTB below
792 * even if reusing skb.
793 */
794 skb = pool->rx_buff[bufidx].skb;
795 if (!skb) {
796 skb = netdev_alloc_skb(dev: adapter->netdev,
797 length: pool->buff_size);
798 if (!skb) {
799 dev_err(dev, "Couldn't replenish rx buff\n");
800 adapter->replenish_no_mem++;
801 break;
802 }
803 }
804
805 pool->free_map[pool->next_free] = IBMVNIC_INVALID_MAP;
806 pool->next_free = (pool->next_free + 1) % pool->size;
807
808 /* Copy the skb to the long term mapped DMA buffer */
809 map_rxpool_buf_to_ltb(rxpool: pool, bufidx, ltbp: &ltb, offset: &offset);
810 dst = ltb->buff + offset;
811 memset(dst, 0, pool->buff_size);
812 dma_addr = ltb->addr + offset;
813
814 /* add the skb to an rx_buff in the pool */
815 pool->rx_buff[bufidx].data = dst;
816 pool->rx_buff[bufidx].dma = dma_addr;
817 pool->rx_buff[bufidx].skb = skb;
818 pool->rx_buff[bufidx].pool_index = pool->index;
819 pool->rx_buff[bufidx].size = pool->buff_size;
820
821 /* queue the rx_buff for the next send_subcrq_indirect */
822 sub_crq = &ind_bufp->indir_arr[ind_bufp->index++];
823 memset(sub_crq, 0, sizeof(*sub_crq));
824 sub_crq->rx_add.first = IBMVNIC_CRQ_CMD;
825 sub_crq->rx_add.correlator =
826 cpu_to_be64((u64)&pool->rx_buff[bufidx]);
827 sub_crq->rx_add.ioba = cpu_to_be32(dma_addr);
828 sub_crq->rx_add.map_id = ltb->map_id;
829
830 /* The length field of the sCRQ is defined to be 24 bits so the
831 * buffer size needs to be left shifted by a byte before it is
832 * converted to big endian to prevent the last byte from being
833 * truncated.
834 */
835#ifdef __LITTLE_ENDIAN__
836 shift = 8;
837#endif
838 sub_crq->rx_add.len = cpu_to_be32(pool->buff_size << shift);
839
840 /* if send_subcrq_indirect queue is full, flush to VIOS */
841 if (ind_bufp->index == IBMVNIC_MAX_IND_DESCS ||
842 i == count - 1) {
843 lpar_rc =
844 send_subcrq_indirect(adapter, handle,
845 (u64)ind_bufp->indir_dma,
846 (u64)ind_bufp->index);
847 if (lpar_rc != H_SUCCESS)
848 goto failure;
849 buffers_added += ind_bufp->index;
850 adapter->replenish_add_buff_success += ind_bufp->index;
851 ind_bufp->index = 0;
852 }
853 }
854 atomic_add(i: buffers_added, v: &pool->available);
855 return;
856
857failure:
858 if (lpar_rc != H_PARAMETER && lpar_rc != H_CLOSED)
859 dev_err_ratelimited(dev, "rx: replenish packet buffer failed\n");
860 for (i = ind_bufp->index - 1; i >= 0; --i) {
861 struct ibmvnic_rx_buff *rx_buff;
862
863 pool->next_free = pool->next_free == 0 ?
864 pool->size - 1 : pool->next_free - 1;
865 sub_crq = &ind_bufp->indir_arr[i];
866 rx_buff = (struct ibmvnic_rx_buff *)
867 be64_to_cpu(sub_crq->rx_add.correlator);
868 bufidx = (int)(rx_buff - pool->rx_buff);
869 pool->free_map[pool->next_free] = bufidx;
870 dev_kfree_skb_any(skb: pool->rx_buff[bufidx].skb);
871 pool->rx_buff[bufidx].skb = NULL;
872 }
873 adapter->replenish_add_buff_failure += ind_bufp->index;
874 atomic_add(i: buffers_added, v: &pool->available);
875 ind_bufp->index = 0;
876 if (lpar_rc == H_CLOSED || adapter->failover_pending) {
877 /* Disable buffer pool replenishment and report carrier off if
878 * queue is closed or pending failover.
879 * Firmware guarantees that a signal will be sent to the
880 * driver, triggering a reset.
881 */
882 deactivate_rx_pools(adapter);
883 netif_carrier_off(dev: adapter->netdev);
884 }
885}
886
887static void replenish_pools(struct ibmvnic_adapter *adapter)
888{
889 int i;
890
891 adapter->replenish_task_cycles++;
892 for (i = 0; i < adapter->num_active_rx_pools; i++) {
893 if (adapter->rx_pool[i].active)
894 replenish_rx_pool(adapter, pool: &adapter->rx_pool[i]);
895 }
896
897 netdev_dbg(adapter->netdev, "Replenished %d pools\n", i);
898}
899
900static void release_stats_buffers(struct ibmvnic_adapter *adapter)
901{
902 kfree(objp: adapter->tx_stats_buffers);
903 kfree(objp: adapter->rx_stats_buffers);
904 adapter->tx_stats_buffers = NULL;
905 adapter->rx_stats_buffers = NULL;
906}
907
908static int init_stats_buffers(struct ibmvnic_adapter *adapter)
909{
910 adapter->tx_stats_buffers =
911 kcalloc(IBMVNIC_MAX_QUEUES,
912 size: sizeof(struct ibmvnic_tx_queue_stats),
913 GFP_KERNEL);
914 if (!adapter->tx_stats_buffers)
915 return -ENOMEM;
916
917 adapter->rx_stats_buffers =
918 kcalloc(IBMVNIC_MAX_QUEUES,
919 size: sizeof(struct ibmvnic_rx_queue_stats),
920 GFP_KERNEL);
921 if (!adapter->rx_stats_buffers)
922 return -ENOMEM;
923
924 return 0;
925}
926
927static void release_stats_token(struct ibmvnic_adapter *adapter)
928{
929 struct device *dev = &adapter->vdev->dev;
930
931 if (!adapter->stats_token)
932 return;
933
934 dma_unmap_single(dev, adapter->stats_token,
935 sizeof(struct ibmvnic_statistics),
936 DMA_FROM_DEVICE);
937 adapter->stats_token = 0;
938}
939
940static int init_stats_token(struct ibmvnic_adapter *adapter)
941{
942 struct device *dev = &adapter->vdev->dev;
943 dma_addr_t stok;
944 int rc;
945
946 stok = dma_map_single(dev, &adapter->stats,
947 sizeof(struct ibmvnic_statistics),
948 DMA_FROM_DEVICE);
949 rc = dma_mapping_error(dev, dma_addr: stok);
950 if (rc) {
951 dev_err(dev, "Couldn't map stats buffer, rc = %d\n", rc);
952 return rc;
953 }
954
955 adapter->stats_token = stok;
956 netdev_dbg(adapter->netdev, "Stats token initialized (%llx)\n", stok);
957 return 0;
958}
959
960/**
961 * release_rx_pools() - Release any rx pools attached to @adapter.
962 * @adapter: ibmvnic adapter
963 *
964 * Safe to call this multiple times - even if no pools are attached.
965 */
966static void release_rx_pools(struct ibmvnic_adapter *adapter)
967{
968 struct ibmvnic_rx_pool *rx_pool;
969 int i, j;
970
971 if (!adapter->rx_pool)
972 return;
973
974 for (i = 0; i < adapter->num_active_rx_pools; i++) {
975 rx_pool = &adapter->rx_pool[i];
976
977 netdev_dbg(adapter->netdev, "Releasing rx_pool[%d]\n", i);
978
979 kfree(objp: rx_pool->free_map);
980
981 free_ltb_set(adapter, ltb_set: &rx_pool->ltb_set);
982
983 if (!rx_pool->rx_buff)
984 continue;
985
986 for (j = 0; j < rx_pool->size; j++) {
987 if (rx_pool->rx_buff[j].skb) {
988 dev_kfree_skb_any(skb: rx_pool->rx_buff[j].skb);
989 rx_pool->rx_buff[j].skb = NULL;
990 }
991 }
992
993 kfree(objp: rx_pool->rx_buff);
994 }
995
996 kfree(objp: adapter->rx_pool);
997 adapter->rx_pool = NULL;
998 adapter->num_active_rx_pools = 0;
999 adapter->prev_rx_pool_size = 0;
1000}
1001
1002/**
1003 * reuse_rx_pools() - Check if the existing rx pools can be reused.
1004 * @adapter: ibmvnic adapter
1005 *
1006 * Check if the existing rx pools in the adapter can be reused. The
1007 * pools can be reused if the pool parameters (number of pools,
1008 * number of buffers in the pool and size of each buffer) have not
1009 * changed.
1010 *
1011 * NOTE: This assumes that all pools have the same number of buffers
1012 * which is the case currently. If that changes, we must fix this.
1013 *
1014 * Return: true if the rx pools can be reused, false otherwise.
1015 */
1016static bool reuse_rx_pools(struct ibmvnic_adapter *adapter)
1017{
1018 u64 old_num_pools, new_num_pools;
1019 u64 old_pool_size, new_pool_size;
1020 u64 old_buff_size, new_buff_size;
1021
1022 if (!adapter->rx_pool)
1023 return false;
1024
1025 old_num_pools = adapter->num_active_rx_pools;
1026 new_num_pools = adapter->req_rx_queues;
1027
1028 old_pool_size = adapter->prev_rx_pool_size;
1029 new_pool_size = adapter->req_rx_add_entries_per_subcrq;
1030
1031 old_buff_size = adapter->prev_rx_buf_sz;
1032 new_buff_size = adapter->cur_rx_buf_sz;
1033
1034 if (old_buff_size != new_buff_size ||
1035 old_num_pools != new_num_pools ||
1036 old_pool_size != new_pool_size)
1037 return false;
1038
1039 return true;
1040}
1041
1042/**
1043 * init_rx_pools(): Initialize the set of receiver pools in the adapter.
1044 * @netdev: net device associated with the vnic interface
1045 *
1046 * Initialize the set of receiver pools in the ibmvnic adapter associated
1047 * with the net_device @netdev. If possible, reuse the existing rx pools.
1048 * Otherwise free any existing pools and allocate a new set of pools
1049 * before initializing them.
1050 *
1051 * Return: 0 on success and negative value on error.
1052 */
1053static int init_rx_pools(struct net_device *netdev)
1054{
1055 struct ibmvnic_adapter *adapter = netdev_priv(dev: netdev);
1056 struct device *dev = &adapter->vdev->dev;
1057 struct ibmvnic_rx_pool *rx_pool;
1058 u64 num_pools;
1059 u64 pool_size; /* # of buffers in one pool */
1060 u64 buff_size;
1061 int i, j, rc;
1062
1063 pool_size = adapter->req_rx_add_entries_per_subcrq;
1064 num_pools = adapter->req_rx_queues;
1065 buff_size = adapter->cur_rx_buf_sz;
1066
1067 if (reuse_rx_pools(adapter)) {
1068 dev_dbg(dev, "Reusing rx pools\n");
1069 goto update_ltb;
1070 }
1071
1072 /* Allocate/populate the pools. */
1073 release_rx_pools(adapter);
1074
1075 adapter->rx_pool = kcalloc(n: num_pools,
1076 size: sizeof(struct ibmvnic_rx_pool),
1077 GFP_KERNEL);
1078 if (!adapter->rx_pool) {
1079 dev_err(dev, "Failed to allocate rx pools\n");
1080 return -ENOMEM;
1081 }
1082
1083 /* Set num_active_rx_pools early. If we fail below after partial
1084 * allocation, release_rx_pools() will know how many to look for.
1085 */
1086 adapter->num_active_rx_pools = num_pools;
1087
1088 for (i = 0; i < num_pools; i++) {
1089 rx_pool = &adapter->rx_pool[i];
1090
1091 netdev_dbg(adapter->netdev,
1092 "Initializing rx_pool[%d], %lld buffs, %lld bytes each\n",
1093 i, pool_size, buff_size);
1094
1095 rx_pool->size = pool_size;
1096 rx_pool->index = i;
1097 rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
1098
1099 rx_pool->free_map = kcalloc(n: rx_pool->size, size: sizeof(int),
1100 GFP_KERNEL);
1101 if (!rx_pool->free_map) {
1102 dev_err(dev, "Couldn't alloc free_map %d\n", i);
1103 rc = -ENOMEM;
1104 goto out_release;
1105 }
1106
1107 rx_pool->rx_buff = kcalloc(n: rx_pool->size,
1108 size: sizeof(struct ibmvnic_rx_buff),
1109 GFP_KERNEL);
1110 if (!rx_pool->rx_buff) {
1111 dev_err(dev, "Couldn't alloc rx buffers\n");
1112 rc = -ENOMEM;
1113 goto out_release;
1114 }
1115 }
1116
1117 adapter->prev_rx_pool_size = pool_size;
1118 adapter->prev_rx_buf_sz = adapter->cur_rx_buf_sz;
1119
1120update_ltb:
1121 for (i = 0; i < num_pools; i++) {
1122 rx_pool = &adapter->rx_pool[i];
1123 dev_dbg(dev, "Updating LTB for rx pool %d [%d, %d]\n",
1124 i, rx_pool->size, rx_pool->buff_size);
1125
1126 rc = alloc_ltb_set(adapter, ltb_set: &rx_pool->ltb_set,
1127 num_buffs: rx_pool->size, buff_size: rx_pool->buff_size);
1128 if (rc)
1129 goto out;
1130
1131 for (j = 0; j < rx_pool->size; ++j) {
1132 struct ibmvnic_rx_buff *rx_buff;
1133
1134 rx_pool->free_map[j] = j;
1135
1136 /* NOTE: Don't clear rx_buff->skb here - will leak
1137 * memory! replenish_rx_pool() will reuse skbs or
1138 * allocate as necessary.
1139 */
1140 rx_buff = &rx_pool->rx_buff[j];
1141 rx_buff->dma = 0;
1142 rx_buff->data = 0;
1143 rx_buff->size = 0;
1144 rx_buff->pool_index = 0;
1145 }
1146
1147 /* Mark pool "empty" so replenish_rx_pools() will
1148 * update the LTB info for each buffer
1149 */
1150 atomic_set(v: &rx_pool->available, i: 0);
1151 rx_pool->next_alloc = 0;
1152 rx_pool->next_free = 0;
1153 /* replenish_rx_pool() may have called deactivate_rx_pools()
1154 * on failover. Ensure pool is active now.
1155 */
1156 rx_pool->active = 1;
1157 }
1158 return 0;
1159out_release:
1160 release_rx_pools(adapter);
1161out:
1162 /* We failed to allocate one or more LTBs or map them on the VIOS.
1163 * Hold onto the pools and any LTBs that we did allocate/map.
1164 */
1165 return rc;
1166}
1167
1168static void release_vpd_data(struct ibmvnic_adapter *adapter)
1169{
1170 if (!adapter->vpd)
1171 return;
1172
1173 kfree(objp: adapter->vpd->buff);
1174 kfree(objp: adapter->vpd);
1175
1176 adapter->vpd = NULL;
1177}
1178
1179static void release_one_tx_pool(struct ibmvnic_adapter *adapter,
1180 struct ibmvnic_tx_pool *tx_pool)
1181{
1182 kfree(objp: tx_pool->tx_buff);
1183 kfree(objp: tx_pool->free_map);
1184 free_ltb_set(adapter, ltb_set: &tx_pool->ltb_set);
1185}
1186
1187/**
1188 * release_tx_pools() - Release any tx pools attached to @adapter.
1189 * @adapter: ibmvnic adapter
1190 *
1191 * Safe to call this multiple times - even if no pools are attached.
1192 */
1193static void release_tx_pools(struct ibmvnic_adapter *adapter)
1194{
1195 int i;
1196
1197 /* init_tx_pools() ensures that ->tx_pool and ->tso_pool are
1198 * both NULL or both non-NULL. So we only need to check one.
1199 */
1200 if (!adapter->tx_pool)
1201 return;
1202
1203 for (i = 0; i < adapter->num_active_tx_pools; i++) {
1204 release_one_tx_pool(adapter, tx_pool: &adapter->tx_pool[i]);
1205 release_one_tx_pool(adapter, tx_pool: &adapter->tso_pool[i]);
1206 }
1207
1208 kfree(objp: adapter->tx_pool);
1209 adapter->tx_pool = NULL;
1210 kfree(objp: adapter->tso_pool);
1211 adapter->tso_pool = NULL;
1212 adapter->num_active_tx_pools = 0;
1213 adapter->prev_tx_pool_size = 0;
1214}
1215
1216static int init_one_tx_pool(struct net_device *netdev,
1217 struct ibmvnic_tx_pool *tx_pool,
1218 int pool_size, int buf_size)
1219{
1220 int i;
1221
1222 tx_pool->tx_buff = kcalloc(n: pool_size,
1223 size: sizeof(struct ibmvnic_tx_buff),
1224 GFP_KERNEL);
1225 if (!tx_pool->tx_buff)
1226 return -ENOMEM;
1227
1228 tx_pool->free_map = kcalloc(n: pool_size, size: sizeof(int), GFP_KERNEL);
1229 if (!tx_pool->free_map) {
1230 kfree(objp: tx_pool->tx_buff);
1231 tx_pool->tx_buff = NULL;
1232 return -ENOMEM;
1233 }
1234
1235 for (i = 0; i < pool_size; i++)
1236 tx_pool->free_map[i] = i;
1237
1238 tx_pool->consumer_index = 0;
1239 tx_pool->producer_index = 0;
1240 tx_pool->num_buffers = pool_size;
1241 tx_pool->buf_size = buf_size;
1242
1243 return 0;
1244}
1245
1246/**
1247 * reuse_tx_pools() - Check if the existing tx pools can be reused.
1248 * @adapter: ibmvnic adapter
1249 *
1250 * Check if the existing tx pools in the adapter can be reused. The
1251 * pools can be reused if the pool parameters (number of pools,
1252 * number of buffers in the pool and mtu) have not changed.
1253 *
1254 * NOTE: This assumes that all pools have the same number of buffers
1255 * which is the case currently. If that changes, we must fix this.
1256 *
1257 * Return: true if the tx pools can be reused, false otherwise.
1258 */
1259static bool reuse_tx_pools(struct ibmvnic_adapter *adapter)
1260{
1261 u64 old_num_pools, new_num_pools;
1262 u64 old_pool_size, new_pool_size;
1263 u64 old_mtu, new_mtu;
1264
1265 if (!adapter->tx_pool)
1266 return false;
1267
1268 old_num_pools = adapter->num_active_tx_pools;
1269 new_num_pools = adapter->num_active_tx_scrqs;
1270 old_pool_size = adapter->prev_tx_pool_size;
1271 new_pool_size = adapter->req_tx_entries_per_subcrq;
1272 old_mtu = adapter->prev_mtu;
1273 new_mtu = adapter->req_mtu;
1274
1275 if (old_mtu != new_mtu ||
1276 old_num_pools != new_num_pools ||
1277 old_pool_size != new_pool_size)
1278 return false;
1279
1280 return true;
1281}
1282
1283/**
1284 * init_tx_pools(): Initialize the set of transmit pools in the adapter.
1285 * @netdev: net device associated with the vnic interface
1286 *
1287 * Initialize the set of transmit pools in the ibmvnic adapter associated
1288 * with the net_device @netdev. If possible, reuse the existing tx pools.
1289 * Otherwise free any existing pools and allocate a new set of pools
1290 * before initializing them.
1291 *
1292 * Return: 0 on success and negative value on error.
1293 */
1294static int init_tx_pools(struct net_device *netdev)
1295{
1296 struct ibmvnic_adapter *adapter = netdev_priv(dev: netdev);
1297 struct device *dev = &adapter->vdev->dev;
1298 int num_pools;
1299 u64 pool_size; /* # of buffers in pool */
1300 u64 buff_size;
1301 int i, j, rc;
1302
1303 num_pools = adapter->req_tx_queues;
1304
1305 /* We must notify the VIOS about the LTB on all resets - but we only
1306 * need to alloc/populate pools if either the number of buffers or
1307 * size of each buffer in the pool has changed.
1308 */
1309 if (reuse_tx_pools(adapter)) {
1310 netdev_dbg(netdev, "Reusing tx pools\n");
1311 goto update_ltb;
1312 }
1313
1314 /* Allocate/populate the pools. */
1315 release_tx_pools(adapter);
1316
1317 pool_size = adapter->req_tx_entries_per_subcrq;
1318 num_pools = adapter->num_active_tx_scrqs;
1319
1320 adapter->tx_pool = kcalloc(n: num_pools,
1321 size: sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
1322 if (!adapter->tx_pool)
1323 return -ENOMEM;
1324
1325 adapter->tso_pool = kcalloc(n: num_pools,
1326 size: sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
1327 /* To simplify release_tx_pools() ensure that ->tx_pool and
1328 * ->tso_pool are either both NULL or both non-NULL.
1329 */
1330 if (!adapter->tso_pool) {
1331 kfree(objp: adapter->tx_pool);
1332 adapter->tx_pool = NULL;
1333 return -ENOMEM;
1334 }
1335
1336 /* Set num_active_tx_pools early. If we fail below after partial
1337 * allocation, release_tx_pools() will know how many to look for.
1338 */
1339 adapter->num_active_tx_pools = num_pools;
1340
1341 buff_size = adapter->req_mtu + VLAN_HLEN;
1342 buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
1343
1344 for (i = 0; i < num_pools; i++) {
1345 dev_dbg(dev, "Init tx pool %d [%llu, %llu]\n",
1346 i, adapter->req_tx_entries_per_subcrq, buff_size);
1347
1348 rc = init_one_tx_pool(netdev, tx_pool: &adapter->tx_pool[i],
1349 pool_size, buf_size: buff_size);
1350 if (rc)
1351 goto out_release;
1352
1353 rc = init_one_tx_pool(netdev, tx_pool: &adapter->tso_pool[i],
1354 IBMVNIC_TSO_BUFS,
1355 IBMVNIC_TSO_BUF_SZ);
1356 if (rc)
1357 goto out_release;
1358 }
1359
1360 adapter->prev_tx_pool_size = pool_size;
1361 adapter->prev_mtu = adapter->req_mtu;
1362
1363update_ltb:
1364 /* NOTE: All tx_pools have the same number of buffers (which is
1365 * same as pool_size). All tso_pools have IBMVNIC_TSO_BUFS
1366 * buffers (see calls init_one_tx_pool() for these).
1367 * For consistency, we use tx_pool->num_buffers and
1368 * tso_pool->num_buffers below.
1369 */
1370 rc = -1;
1371 for (i = 0; i < num_pools; i++) {
1372 struct ibmvnic_tx_pool *tso_pool;
1373 struct ibmvnic_tx_pool *tx_pool;
1374
1375 tx_pool = &adapter->tx_pool[i];
1376
1377 dev_dbg(dev, "Updating LTB for tx pool %d [%d, %d]\n",
1378 i, tx_pool->num_buffers, tx_pool->buf_size);
1379
1380 rc = alloc_ltb_set(adapter, ltb_set: &tx_pool->ltb_set,
1381 num_buffs: tx_pool->num_buffers, buff_size: tx_pool->buf_size);
1382 if (rc)
1383 goto out;
1384
1385 tx_pool->consumer_index = 0;
1386 tx_pool->producer_index = 0;
1387
1388 for (j = 0; j < tx_pool->num_buffers; j++)
1389 tx_pool->free_map[j] = j;
1390
1391 tso_pool = &adapter->tso_pool[i];
1392
1393 dev_dbg(dev, "Updating LTB for tso pool %d [%d, %d]\n",
1394 i, tso_pool->num_buffers, tso_pool->buf_size);
1395
1396 rc = alloc_ltb_set(adapter, ltb_set: &tso_pool->ltb_set,
1397 num_buffs: tso_pool->num_buffers, buff_size: tso_pool->buf_size);
1398 if (rc)
1399 goto out;
1400
1401 tso_pool->consumer_index = 0;
1402 tso_pool->producer_index = 0;
1403
1404 for (j = 0; j < tso_pool->num_buffers; j++)
1405 tso_pool->free_map[j] = j;
1406 }
1407
1408 return 0;
1409out_release:
1410 release_tx_pools(adapter);
1411out:
1412 /* We failed to allocate one or more LTBs or map them on the VIOS.
1413 * Hold onto the pools and any LTBs that we did allocate/map.
1414 */
1415 return rc;
1416}
1417
1418static void ibmvnic_napi_enable(struct ibmvnic_adapter *adapter)
1419{
1420 int i;
1421
1422 if (adapter->napi_enabled)
1423 return;
1424
1425 for (i = 0; i < adapter->req_rx_queues; i++)
1426 napi_enable(n: &adapter->napi[i]);
1427
1428 adapter->napi_enabled = true;
1429}
1430
1431static void ibmvnic_napi_disable(struct ibmvnic_adapter *adapter)
1432{
1433 int i;
1434
1435 if (!adapter->napi_enabled)
1436 return;
1437
1438 for (i = 0; i < adapter->req_rx_queues; i++) {
1439 netdev_dbg(adapter->netdev, "Disabling napi[%d]\n", i);
1440 napi_disable(n: &adapter->napi[i]);
1441 }
1442
1443 adapter->napi_enabled = false;
1444}
1445
1446static int init_napi(struct ibmvnic_adapter *adapter)
1447{
1448 int i;
1449
1450 adapter->napi = kcalloc(n: adapter->req_rx_queues,
1451 size: sizeof(struct napi_struct), GFP_KERNEL);
1452 if (!adapter->napi)
1453 return -ENOMEM;
1454
1455 for (i = 0; i < adapter->req_rx_queues; i++) {
1456 netdev_dbg(adapter->netdev, "Adding napi[%d]\n", i);
1457 netif_napi_add(dev: adapter->netdev, napi: &adapter->napi[i],
1458 poll: ibmvnic_poll);
1459 }
1460
1461 adapter->num_active_rx_napi = adapter->req_rx_queues;
1462 return 0;
1463}
1464
1465static void release_napi(struct ibmvnic_adapter *adapter)
1466{
1467 int i;
1468
1469 if (!adapter->napi)
1470 return;
1471
1472 for (i = 0; i < adapter->num_active_rx_napi; i++) {
1473 netdev_dbg(adapter->netdev, "Releasing napi[%d]\n", i);
1474 netif_napi_del(napi: &adapter->napi[i]);
1475 }
1476
1477 kfree(objp: adapter->napi);
1478 adapter->napi = NULL;
1479 adapter->num_active_rx_napi = 0;
1480 adapter->napi_enabled = false;
1481}
1482
1483static const char *adapter_state_to_string(enum vnic_state state)
1484{
1485 switch (state) {
1486 case VNIC_PROBING:
1487 return "PROBING";
1488 case VNIC_PROBED:
1489 return "PROBED";
1490 case VNIC_OPENING:
1491 return "OPENING";
1492 case VNIC_OPEN:
1493 return "OPEN";
1494 case VNIC_CLOSING:
1495 return "CLOSING";
1496 case VNIC_CLOSED:
1497 return "CLOSED";
1498 case VNIC_REMOVING:
1499 return "REMOVING";
1500 case VNIC_REMOVED:
1501 return "REMOVED";
1502 case VNIC_DOWN:
1503 return "DOWN";
1504 }
1505 return "UNKNOWN";
1506}
1507
1508static int ibmvnic_login(struct net_device *netdev)
1509{
1510 unsigned long flags, timeout = msecs_to_jiffies(m: 20000);
1511 struct ibmvnic_adapter *adapter = netdev_priv(dev: netdev);
1512 int retry_count = 0;
1513 int retries = 10;
1514 bool retry;
1515 int rc;
1516
1517 do {
1518 retry = false;
1519 if (retry_count > retries) {
1520 netdev_warn(dev: netdev, format: "Login attempts exceeded\n");
1521 return -EACCES;
1522 }
1523
1524 adapter->init_done_rc = 0;
1525 reinit_completion(x: &adapter->init_done);
1526 rc = send_login(adapter);
1527 if (rc)
1528 return rc;
1529
1530 if (!wait_for_completion_timeout(x: &adapter->init_done,
1531 timeout)) {
1532 netdev_warn(dev: netdev, format: "Login timed out\n");
1533 adapter->login_pending = false;
1534 goto partial_reset;
1535 }
1536
1537 if (adapter->init_done_rc == ABORTED) {
1538 netdev_warn(dev: netdev, format: "Login aborted, retrying...\n");
1539 retry = true;
1540 adapter->init_done_rc = 0;
1541 retry_count++;
1542 /* FW or device may be busy, so
1543 * wait a bit before retrying login
1544 */
1545 msleep(msecs: 500);
1546 } else if (adapter->init_done_rc == PARTIALSUCCESS) {
1547 retry_count++;
1548 release_sub_crqs(adapter, 1);
1549
1550 retry = true;
1551 netdev_dbg(netdev,
1552 "Received partial success, retrying...\n");
1553 adapter->init_done_rc = 0;
1554 reinit_completion(x: &adapter->init_done);
1555 send_query_cap(adapter);
1556 if (!wait_for_completion_timeout(x: &adapter->init_done,
1557 timeout)) {
1558 netdev_warn(dev: netdev,
1559 format: "Capabilities query timed out\n");
1560 return -ETIMEDOUT;
1561 }
1562
1563 rc = init_sub_crqs(adapter);
1564 if (rc) {
1565 netdev_warn(dev: netdev,
1566 format: "SCRQ initialization failed\n");
1567 return rc;
1568 }
1569
1570 rc = init_sub_crq_irqs(adapter);
1571 if (rc) {
1572 netdev_warn(dev: netdev,
1573 format: "SCRQ irq initialization failed\n");
1574 return rc;
1575 }
1576 /* Default/timeout error handling, reset and start fresh */
1577 } else if (adapter->init_done_rc) {
1578 netdev_warn(dev: netdev, format: "Adapter login failed, init_done_rc = %d\n",
1579 adapter->init_done_rc);
1580
1581partial_reset:
1582 /* adapter login failed, so free any CRQs or sub-CRQs
1583 * and register again before attempting to login again.
1584 * If we don't do this then the VIOS may think that
1585 * we are already logged in and reject any subsequent
1586 * attempts
1587 */
1588 netdev_warn(dev: netdev,
1589 format: "Freeing and re-registering CRQs before attempting to login again\n");
1590 retry = true;
1591 adapter->init_done_rc = 0;
1592 release_sub_crqs(adapter, true);
1593 /* Much of this is similar logic as ibmvnic_probe(),
1594 * we are essentially re-initializing communication
1595 * with the server. We really should not run any
1596 * resets/failovers here because this is already a form
1597 * of reset and we do not want parallel resets occurring
1598 */
1599 do {
1600 reinit_init_done(adapter);
1601 /* Clear any failovers we got in the previous
1602 * pass since we are re-initializing the CRQ
1603 */
1604 adapter->failover_pending = false;
1605 release_crq_queue(adapter);
1606 /* If we don't sleep here then we risk an
1607 * unnecessary failover event from the VIOS.
1608 * This is a known VIOS issue caused by a vnic
1609 * device freeing and registering a CRQ too
1610 * quickly.
1611 */
1612 msleep(msecs: 1500);
1613 /* Avoid any resets, since we are currently
1614 * resetting.
1615 */
1616 spin_lock_irqsave(&adapter->rwi_lock, flags);
1617 flush_reset_queue(adapter);
1618 spin_unlock_irqrestore(lock: &adapter->rwi_lock,
1619 flags);
1620
1621 rc = init_crq_queue(adapter);
1622 if (rc) {
1623 netdev_err(dev: netdev, format: "login recovery: init CRQ failed %d\n",
1624 rc);
1625 return -EIO;
1626 }
1627
1628 rc = ibmvnic_reset_init(adapter, reset: false);
1629 if (rc)
1630 netdev_err(dev: netdev, format: "login recovery: Reset init failed %d\n",
1631 rc);
1632 /* IBMVNIC_CRQ_INIT will return EAGAIN if it
1633 * fails, since ibmvnic_reset_init will free
1634 * irq's in failure, we won't be able to receive
1635 * new CRQs so we need to keep trying. probe()
1636 * handles this similarly.
1637 */
1638 } while (rc == -EAGAIN && retry_count++ < retries);
1639 }
1640 } while (retry);
1641
1642 __ibmvnic_set_mac(netdev, adapter->mac_addr);
1643
1644 netdev_dbg(netdev, "[S:%s] Login succeeded\n", adapter_state_to_string(adapter->state));
1645 return 0;
1646}
1647
1648static void release_login_buffer(struct ibmvnic_adapter *adapter)
1649{
1650 if (!adapter->login_buf)
1651 return;
1652
1653 dma_unmap_single(&adapter->vdev->dev, adapter->login_buf_token,
1654 adapter->login_buf_sz, DMA_TO_DEVICE);
1655 kfree(objp: adapter->login_buf);
1656 adapter->login_buf = NULL;
1657}
1658
1659static void release_login_rsp_buffer(struct ibmvnic_adapter *adapter)
1660{
1661 if (!adapter->login_rsp_buf)
1662 return;
1663
1664 dma_unmap_single(&adapter->vdev->dev, adapter->login_rsp_buf_token,
1665 adapter->login_rsp_buf_sz, DMA_FROM_DEVICE);
1666 kfree(objp: adapter->login_rsp_buf);
1667 adapter->login_rsp_buf = NULL;
1668}
1669
1670static void release_resources(struct ibmvnic_adapter *adapter)
1671{
1672 release_vpd_data(adapter);
1673
1674 release_napi(adapter);
1675 release_login_buffer(adapter);
1676 release_login_rsp_buffer(adapter);
1677}
1678
1679static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state)
1680{
1681 struct net_device *netdev = adapter->netdev;
1682 unsigned long timeout = msecs_to_jiffies(m: 20000);
1683 union ibmvnic_crq crq;
1684 bool resend;
1685 int rc;
1686
1687 netdev_dbg(netdev, "setting link state %d\n", link_state);
1688
1689 memset(&crq, 0, sizeof(crq));
1690 crq.logical_link_state.first = IBMVNIC_CRQ_CMD;
1691 crq.logical_link_state.cmd = LOGICAL_LINK_STATE;
1692 crq.logical_link_state.link_state = link_state;
1693
1694 do {
1695 resend = false;
1696
1697 reinit_completion(x: &adapter->init_done);
1698 rc = ibmvnic_send_crq(adapter, &crq);
1699 if (rc) {
1700 netdev_err(dev: netdev, format: "Failed to set link state\n");
1701 return rc;
1702 }
1703
1704 if (!wait_for_completion_timeout(x: &adapter->init_done,
1705 timeout)) {
1706 netdev_err(dev: netdev, format: "timeout setting link state\n");
1707 return -ETIMEDOUT;
1708 }
1709
1710 if (adapter->init_done_rc == PARTIALSUCCESS) {
1711 /* Partuial success, delay and re-send */
1712 mdelay(1000);
1713 resend = true;
1714 } else if (adapter->init_done_rc) {
1715 netdev_warn(dev: netdev, format: "Unable to set link state, rc=%d\n",
1716 adapter->init_done_rc);
1717 return adapter->init_done_rc;
1718 }
1719 } while (resend);
1720
1721 return 0;
1722}
1723
1724static int set_real_num_queues(struct net_device *netdev)
1725{
1726 struct ibmvnic_adapter *adapter = netdev_priv(dev: netdev);
1727 int rc;
1728
1729 netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n",
1730 adapter->req_tx_queues, adapter->req_rx_queues);
1731
1732 rc = netif_set_real_num_tx_queues(dev: netdev, txq: adapter->req_tx_queues);
1733 if (rc) {
1734 netdev_err(dev: netdev, format: "failed to set the number of tx queues\n");
1735 return rc;
1736 }
1737
1738 rc = netif_set_real_num_rx_queues(dev: netdev, rxq: adapter->req_rx_queues);
1739 if (rc)
1740 netdev_err(dev: netdev, format: "failed to set the number of rx queues\n");
1741
1742 return rc;
1743}
1744
1745static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter)
1746{
1747 struct device *dev = &adapter->vdev->dev;
1748 union ibmvnic_crq crq;
1749 int len = 0;
1750 int rc;
1751
1752 if (adapter->vpd->buff)
1753 len = adapter->vpd->len;
1754
1755 mutex_lock(&adapter->fw_lock);
1756 adapter->fw_done_rc = 0;
1757 reinit_completion(x: &adapter->fw_done);
1758
1759 crq.get_vpd_size.first = IBMVNIC_CRQ_CMD;
1760 crq.get_vpd_size.cmd = GET_VPD_SIZE;
1761 rc = ibmvnic_send_crq(adapter, &crq);
1762 if (rc) {
1763 mutex_unlock(lock: &adapter->fw_lock);
1764 return rc;
1765 }
1766
1767 rc = ibmvnic_wait_for_completion(adapter, comp_done: &adapter->fw_done, timeout: 10000);
1768 if (rc) {
1769 dev_err(dev, "Could not retrieve VPD size, rc = %d\n", rc);
1770 mutex_unlock(lock: &adapter->fw_lock);
1771 return rc;
1772 }
1773 mutex_unlock(lock: &adapter->fw_lock);
1774
1775 if (!adapter->vpd->len)
1776 return -ENODATA;
1777
1778 if (!adapter->vpd->buff)
1779 adapter->vpd->buff = kzalloc(size: adapter->vpd->len, GFP_KERNEL);
1780 else if (adapter->vpd->len != len)
1781 adapter->vpd->buff =
1782 krealloc(objp: adapter->vpd->buff,
1783 new_size: adapter->vpd->len, GFP_KERNEL);
1784
1785 if (!adapter->vpd->buff) {
1786 dev_err(dev, "Could allocate VPD buffer\n");
1787 return -ENOMEM;
1788 }
1789
1790 adapter->vpd->dma_addr =
1791 dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len,
1792 DMA_FROM_DEVICE);
1793 if (dma_mapping_error(dev, dma_addr: adapter->vpd->dma_addr)) {
1794 dev_err(dev, "Could not map VPD buffer\n");
1795 kfree(objp: adapter->vpd->buff);
1796 adapter->vpd->buff = NULL;
1797 return -ENOMEM;
1798 }
1799
1800 mutex_lock(&adapter->fw_lock);
1801 adapter->fw_done_rc = 0;
1802 reinit_completion(x: &adapter->fw_done);
1803
1804 crq.get_vpd.first = IBMVNIC_CRQ_CMD;
1805 crq.get_vpd.cmd = GET_VPD;
1806 crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr);
1807 crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len);
1808 rc = ibmvnic_send_crq(adapter, &crq);
1809 if (rc) {
1810 kfree(objp: adapter->vpd->buff);
1811 adapter->vpd->buff = NULL;
1812 mutex_unlock(lock: &adapter->fw_lock);
1813 return rc;
1814 }
1815
1816 rc = ibmvnic_wait_for_completion(adapter, comp_done: &adapter->fw_done, timeout: 10000);
1817 if (rc) {
1818 dev_err(dev, "Unable to retrieve VPD, rc = %d\n", rc);
1819 kfree(objp: adapter->vpd->buff);
1820 adapter->vpd->buff = NULL;
1821 mutex_unlock(lock: &adapter->fw_lock);
1822 return rc;
1823 }
1824
1825 mutex_unlock(lock: &adapter->fw_lock);
1826 return 0;
1827}
1828
1829static int init_resources(struct ibmvnic_adapter *adapter)
1830{
1831 struct net_device *netdev = adapter->netdev;
1832 int rc;
1833
1834 rc = set_real_num_queues(netdev);
1835 if (rc)
1836 return rc;
1837
1838 adapter->vpd = kzalloc(size: sizeof(*adapter->vpd), GFP_KERNEL);
1839 if (!adapter->vpd)
1840 return -ENOMEM;
1841
1842 /* Vital Product Data (VPD) */
1843 rc = ibmvnic_get_vpd(adapter);
1844 if (rc) {
1845 netdev_err(dev: netdev, format: "failed to initialize Vital Product Data (VPD)\n");
1846 return rc;
1847 }
1848
1849 rc = init_napi(adapter);
1850 if (rc)
1851 return rc;
1852
1853 send_query_map(adapter);
1854
1855 rc = init_rx_pools(netdev);
1856 if (rc)
1857 return rc;
1858
1859 rc = init_tx_pools(netdev);
1860 return rc;
1861}
1862
1863static int __ibmvnic_open(struct net_device *netdev)
1864{
1865 struct ibmvnic_adapter *adapter = netdev_priv(dev: netdev);
1866 enum vnic_state prev_state = adapter->state;
1867 int i, rc;
1868
1869 adapter->state = VNIC_OPENING;
1870 replenish_pools(adapter);
1871 ibmvnic_napi_enable(adapter);
1872
1873 /* We're ready to receive frames, enable the sub-crq interrupts and
1874 * set the logical link state to up
1875 */
1876 for (i = 0; i < adapter->req_rx_queues; i++) {
1877 netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i);
1878 if (prev_state == VNIC_CLOSED)
1879 enable_irq(irq: adapter->rx_scrq[i]->irq);
1880 enable_scrq_irq(adapter, adapter->rx_scrq[i]);
1881 }
1882
1883 for (i = 0; i < adapter->req_tx_queues; i++) {
1884 netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i);
1885 if (prev_state == VNIC_CLOSED)
1886 enable_irq(irq: adapter->tx_scrq[i]->irq);
1887 enable_scrq_irq(adapter, adapter->tx_scrq[i]);
1888 /* netdev_tx_reset_queue will reset dql stats. During NON_FATAL
1889 * resets, don't reset the stats because there could be batched
1890 * skb's waiting to be sent. If we reset dql stats, we risk
1891 * num_completed being greater than num_queued. This will cause
1892 * a BUG_ON in dql_completed().
1893 */
1894 if (adapter->reset_reason != VNIC_RESET_NON_FATAL)
1895 netdev_tx_reset_queue(q: netdev_get_tx_queue(dev: netdev, index: i));
1896 }
1897
1898 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP);
1899 if (rc) {
1900 ibmvnic_napi_disable(adapter);
1901 ibmvnic_disable_irqs(adapter);
1902 return rc;
1903 }
1904
1905 adapter->tx_queues_active = true;
1906
1907 /* Since queues were stopped until now, there shouldn't be any
1908 * one in ibmvnic_complete_tx() or ibmvnic_xmit() so maybe we
1909 * don't need the synchronize_rcu()? Leaving it for consistency
1910 * with setting ->tx_queues_active = false.
1911 */
1912 synchronize_rcu();
1913
1914 netif_tx_start_all_queues(dev: netdev);
1915
1916 if (prev_state == VNIC_CLOSED) {
1917 for (i = 0; i < adapter->req_rx_queues; i++)
1918 napi_schedule(n: &adapter->napi[i]);
1919 }
1920
1921 adapter->state = VNIC_OPEN;
1922 return rc;
1923}
1924
1925static int ibmvnic_open(struct net_device *netdev)
1926{
1927 struct ibmvnic_adapter *adapter = netdev_priv(dev: netdev);
1928 int rc;
1929
1930 ASSERT_RTNL();
1931
1932 /* If device failover is pending or we are about to reset, just set
1933 * device state and return. Device operation will be handled by reset
1934 * routine.
1935 *
1936 * It should be safe to overwrite the adapter->state here. Since
1937 * we hold the rtnl, either the reset has not actually started or
1938 * the rtnl got dropped during the set_link_state() in do_reset().
1939 * In the former case, no one else is changing the state (again we
1940 * have the rtnl) and in the latter case, do_reset() will detect and
1941 * honor our setting below.
1942 */
1943 if (adapter->failover_pending || (test_bit(0, &adapter->resetting))) {
1944 netdev_dbg(netdev, "[S:%s FOP:%d] Resetting, deferring open\n",
1945 adapter_state_to_string(adapter->state),
1946 adapter->failover_pending);
1947 adapter->state = VNIC_OPEN;
1948 rc = 0;
1949 goto out;
1950 }
1951
1952 if (adapter->state != VNIC_CLOSED) {
1953 rc = ibmvnic_login(netdev);
1954 if (rc)
1955 goto out;
1956
1957 rc = init_resources(adapter);
1958 if (rc) {
1959 netdev_err(dev: netdev, format: "failed to initialize resources\n");
1960 goto out;
1961 }
1962 }
1963
1964 rc = __ibmvnic_open(netdev);
1965
1966out:
1967 /* If open failed and there is a pending failover or in-progress reset,
1968 * set device state and return. Device operation will be handled by
1969 * reset routine. See also comments above regarding rtnl.
1970 */
1971 if (rc &&
1972 (adapter->failover_pending || (test_bit(0, &adapter->resetting)))) {
1973 adapter->state = VNIC_OPEN;
1974 rc = 0;
1975 }
1976
1977 if (rc) {
1978 release_resources(adapter);
1979 release_rx_pools(adapter);
1980 release_tx_pools(adapter);
1981 }
1982
1983 return rc;
1984}
1985
1986static void clean_rx_pools(struct ibmvnic_adapter *adapter)
1987{
1988 struct ibmvnic_rx_pool *rx_pool;
1989 struct ibmvnic_rx_buff *rx_buff;
1990 u64 rx_entries;
1991 int rx_scrqs;
1992 int i, j;
1993
1994 if (!adapter->rx_pool)
1995 return;
1996
1997 rx_scrqs = adapter->num_active_rx_pools;
1998 rx_entries = adapter->req_rx_add_entries_per_subcrq;
1999
2000 /* Free any remaining skbs in the rx buffer pools */
2001 for (i = 0; i < rx_scrqs; i++) {
2002 rx_pool = &adapter->rx_pool[i];
2003 if (!rx_pool || !rx_pool->rx_buff)
2004 continue;
2005
2006 netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i);
2007 for (j = 0; j < rx_entries; j++) {
2008 rx_buff = &rx_pool->rx_buff[j];
2009 if (rx_buff && rx_buff->skb) {
2010 dev_kfree_skb_any(skb: rx_buff->skb);
2011 rx_buff->skb = NULL;
2012 }
2013 }
2014 }
2015}
2016
2017static void clean_one_tx_pool(struct ibmvnic_adapter *adapter,
2018 struct ibmvnic_tx_pool *tx_pool)
2019{
2020 struct ibmvnic_tx_buff *tx_buff;
2021 u64 tx_entries;
2022 int i;
2023
2024 if (!tx_pool || !tx_pool->tx_buff)
2025 return;
2026
2027 tx_entries = tx_pool->num_buffers;
2028
2029 for (i = 0; i < tx_entries; i++) {
2030 tx_buff = &tx_pool->tx_buff[i];
2031 if (tx_buff && tx_buff->skb) {
2032 dev_kfree_skb_any(skb: tx_buff->skb);
2033 tx_buff->skb = NULL;
2034 }
2035 }
2036}
2037
2038static void clean_tx_pools(struct ibmvnic_adapter *adapter)
2039{
2040 int tx_scrqs;
2041 int i;
2042
2043 if (!adapter->tx_pool || !adapter->tso_pool)
2044 return;
2045
2046 tx_scrqs = adapter->num_active_tx_pools;
2047
2048 /* Free any remaining skbs in the tx buffer pools */
2049 for (i = 0; i < tx_scrqs; i++) {
2050 netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i);
2051 clean_one_tx_pool(adapter, tx_pool: &adapter->tx_pool[i]);
2052 clean_one_tx_pool(adapter, tx_pool: &adapter->tso_pool[i]);
2053 }
2054}
2055
2056static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter)
2057{
2058 struct net_device *netdev = adapter->netdev;
2059 int i;
2060
2061 if (adapter->tx_scrq) {
2062 for (i = 0; i < adapter->req_tx_queues; i++)
2063 if (adapter->tx_scrq[i]->irq) {
2064 netdev_dbg(netdev,
2065 "Disabling tx_scrq[%d] irq\n", i);
2066 disable_scrq_irq(adapter, adapter->tx_scrq[i]);
2067 disable_irq(irq: adapter->tx_scrq[i]->irq);
2068 }
2069 }
2070
2071 if (adapter->rx_scrq) {
2072 for (i = 0; i < adapter->req_rx_queues; i++) {
2073 if (adapter->rx_scrq[i]->irq) {
2074 netdev_dbg(netdev,
2075 "Disabling rx_scrq[%d] irq\n", i);
2076 disable_scrq_irq(adapter, adapter->rx_scrq[i]);
2077 disable_irq(irq: adapter->rx_scrq[i]->irq);
2078 }
2079 }
2080 }
2081}
2082
2083static void ibmvnic_cleanup(struct net_device *netdev)
2084{
2085 struct ibmvnic_adapter *adapter = netdev_priv(dev: netdev);
2086
2087 /* ensure that transmissions are stopped if called by do_reset */
2088
2089 adapter->tx_queues_active = false;
2090
2091 /* Ensure complete_tx() and ibmvnic_xmit() see ->tx_queues_active
2092 * update so they don't restart a queue after we stop it below.
2093 */
2094 synchronize_rcu();
2095
2096 if (test_bit(0, &adapter->resetting))
2097 netif_tx_disable(dev: netdev);
2098 else
2099 netif_tx_stop_all_queues(dev: netdev);
2100
2101 ibmvnic_napi_disable(adapter);
2102 ibmvnic_disable_irqs(adapter);
2103}
2104
2105static int __ibmvnic_close(struct net_device *netdev)
2106{
2107 struct ibmvnic_adapter *adapter = netdev_priv(dev: netdev);
2108 int rc = 0;
2109
2110 adapter->state = VNIC_CLOSING;
2111 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
2112 adapter->state = VNIC_CLOSED;
2113 return rc;
2114}
2115
2116static int ibmvnic_close(struct net_device *netdev)
2117{
2118 struct ibmvnic_adapter *adapter = netdev_priv(dev: netdev);
2119 int rc;
2120
2121 netdev_dbg(netdev, "[S:%s FOP:%d FRR:%d] Closing\n",
2122 adapter_state_to_string(adapter->state),
2123 adapter->failover_pending,
2124 adapter->force_reset_recovery);
2125
2126 /* If device failover is pending, just set device state and return.
2127 * Device operation will be handled by reset routine.
2128 */
2129 if (adapter->failover_pending) {
2130 adapter->state = VNIC_CLOSED;
2131 return 0;
2132 }
2133
2134 rc = __ibmvnic_close(netdev);
2135 ibmvnic_cleanup(netdev);
2136 clean_rx_pools(adapter);
2137 clean_tx_pools(adapter);
2138
2139 return rc;
2140}
2141
2142/**
2143 * build_hdr_data - creates L2/L3/L4 header data buffer
2144 * @hdr_field: bitfield determining needed headers
2145 * @skb: socket buffer
2146 * @hdr_len: array of header lengths
2147 * @hdr_data: buffer to write the header to
2148 *
2149 * Reads hdr_field to determine which headers are needed by firmware.
2150 * Builds a buffer containing these headers. Saves individual header
2151 * lengths and total buffer length to be used to build descriptors.
2152 */
2153static int build_hdr_data(u8 hdr_field, struct sk_buff *skb,
2154 int *hdr_len, u8 *hdr_data)
2155{
2156 int len = 0;
2157 u8 *hdr;
2158
2159 if (skb_vlan_tagged(skb) && !skb_vlan_tag_present(skb))
2160 hdr_len[0] = sizeof(struct vlan_ethhdr);
2161 else
2162 hdr_len[0] = sizeof(struct ethhdr);
2163
2164 if (skb->protocol == htons(ETH_P_IP)) {
2165 hdr_len[1] = ip_hdr(skb)->ihl * 4;
2166 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2167 hdr_len[2] = tcp_hdrlen(skb);
2168 else if (ip_hdr(skb)->protocol == IPPROTO_UDP)
2169 hdr_len[2] = sizeof(struct udphdr);
2170 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2171 hdr_len[1] = sizeof(struct ipv6hdr);
2172 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2173 hdr_len[2] = tcp_hdrlen(skb);
2174 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
2175 hdr_len[2] = sizeof(struct udphdr);
2176 } else if (skb->protocol == htons(ETH_P_ARP)) {
2177 hdr_len[1] = arp_hdr_len(dev: skb->dev);
2178 hdr_len[2] = 0;
2179 }
2180
2181 memset(hdr_data, 0, 120);
2182 if ((hdr_field >> 6) & 1) {
2183 hdr = skb_mac_header(skb);
2184 memcpy(hdr_data, hdr, hdr_len[0]);
2185 len += hdr_len[0];
2186 }
2187
2188 if ((hdr_field >> 5) & 1) {
2189 hdr = skb_network_header(skb);
2190 memcpy(hdr_data + len, hdr, hdr_len[1]);
2191 len += hdr_len[1];
2192 }
2193
2194 if ((hdr_field >> 4) & 1) {
2195 hdr = skb_transport_header(skb);
2196 memcpy(hdr_data + len, hdr, hdr_len[2]);
2197 len += hdr_len[2];
2198 }
2199 return len;
2200}
2201
2202/**
2203 * create_hdr_descs - create header and header extension descriptors
2204 * @hdr_field: bitfield determining needed headers
2205 * @hdr_data: buffer containing header data
2206 * @len: length of data buffer
2207 * @hdr_len: array of individual header lengths
2208 * @scrq_arr: descriptor array
2209 *
2210 * Creates header and, if needed, header extension descriptors and
2211 * places them in a descriptor array, scrq_arr
2212 */
2213
2214static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len,
2215 union sub_crq *scrq_arr)
2216{
2217 union sub_crq hdr_desc;
2218 int tmp_len = len;
2219 int num_descs = 0;
2220 u8 *data, *cur;
2221 int tmp;
2222
2223 while (tmp_len > 0) {
2224 cur = hdr_data + len - tmp_len;
2225
2226 memset(&hdr_desc, 0, sizeof(hdr_desc));
2227 if (cur != hdr_data) {
2228 data = hdr_desc.hdr_ext.data;
2229 tmp = tmp_len > 29 ? 29 : tmp_len;
2230 hdr_desc.hdr_ext.first = IBMVNIC_CRQ_CMD;
2231 hdr_desc.hdr_ext.type = IBMVNIC_HDR_EXT_DESC;
2232 hdr_desc.hdr_ext.len = tmp;
2233 } else {
2234 data = hdr_desc.hdr.data;
2235 tmp = tmp_len > 24 ? 24 : tmp_len;
2236 hdr_desc.hdr.first = IBMVNIC_CRQ_CMD;
2237 hdr_desc.hdr.type = IBMVNIC_HDR_DESC;
2238 hdr_desc.hdr.len = tmp;
2239 hdr_desc.hdr.l2_len = (u8)hdr_len[0];
2240 hdr_desc.hdr.l3_len = cpu_to_be16((u16)hdr_len[1]);
2241 hdr_desc.hdr.l4_len = (u8)hdr_len[2];
2242 hdr_desc.hdr.flag = hdr_field << 1;
2243 }
2244 memcpy(data, cur, tmp);
2245 tmp_len -= tmp;
2246 *scrq_arr = hdr_desc;
2247 scrq_arr++;
2248 num_descs++;
2249 }
2250
2251 return num_descs;
2252}
2253
2254/**
2255 * build_hdr_descs_arr - build a header descriptor array
2256 * @skb: tx socket buffer
2257 * @indir_arr: indirect array
2258 * @num_entries: number of descriptors to be sent
2259 * @hdr_field: bit field determining which headers will be sent
2260 *
2261 * This function will build a TX descriptor array with applicable
2262 * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect.
2263 */
2264
2265static void build_hdr_descs_arr(struct sk_buff *skb,
2266 union sub_crq *indir_arr,
2267 int *num_entries, u8 hdr_field)
2268{
2269 int hdr_len[3] = {0, 0, 0};
2270 u8 hdr_data[140] = {0};
2271 int tot_len;
2272
2273 tot_len = build_hdr_data(hdr_field, skb, hdr_len,
2274 hdr_data);
2275 *num_entries += create_hdr_descs(hdr_field, hdr_data, len: tot_len, hdr_len,
2276 scrq_arr: indir_arr + 1);
2277}
2278
2279static int ibmvnic_xmit_workarounds(struct sk_buff *skb,
2280 struct net_device *netdev)
2281{
2282 /* For some backing devices, mishandling of small packets
2283 * can result in a loss of connection or TX stall. Device
2284 * architects recommend that no packet should be smaller
2285 * than the minimum MTU value provided to the driver, so
2286 * pad any packets to that length
2287 */
2288 if (skb->len < netdev->min_mtu)
2289 return skb_put_padto(skb, len: netdev->min_mtu);
2290
2291 return 0;
2292}
2293
2294static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
2295 struct ibmvnic_sub_crq_queue *tx_scrq)
2296{
2297 struct ibmvnic_ind_xmit_queue *ind_bufp;
2298 struct ibmvnic_tx_buff *tx_buff;
2299 struct ibmvnic_tx_pool *tx_pool;
2300 union sub_crq tx_scrq_entry;
2301 int queue_num;
2302 int entries;
2303 int index;
2304 int i;
2305
2306 ind_bufp = &tx_scrq->ind_buf;
2307 entries = (u64)ind_bufp->index;
2308 queue_num = tx_scrq->pool_index;
2309
2310 for (i = entries - 1; i >= 0; --i) {
2311 tx_scrq_entry = ind_bufp->indir_arr[i];
2312 if (tx_scrq_entry.v1.type != IBMVNIC_TX_DESC)
2313 continue;
2314 index = be32_to_cpu(tx_scrq_entry.v1.correlator);
2315 if (index & IBMVNIC_TSO_POOL_MASK) {
2316 tx_pool = &adapter->tso_pool[queue_num];
2317 index &= ~IBMVNIC_TSO_POOL_MASK;
2318 } else {
2319 tx_pool = &adapter->tx_pool[queue_num];
2320 }
2321 tx_pool->free_map[tx_pool->consumer_index] = index;
2322 tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
2323 tx_pool->num_buffers - 1 :
2324 tx_pool->consumer_index - 1;
2325 tx_buff = &tx_pool->tx_buff[index];
2326 adapter->netdev->stats.tx_packets--;
2327 adapter->netdev->stats.tx_bytes -= tx_buff->skb->len;
2328 adapter->tx_stats_buffers[queue_num].packets--;
2329 adapter->tx_stats_buffers[queue_num].bytes -=
2330 tx_buff->skb->len;
2331 dev_kfree_skb_any(skb: tx_buff->skb);
2332 tx_buff->skb = NULL;
2333 adapter->netdev->stats.tx_dropped++;
2334 }
2335
2336 ind_bufp->index = 0;
2337
2338 if (atomic_sub_return(i: entries, v: &tx_scrq->used) <=
2339 (adapter->req_tx_entries_per_subcrq / 2) &&
2340 __netif_subqueue_stopped(dev: adapter->netdev, queue_index: queue_num)) {
2341 rcu_read_lock();
2342
2343 if (adapter->tx_queues_active) {
2344 netif_wake_subqueue(dev: adapter->netdev, queue_index: queue_num);
2345 netdev_dbg(adapter->netdev, "Started queue %d\n",
2346 queue_num);
2347 }
2348
2349 rcu_read_unlock();
2350 }
2351}
2352
2353static int ibmvnic_tx_scrq_flush(struct ibmvnic_adapter *adapter,
2354 struct ibmvnic_sub_crq_queue *tx_scrq)
2355{
2356 struct ibmvnic_ind_xmit_queue *ind_bufp;
2357 u64 dma_addr;
2358 u64 entries;
2359 u64 handle;
2360 int rc;
2361
2362 ind_bufp = &tx_scrq->ind_buf;
2363 dma_addr = (u64)ind_bufp->indir_dma;
2364 entries = (u64)ind_bufp->index;
2365 handle = tx_scrq->handle;
2366
2367 if (!entries)
2368 return 0;
2369 rc = send_subcrq_indirect(adapter, handle, dma_addr, entries);
2370 if (rc)
2371 ibmvnic_tx_scrq_clean_buffer(adapter, tx_scrq);
2372 else
2373 ind_bufp->index = 0;
2374 return 0;
2375}
2376
2377static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
2378{
2379 struct ibmvnic_adapter *adapter = netdev_priv(dev: netdev);
2380 int queue_num = skb_get_queue_mapping(skb);
2381 u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req;
2382 struct device *dev = &adapter->vdev->dev;
2383 struct ibmvnic_ind_xmit_queue *ind_bufp;
2384 struct ibmvnic_tx_buff *tx_buff = NULL;
2385 struct ibmvnic_sub_crq_queue *tx_scrq;
2386 struct ibmvnic_long_term_buff *ltb;
2387 struct ibmvnic_tx_pool *tx_pool;
2388 unsigned int tx_send_failed = 0;
2389 netdev_tx_t ret = NETDEV_TX_OK;
2390 unsigned int tx_map_failed = 0;
2391 union sub_crq indir_arr[16];
2392 unsigned int tx_dropped = 0;
2393 unsigned int tx_packets = 0;
2394 unsigned int tx_bytes = 0;
2395 dma_addr_t data_dma_addr;
2396 struct netdev_queue *txq;
2397 unsigned long lpar_rc;
2398 union sub_crq tx_crq;
2399 unsigned int offset;
2400 int num_entries = 1;
2401 unsigned char *dst;
2402 int bufidx = 0;
2403 u8 proto = 0;
2404
2405 /* If a reset is in progress, drop the packet since
2406 * the scrqs may get torn down. Otherwise use the
2407 * rcu to ensure reset waits for us to complete.
2408 */
2409 rcu_read_lock();
2410 if (!adapter->tx_queues_active) {
2411 dev_kfree_skb_any(skb);
2412
2413 tx_send_failed++;
2414 tx_dropped++;
2415 ret = NETDEV_TX_OK;
2416 goto out;
2417 }
2418
2419 tx_scrq = adapter->tx_scrq[queue_num];
2420 txq = netdev_get_tx_queue(dev: netdev, index: queue_num);
2421 ind_bufp = &tx_scrq->ind_buf;
2422
2423 if (ibmvnic_xmit_workarounds(skb, netdev)) {
2424 tx_dropped++;
2425 tx_send_failed++;
2426 ret = NETDEV_TX_OK;
2427 ibmvnic_tx_scrq_flush(adapter, tx_scrq);
2428 goto out;
2429 }
2430
2431 if (skb_is_gso(skb))
2432 tx_pool = &adapter->tso_pool[queue_num];
2433 else
2434 tx_pool = &adapter->tx_pool[queue_num];
2435
2436 bufidx = tx_pool->free_map[tx_pool->consumer_index];
2437
2438 if (bufidx == IBMVNIC_INVALID_MAP) {
2439 dev_kfree_skb_any(skb);
2440 tx_send_failed++;
2441 tx_dropped++;
2442 ibmvnic_tx_scrq_flush(adapter, tx_scrq);
2443 ret = NETDEV_TX_OK;
2444 goto out;
2445 }
2446
2447 tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP;
2448
2449 map_txpool_buf_to_ltb(txpool: tx_pool, bufidx, ltbp: &ltb, offset: &offset);
2450
2451 dst = ltb->buff + offset;
2452 memset(dst, 0, tx_pool->buf_size);
2453 data_dma_addr = ltb->addr + offset;
2454
2455 if (skb_shinfo(skb)->nr_frags) {
2456 int cur, i;
2457
2458 /* Copy the head */
2459 skb_copy_from_linear_data(skb, to: dst, len: skb_headlen(skb));
2460 cur = skb_headlen(skb);
2461
2462 /* Copy the frags */
2463 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2464 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2465
2466 memcpy(dst + cur, skb_frag_address(frag),
2467 skb_frag_size(frag));
2468 cur += skb_frag_size(frag);
2469 }
2470 } else {
2471 skb_copy_from_linear_data(skb, to: dst, len: skb->len);
2472 }
2473
2474 /* post changes to long_term_buff *dst before VIOS accessing it */
2475 dma_wmb();
2476
2477 tx_pool->consumer_index =
2478 (tx_pool->consumer_index + 1) % tx_pool->num_buffers;
2479
2480 tx_buff = &tx_pool->tx_buff[bufidx];
2481 tx_buff->skb = skb;
2482 tx_buff->index = bufidx;
2483 tx_buff->pool_index = queue_num;
2484
2485 memset(&tx_crq, 0, sizeof(tx_crq));
2486 tx_crq.v1.first = IBMVNIC_CRQ_CMD;
2487 tx_crq.v1.type = IBMVNIC_TX_DESC;
2488 tx_crq.v1.n_crq_elem = 1;
2489 tx_crq.v1.n_sge = 1;
2490 tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED;
2491
2492 if (skb_is_gso(skb))
2493 tx_crq.v1.correlator =
2494 cpu_to_be32(bufidx | IBMVNIC_TSO_POOL_MASK);
2495 else
2496 tx_crq.v1.correlator = cpu_to_be32(bufidx);
2497 tx_crq.v1.dma_reg = cpu_to_be16(ltb->map_id);
2498 tx_crq.v1.sge_len = cpu_to_be32(skb->len);
2499 tx_crq.v1.ioba = cpu_to_be64(data_dma_addr);
2500
2501 if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) {
2502 tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT;
2503 tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci);
2504 }
2505
2506 if (skb->protocol == htons(ETH_P_IP)) {
2507 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4;
2508 proto = ip_hdr(skb)->protocol;
2509 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2510 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6;
2511 proto = ipv6_hdr(skb)->nexthdr;
2512 }
2513
2514 if (proto == IPPROTO_TCP)
2515 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP;
2516 else if (proto == IPPROTO_UDP)
2517 tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP;
2518
2519 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2520 tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD;
2521 hdrs += 2;
2522 }
2523 if (skb_is_gso(skb)) {
2524 tx_crq.v1.flags1 |= IBMVNIC_TX_LSO;
2525 tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size);
2526 hdrs += 2;
2527 }
2528
2529 if ((*hdrs >> 7) & 1)
2530 build_hdr_descs_arr(skb, indir_arr, num_entries: &num_entries, hdr_field: *hdrs);
2531
2532 tx_crq.v1.n_crq_elem = num_entries;
2533 tx_buff->num_entries = num_entries;
2534 /* flush buffer if current entry can not fit */
2535 if (num_entries + ind_bufp->index > IBMVNIC_MAX_IND_DESCS) {
2536 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
2537 if (lpar_rc != H_SUCCESS)
2538 goto tx_flush_err;
2539 }
2540
2541 indir_arr[0] = tx_crq;
2542 memcpy(&ind_bufp->indir_arr[ind_bufp->index], &indir_arr[0],
2543 num_entries * sizeof(struct ibmvnic_generic_scrq));
2544 ind_bufp->index += num_entries;
2545 if (__netdev_tx_sent_queue(dev_queue: txq, bytes: skb->len,
2546 xmit_more: netdev_xmit_more() &&
2547 ind_bufp->index < IBMVNIC_MAX_IND_DESCS)) {
2548 lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
2549 if (lpar_rc != H_SUCCESS)
2550 goto tx_err;
2551 }
2552
2553 if (atomic_add_return(i: num_entries, v: &tx_scrq->used)
2554 >= adapter->req_tx_entries_per_subcrq) {
2555 netdev_dbg(netdev, "Stopping queue %d\n", queue_num);
2556 netif_stop_subqueue(dev: netdev, queue_index: queue_num);
2557 }
2558
2559 tx_packets++;
2560 tx_bytes += skb->len;
2561 txq_trans_cond_update(txq);
2562 ret = NETDEV_TX_OK;
2563 goto out;
2564
2565tx_flush_err:
2566 dev_kfree_skb_any(skb);
2567 tx_buff->skb = NULL;
2568 tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
2569 tx_pool->num_buffers - 1 :
2570 tx_pool->consumer_index - 1;
2571 tx_dropped++;
2572tx_err:
2573 if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER)
2574 dev_err_ratelimited(dev, "tx: send failed\n");
2575
2576 if (lpar_rc == H_CLOSED || adapter->failover_pending) {
2577 /* Disable TX and report carrier off if queue is closed
2578 * or pending failover.
2579 * Firmware guarantees that a signal will be sent to the
2580 * driver, triggering a reset or some other action.
2581 */
2582 netif_tx_stop_all_queues(dev: netdev);
2583 netif_carrier_off(dev: netdev);
2584 }
2585out:
2586 rcu_read_unlock();
2587 netdev->stats.tx_dropped += tx_dropped;
2588 netdev->stats.tx_bytes += tx_bytes;
2589 netdev->stats.tx_packets += tx_packets;
2590 adapter->tx_send_failed += tx_send_failed;
2591 adapter->tx_map_failed += tx_map_failed;
2592 adapter->tx_stats_buffers[queue_num].packets += tx_packets;
2593 adapter->tx_stats_buffers[queue_num].bytes += tx_bytes;
2594 adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped;
2595
2596 return ret;
2597}
2598
2599static void ibmvnic_set_multi(struct net_device *netdev)
2600{
2601 struct ibmvnic_adapter *adapter = netdev_priv(dev: netdev);
2602 struct netdev_hw_addr *ha;
2603 union ibmvnic_crq crq;
2604
2605 memset(&crq, 0, sizeof(crq));
2606 crq.request_capability.first = IBMVNIC_CRQ_CMD;
2607 crq.request_capability.cmd = REQUEST_CAPABILITY;
2608
2609 if (netdev->flags & IFF_PROMISC) {
2610 if (!adapter->promisc_supported)
2611 return;
2612 } else {
2613 if (netdev->flags & IFF_ALLMULTI) {
2614 /* Accept all multicast */
2615 memset(&crq, 0, sizeof(crq));
2616 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2617 crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2618 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL;
2619 ibmvnic_send_crq(adapter, &crq);
2620 } else if (netdev_mc_empty(netdev)) {
2621 /* Reject all multicast */
2622 memset(&crq, 0, sizeof(crq));
2623 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2624 crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2625 crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL;
2626 ibmvnic_send_crq(adapter, &crq);
2627 } else {
2628 /* Accept one or more multicast(s) */
2629 netdev_for_each_mc_addr(ha, netdev) {
2630 memset(&crq, 0, sizeof(crq));
2631 crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2632 crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2633 crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC;
2634 ether_addr_copy(dst: &crq.multicast_ctrl.mac_addr[0],
2635 src: ha->addr);
2636 ibmvnic_send_crq(adapter, &crq);
2637 }
2638 }
2639 }
2640}
2641
2642static int __ibmvnic_set_mac(struct net_device *netdev, u8 *dev_addr)
2643{
2644 struct ibmvnic_adapter *adapter = netdev_priv(dev: netdev);
2645 union ibmvnic_crq crq;
2646 int rc;
2647
2648 if (!is_valid_ether_addr(addr: dev_addr)) {
2649 rc = -EADDRNOTAVAIL;
2650 goto err;
2651 }
2652
2653 memset(&crq, 0, sizeof(crq));
2654 crq.change_mac_addr.first = IBMVNIC_CRQ_CMD;
2655 crq.change_mac_addr.cmd = CHANGE_MAC_ADDR;
2656 ether_addr_copy(dst: &crq.change_mac_addr.mac_addr[0], src: dev_addr);
2657
2658 mutex_lock(&adapter->fw_lock);
2659 adapter->fw_done_rc = 0;
2660 reinit_completion(x: &adapter->fw_done);
2661
2662 rc = ibmvnic_send_crq(adapter, &crq);
2663 if (rc) {
2664 rc = -EIO;
2665 mutex_unlock(lock: &adapter->fw_lock);
2666 goto err;
2667 }
2668
2669 rc = ibmvnic_wait_for_completion(adapter, comp_done: &adapter->fw_done, timeout: 10000);
2670 /* netdev->dev_addr is changed in handle_change_mac_rsp function */
2671 if (rc || adapter->fw_done_rc) {
2672 rc = -EIO;
2673 mutex_unlock(lock: &adapter->fw_lock);
2674 goto err;
2675 }
2676 mutex_unlock(lock: &adapter->fw_lock);
2677 return 0;
2678err:
2679 ether_addr_copy(dst: adapter->mac_addr, src: netdev->dev_addr);
2680 return rc;
2681}
2682
2683static int ibmvnic_set_mac(struct net_device *netdev, void *p)
2684{
2685 struct ibmvnic_adapter *adapter = netdev_priv(dev: netdev);
2686 struct sockaddr *addr = p;
2687 int rc;
2688
2689 rc = 0;
2690 if (!is_valid_ether_addr(addr: addr->sa_data))
2691 return -EADDRNOTAVAIL;
2692
2693 ether_addr_copy(dst: adapter->mac_addr, src: addr->sa_data);
2694 if (adapter->state != VNIC_PROBED)
2695 rc = __ibmvnic_set_mac(netdev, dev_addr: addr->sa_data);
2696
2697 return rc;
2698}
2699
2700static const char *reset_reason_to_string(enum ibmvnic_reset_reason reason)
2701{
2702 switch (reason) {
2703 case VNIC_RESET_FAILOVER:
2704 return "FAILOVER";
2705 case VNIC_RESET_MOBILITY:
2706 return "MOBILITY";
2707 case VNIC_RESET_FATAL:
2708 return "FATAL";
2709 case VNIC_RESET_NON_FATAL:
2710 return "NON_FATAL";
2711 case VNIC_RESET_TIMEOUT:
2712 return "TIMEOUT";
2713 case VNIC_RESET_CHANGE_PARAM:
2714 return "CHANGE_PARAM";
2715 case VNIC_RESET_PASSIVE_INIT:
2716 return "PASSIVE_INIT";
2717 }
2718 return "UNKNOWN";
2719}
2720
2721/*
2722 * Initialize the init_done completion and return code values. We
2723 * can get a transport event just after registering the CRQ and the
2724 * tasklet will use this to communicate the transport event. To ensure
2725 * we don't miss the notification/error, initialize these _before_
2726 * regisering the CRQ.
2727 */
2728static inline void reinit_init_done(struct ibmvnic_adapter *adapter)
2729{
2730 reinit_completion(x: &adapter->init_done);
2731 adapter->init_done_rc = 0;
2732}
2733
2734/*
2735 * do_reset returns zero if we are able to keep processing reset events, or
2736 * non-zero if we hit a fatal error and must halt.
2737 */
2738static int do_reset(struct ibmvnic_adapter *adapter,
2739 struct ibmvnic_rwi *rwi, u32 reset_state)
2740{
2741 struct net_device *netdev = adapter->netdev;
2742 u64 old_num_rx_queues, old_num_tx_queues;
2743 u64 old_num_rx_slots, old_num_tx_slots;
2744 int rc;
2745
2746 netdev_dbg(adapter->netdev,
2747 "[S:%s FOP:%d] Reset reason: %s, reset_state: %s\n",
2748 adapter_state_to_string(adapter->state),
2749 adapter->failover_pending,
2750 reset_reason_to_string(rwi->reset_reason),
2751 adapter_state_to_string(reset_state));
2752
2753 adapter->reset_reason = rwi->reset_reason;
2754 /* requestor of VNIC_RESET_CHANGE_PARAM already has the rtnl lock */
2755 if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
2756 rtnl_lock();
2757
2758 /* Now that we have the rtnl lock, clear any pending failover.
2759 * This will ensure ibmvnic_open() has either completed or will
2760 * block until failover is complete.
2761 */
2762 if (rwi->reset_reason == VNIC_RESET_FAILOVER)
2763 adapter->failover_pending = false;
2764
2765 /* read the state and check (again) after getting rtnl */
2766 reset_state = adapter->state;
2767
2768 if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
2769 rc = -EBUSY;
2770 goto out;
2771 }
2772
2773 netif_carrier_off(dev: netdev);
2774
2775 old_num_rx_queues = adapter->req_rx_queues;
2776 old_num_tx_queues = adapter->req_tx_queues;
2777 old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq;
2778 old_num_tx_slots = adapter->req_tx_entries_per_subcrq;
2779
2780 ibmvnic_cleanup(netdev);
2781
2782 if (reset_state == VNIC_OPEN &&
2783 adapter->reset_reason != VNIC_RESET_MOBILITY &&
2784 adapter->reset_reason != VNIC_RESET_FAILOVER) {
2785 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2786 rc = __ibmvnic_close(netdev);
2787 if (rc)
2788 goto out;
2789 } else {
2790 adapter->state = VNIC_CLOSING;
2791
2792 /* Release the RTNL lock before link state change and
2793 * re-acquire after the link state change to allow
2794 * linkwatch_event to grab the RTNL lock and run during
2795 * a reset.
2796 */
2797 rtnl_unlock();
2798 rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
2799 rtnl_lock();
2800 if (rc)
2801 goto out;
2802
2803 if (adapter->state == VNIC_OPEN) {
2804 /* When we dropped rtnl, ibmvnic_open() got
2805 * it and noticed that we are resetting and
2806 * set the adapter state to OPEN. Update our
2807 * new "target" state, and resume the reset
2808 * from VNIC_CLOSING state.
2809 */
2810 netdev_dbg(netdev,
2811 "Open changed state from %s, updating.\n",
2812 adapter_state_to_string(reset_state));
2813 reset_state = VNIC_OPEN;
2814 adapter->state = VNIC_CLOSING;
2815 }
2816
2817 if (adapter->state != VNIC_CLOSING) {
2818 /* If someone else changed the adapter state
2819 * when we dropped the rtnl, fail the reset
2820 */
2821 rc = -EAGAIN;
2822 goto out;
2823 }
2824 adapter->state = VNIC_CLOSED;
2825 }
2826 }
2827
2828 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2829 release_resources(adapter);
2830 release_sub_crqs(adapter, 1);
2831 release_crq_queue(adapter);
2832 }
2833
2834 if (adapter->reset_reason != VNIC_RESET_NON_FATAL) {
2835 /* remove the closed state so when we call open it appears
2836 * we are coming from the probed state.
2837 */
2838 adapter->state = VNIC_PROBED;
2839
2840 reinit_init_done(adapter);
2841
2842 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2843 rc = init_crq_queue(adapter);
2844 } else if (adapter->reset_reason == VNIC_RESET_MOBILITY) {
2845 rc = ibmvnic_reenable_crq_queue(adapter);
2846 release_sub_crqs(adapter, 1);
2847 } else {
2848 rc = ibmvnic_reset_crq(adapter);
2849 if (rc == H_CLOSED || rc == H_SUCCESS) {
2850 rc = vio_enable_interrupts(adapter->vdev);
2851 if (rc)
2852 netdev_err(dev: adapter->netdev,
2853 format: "Reset failed to enable interrupts. rc=%d\n",
2854 rc);
2855 }
2856 }
2857
2858 if (rc) {
2859 netdev_err(dev: adapter->netdev,
2860 format: "Reset couldn't initialize crq. rc=%d\n", rc);
2861 goto out;
2862 }
2863
2864 rc = ibmvnic_reset_init(adapter, reset: true);
2865 if (rc)
2866 goto out;
2867
2868 /* If the adapter was in PROBE or DOWN state prior to the reset,
2869 * exit here.
2870 */
2871 if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN) {
2872 rc = 0;
2873 goto out;
2874 }
2875
2876 rc = ibmvnic_login(netdev);
2877 if (rc)
2878 goto out;
2879
2880 if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2881 rc = init_resources(adapter);
2882 if (rc)
2883 goto out;
2884 } else if (adapter->req_rx_queues != old_num_rx_queues ||
2885 adapter->req_tx_queues != old_num_tx_queues ||
2886 adapter->req_rx_add_entries_per_subcrq !=
2887 old_num_rx_slots ||
2888 adapter->req_tx_entries_per_subcrq !=
2889 old_num_tx_slots ||
2890 !adapter->rx_pool ||
2891 !adapter->tso_pool ||
2892 !adapter->tx_pool) {
2893 release_napi(adapter);
2894 release_vpd_data(adapter);
2895
2896 rc = init_resources(adapter);
2897 if (rc)
2898 goto out;
2899
2900 } else {
2901 rc = init_tx_pools(netdev);
2902 if (rc) {
2903 netdev_dbg(netdev,
2904 "init tx pools failed (%d)\n",
2905 rc);
2906 goto out;
2907 }
2908
2909 rc = init_rx_pools(netdev);
2910 if (rc) {
2911 netdev_dbg(netdev,
2912 "init rx pools failed (%d)\n",
2913 rc);
2914 goto out;
2915 }
2916 }
2917 ibmvnic_disable_irqs(adapter);
2918 }
2919 adapter->state = VNIC_CLOSED;
2920
2921 if (reset_state == VNIC_CLOSED) {
2922 rc = 0;
2923 goto out;
2924 }
2925
2926 rc = __ibmvnic_open(netdev);
2927 if (rc) {
2928 rc = IBMVNIC_OPEN_FAILED;
2929 goto out;
2930 }
2931
2932 /* refresh device's multicast list */
2933 ibmvnic_set_multi(netdev);
2934
2935 if (adapter->reset_reason == VNIC_RESET_FAILOVER ||
2936 adapter->reset_reason == VNIC_RESET_MOBILITY)
2937 __netdev_notify_peers(dev: netdev);
2938
2939 rc = 0;
2940
2941out:
2942 /* restore the adapter state if reset failed */
2943 if (rc)
2944 adapter->state = reset_state;
2945 /* requestor of VNIC_RESET_CHANGE_PARAM should still hold the rtnl lock */
2946 if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
2947 rtnl_unlock();
2948
2949 netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Reset done, rc %d\n",
2950 adapter_state_to_string(adapter->state),
2951 adapter->failover_pending, rc);
2952 return rc;
2953}
2954
2955static int do_hard_reset(struct ibmvnic_adapter *adapter,
2956 struct ibmvnic_rwi *rwi, u32 reset_state)
2957{
2958 struct net_device *netdev = adapter->netdev;
2959 int rc;
2960
2961 netdev_dbg(adapter->netdev, "Hard resetting driver (%s)\n",
2962 reset_reason_to_string(rwi->reset_reason));
2963
2964 /* read the state and check (again) after getting rtnl */
2965 reset_state = adapter->state;
2966
2967 if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
2968 rc = -EBUSY;
2969 goto out;
2970 }
2971
2972 netif_carrier_off(dev: netdev);
2973 adapter->reset_reason = rwi->reset_reason;
2974
2975 ibmvnic_cleanup(netdev);
2976 release_resources(adapter);
2977 release_sub_crqs(adapter, 0);
2978 release_crq_queue(adapter);
2979
2980 /* remove the closed state so when we call open it appears
2981 * we are coming from the probed state.
2982 */
2983 adapter->state = VNIC_PROBED;
2984
2985 reinit_init_done(adapter);
2986
2987 rc = init_crq_queue(adapter);
2988 if (rc) {
2989 netdev_err(dev: adapter->netdev,
2990 format: "Couldn't initialize crq. rc=%d\n", rc);
2991 goto out;
2992 }
2993
2994 rc = ibmvnic_reset_init(adapter, reset: false);
2995 if (rc)
2996 goto out;
2997
2998 /* If the adapter was in PROBE or DOWN state prior to the reset,
2999 * exit here.
3000 */
3001 if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN)
3002 goto out;
3003
3004 rc = ibmvnic_login(netdev);
3005 if (rc)
3006 goto out;
3007
3008 rc = init_resources(adapter);
3009 if (rc)
3010 goto out;
3011
3012 ibmvnic_disable_irqs(adapter);
3013 adapter->state = VNIC_CLOSED;
3014
3015 if (reset_state == VNIC_CLOSED)
3016 goto out;
3017
3018 rc = __ibmvnic_open(netdev);
3019 if (rc) {
3020 rc = IBMVNIC_OPEN_FAILED;
3021 goto out;
3022 }
3023
3024 __netdev_notify_peers(dev: netdev);
3025out:
3026 /* restore adapter state if reset failed */
3027 if (rc)
3028 adapter->state = reset_state;
3029 netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Hard reset done, rc %d\n",
3030 adapter_state_to_string(adapter->state),
3031 adapter->failover_pending, rc);
3032 return rc;
3033}
3034
3035static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter)
3036{
3037 struct ibmvnic_rwi *rwi;
3038 unsigned long flags;
3039
3040 spin_lock_irqsave(&adapter->rwi_lock, flags);
3041
3042 if (!list_empty(head: &adapter->rwi_list)) {
3043 rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi,
3044 list);
3045 list_del(entry: &rwi->list);
3046 } else {
3047 rwi = NULL;
3048 }
3049
3050 spin_unlock_irqrestore(lock: &adapter->rwi_lock, flags);
3051 return rwi;
3052}
3053
3054/**
3055 * do_passive_init - complete probing when partner device is detected.
3056 * @adapter: ibmvnic_adapter struct
3057 *
3058 * If the ibmvnic device does not have a partner device to communicate with at boot
3059 * and that partner device comes online at a later time, this function is called
3060 * to complete the initialization process of ibmvnic device.
3061 * Caller is expected to hold rtnl_lock().
3062 *
3063 * Returns non-zero if sub-CRQs are not initialized properly leaving the device
3064 * in the down state.
3065 * Returns 0 upon success and the device is in PROBED state.
3066 */
3067
3068static int do_passive_init(struct ibmvnic_adapter *adapter)
3069{
3070 unsigned long timeout = msecs_to_jiffies(m: 30000);
3071 struct net_device *netdev = adapter->netdev;
3072 struct device *dev = &adapter->vdev->dev;
3073 int rc;
3074
3075 netdev_dbg(netdev, "Partner device found, probing.\n");
3076
3077 adapter->state = VNIC_PROBING;
3078 reinit_completion(x: &adapter->init_done);
3079 adapter->init_done_rc = 0;
3080 adapter->crq.active = true;
3081
3082 rc = send_crq_init_complete(adapter);
3083 if (rc)
3084 goto out;
3085
3086 rc = send_version_xchg(adapter);
3087 if (rc)
3088 netdev_dbg(adapter->netdev, "send_version_xchg failed, rc=%d\n", rc);
3089
3090 if (!wait_for_completion_timeout(x: &adapter->init_done, timeout)) {
3091 dev_err(dev, "Initialization sequence timed out\n");
3092 rc = -ETIMEDOUT;
3093 goto out;
3094 }
3095
3096 rc = init_sub_crqs(adapter);
3097 if (rc) {
3098 dev_err(dev, "Initialization of sub crqs failed, rc=%d\n", rc);
3099 goto out;
3100 }
3101
3102 rc = init_sub_crq_irqs(adapter);
3103 if (rc) {
3104 dev_err(dev, "Failed to initialize sub crq irqs\n, rc=%d", rc);
3105 goto init_failed;
3106 }
3107
3108 netdev->mtu = adapter->req_mtu - ETH_HLEN;
3109 netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
3110 netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
3111
3112 adapter->state = VNIC_PROBED;
3113 netdev_dbg(netdev, "Probed successfully. Waiting for signal from partner device.\n");
3114
3115 return 0;
3116
3117init_failed:
3118 release_sub_crqs(adapter, 1);
3119out:
3120 adapter->state = VNIC_DOWN;
3121 return rc;
3122}
3123
3124static void __ibmvnic_reset(struct work_struct *work)
3125{
3126 struct ibmvnic_adapter *adapter;
3127 unsigned int timeout = 5000;
3128 struct ibmvnic_rwi *tmprwi;
3129 bool saved_state = false;
3130 struct ibmvnic_rwi *rwi;
3131 unsigned long flags;
3132 struct device *dev;
3133 bool need_reset;
3134 int num_fails = 0;
3135 u32 reset_state;
3136 int rc = 0;
3137
3138 adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset);
3139 dev = &adapter->vdev->dev;
3140
3141 /* Wait for ibmvnic_probe() to complete. If probe is taking too long
3142 * or if another reset is in progress, defer work for now. If probe
3143 * eventually fails it will flush and terminate our work.
3144 *
3145 * Three possibilities here:
3146 * 1. Adpater being removed - just return
3147 * 2. Timed out on probe or another reset in progress - delay the work
3148 * 3. Completed probe - perform any resets in queue
3149 */
3150 if (adapter->state == VNIC_PROBING &&
3151 !wait_for_completion_timeout(x: &adapter->probe_done, timeout)) {
3152 dev_err(dev, "Reset thread timed out on probe");
3153 queue_delayed_work(wq: system_long_wq,
3154 dwork: &adapter->ibmvnic_delayed_reset,
3155 IBMVNIC_RESET_DELAY);
3156 return;
3157 }
3158
3159 /* adapter is done with probe (i.e state is never VNIC_PROBING now) */
3160 if (adapter->state == VNIC_REMOVING)
3161 return;
3162
3163 /* ->rwi_list is stable now (no one else is removing entries) */
3164
3165 /* ibmvnic_probe() may have purged the reset queue after we were
3166 * scheduled to process a reset so there maybe no resets to process.
3167 * Before setting the ->resetting bit though, we have to make sure
3168 * that there is infact a reset to process. Otherwise we may race
3169 * with ibmvnic_open() and end up leaving the vnic down:
3170 *
3171 * __ibmvnic_reset() ibmvnic_open()
3172 * ----------------- --------------
3173 *
3174 * set ->resetting bit
3175 * find ->resetting bit is set
3176 * set ->state to IBMVNIC_OPEN (i.e
3177 * assume reset will open device)
3178 * return
3179 * find reset queue empty
3180 * return
3181 *
3182 * Neither performed vnic login/open and vnic stays down
3183 *
3184 * If we hold the lock and conditionally set the bit, either we
3185 * or ibmvnic_open() will complete the open.
3186 */
3187 need_reset = false;
3188 spin_lock(lock: &adapter->rwi_lock);
3189 if (!list_empty(head: &adapter->rwi_list)) {
3190 if (test_and_set_bit_lock(nr: 0, addr: &adapter->resetting)) {
3191 queue_delayed_work(wq: system_long_wq,
3192 dwork: &adapter->ibmvnic_delayed_reset,
3193 IBMVNIC_RESET_DELAY);
3194 } else {
3195 need_reset = true;
3196 }
3197 }
3198 spin_unlock(lock: &adapter->rwi_lock);
3199
3200 if (!need_reset)
3201 return;
3202
3203 rwi = get_next_rwi(adapter);
3204 while (rwi) {
3205 spin_lock_irqsave(&adapter->state_lock, flags);
3206
3207 if (adapter->state == VNIC_REMOVING ||
3208 adapter->state == VNIC_REMOVED) {
3209 spin_unlock_irqrestore(lock: &adapter->state_lock, flags);
3210 kfree(objp: rwi);
3211 rc = EBUSY;
3212 break;
3213 }
3214
3215 if (!saved_state) {
3216 reset_state = adapter->state;
3217 saved_state = true;
3218 }
3219 spin_unlock_irqrestore(lock: &adapter->state_lock, flags);
3220
3221 if (rwi->reset_reason == VNIC_RESET_PASSIVE_INIT) {
3222 rtnl_lock();
3223 rc = do_passive_init(adapter);
3224 rtnl_unlock();
3225 if (!rc)
3226 netif_carrier_on(dev: adapter->netdev);
3227 } else if (adapter->force_reset_recovery) {
3228 /* Since we are doing a hard reset now, clear the
3229 * failover_pending flag so we don't ignore any
3230 * future MOBILITY or other resets.
3231 */
3232 adapter->failover_pending = false;
3233
3234 /* Transport event occurred during previous reset */
3235 if (adapter->wait_for_reset) {
3236 /* Previous was CHANGE_PARAM; caller locked */
3237 adapter->force_reset_recovery = false;
3238 rc = do_hard_reset(adapter, rwi, reset_state);
3239 } else {
3240 rtnl_lock();
3241 adapter->force_reset_recovery = false;
3242 rc = do_hard_reset(adapter, rwi, reset_state);
3243 rtnl_unlock();
3244 }
3245 if (rc)
3246 num_fails++;
3247 else
3248 num_fails = 0;
3249
3250 /* If auto-priority-failover is enabled we can get
3251 * back to back failovers during resets, resulting
3252 * in at least two failed resets (from high-priority
3253 * backing device to low-priority one and then back)
3254 * If resets continue to fail beyond that, give the
3255 * adapter some time to settle down before retrying.
3256 */
3257 if (num_fails >= 3) {
3258 netdev_dbg(adapter->netdev,
3259 "[S:%s] Hard reset failed %d times, waiting 60 secs\n",
3260 adapter_state_to_string(adapter->state),
3261 num_fails);
3262 set_current_state(TASK_UNINTERRUPTIBLE);
3263 schedule_timeout(timeout: 60 * HZ);
3264 }
3265 } else {
3266 rc = do_reset(adapter, rwi, reset_state);
3267 }
3268 tmprwi = rwi;
3269 adapter->last_reset_time = jiffies;
3270
3271 if (rc)
3272 netdev_dbg(adapter->netdev, "Reset failed, rc=%d\n", rc);
3273
3274 rwi = get_next_rwi(adapter);
3275
3276 /*
3277 * If there are no resets queued and the previous reset failed,
3278 * the adapter would be in an undefined state. So retry the
3279 * previous reset as a hard reset.
3280 *
3281 * Else, free the previous rwi and, if there is another reset
3282 * queued, process the new reset even if previous reset failed
3283 * (the previous reset could have failed because of a fail
3284 * over for instance, so process the fail over).
3285 */
3286 if (!rwi && rc)
3287 rwi = tmprwi;
3288 else
3289 kfree(objp: tmprwi);
3290
3291 if (rwi && (rwi->reset_reason == VNIC_RESET_FAILOVER ||
3292 rwi->reset_reason == VNIC_RESET_MOBILITY || rc))
3293 adapter->force_reset_recovery = true;
3294 }
3295
3296 if (adapter->wait_for_reset) {
3297 adapter->reset_done_rc = rc;
3298 complete(&adapter->reset_done);
3299 }
3300
3301 clear_bit_unlock(nr: 0, addr: &adapter->resetting);
3302
3303 netdev_dbg(adapter->netdev,
3304 "[S:%s FRR:%d WFR:%d] Done processing resets\n",
3305 adapter_state_to_string(adapter->state),
3306 adapter->force_reset_recovery,
3307 adapter->wait_for_reset);
3308}
3309
3310static void __ibmvnic_delayed_reset(struct work_struct *work)
3311{
3312 struct ibmvnic_adapter *adapter;
3313
3314 adapter = container_of(work, struct ibmvnic_adapter,
3315 ibmvnic_delayed_reset.work);
3316 __ibmvnic_reset(work: &adapter->ibmvnic_reset);
3317}
3318
3319static void flush_reset_queue(struct ibmvnic_adapter *adapter)
3320{
3321 struct list_head *entry, *tmp_entry;
3322
3323 if (!list_empty(head: &adapter->rwi_list)) {
3324 list_for_each_safe(entry, tmp_entry, &adapter->rwi_list) {
3325 list_del(entry);
3326 kfree(list_entry(entry, struct ibmvnic_rwi, list));
3327 }
3328 }
3329}
3330
3331static int ibmvnic_reset(struct ibmvnic_adapter *adapter,
3332 enum ibmvnic_reset_reason reason)
3333{
3334 struct net_device *netdev = adapter->netdev;
3335 struct ibmvnic_rwi *rwi, *tmp;
3336 unsigned long flags;
3337 int ret;
3338
3339 spin_lock_irqsave(&adapter->rwi_lock, flags);
3340
3341 /* If failover is pending don't schedule any other reset.
3342 * Instead let the failover complete. If there is already a
3343 * a failover reset scheduled, we will detect and drop the
3344 * duplicate reset when walking the ->rwi_list below.
3345 */
3346 if (adapter->state == VNIC_REMOVING ||
3347 adapter->state == VNIC_REMOVED ||
3348 (adapter->failover_pending && reason != VNIC_RESET_FAILOVER)) {
3349 ret = EBUSY;
3350 netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n");
3351 goto err;
3352 }
3353
3354 list_for_each_entry(tmp, &adapter->rwi_list, list) {
3355 if (tmp->reset_reason == reason) {
3356 netdev_dbg(netdev, "Skipping matching reset, reason=%s\n",
3357 reset_reason_to_string(reason));
3358 ret = EBUSY;
3359 goto err;
3360 }
3361 }
3362
3363 rwi = kzalloc(size: sizeof(*rwi), GFP_ATOMIC);
3364 if (!rwi) {
3365 ret = ENOMEM;
3366 goto err;
3367 }
3368 /* if we just received a transport event,
3369 * flush reset queue and process this reset
3370 */
3371 if (adapter->force_reset_recovery)
3372 flush_reset_queue(adapter);
3373
3374 rwi->reset_reason = reason;
3375 list_add_tail(new: &rwi->list, head: &adapter->rwi_list);
3376 netdev_dbg(adapter->netdev, "Scheduling reset (reason %s)\n",
3377 reset_reason_to_string(reason));
3378 queue_work(wq: system_long_wq, work: &adapter->ibmvnic_reset);
3379
3380 ret = 0;
3381err:
3382 /* ibmvnic_close() below can block, so drop the lock first */
3383 spin_unlock_irqrestore(lock: &adapter->rwi_lock, flags);
3384
3385 if (ret == ENOMEM)
3386 ibmvnic_close(netdev);
3387
3388 return -ret;
3389}
3390
3391static void ibmvnic_tx_timeout(struct net_device *dev, unsigned int txqueue)
3392{
3393 struct ibmvnic_adapter *adapter = netdev_priv(dev);
3394
3395 if (test_bit(0, &adapter->resetting)) {
3396 netdev_err(dev: adapter->netdev,
3397 format: "Adapter is resetting, skip timeout reset\n");
3398 return;
3399 }
3400 /* No queuing up reset until at least 5 seconds (default watchdog val)
3401 * after last reset
3402 */
3403 if (time_before(jiffies, (adapter->last_reset_time + dev->watchdog_timeo))) {
3404 netdev_dbg(dev, "Not yet time to tx timeout.\n");
3405 return;
3406 }
3407 ibmvnic_reset(adapter, reason: VNIC_RESET_TIMEOUT);
3408}
3409
3410static void remove_buff_from_pool(struct ibmvnic_adapter *adapter,
3411 struct ibmvnic_rx_buff *rx_buff)
3412{
3413 struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index];
3414
3415 rx_buff->skb = NULL;
3416
3417 pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff);
3418 pool->next_alloc = (pool->next_alloc + 1) % pool->size;
3419
3420 atomic_dec(v: &pool->available);
3421}
3422
3423static int ibmvnic_poll(struct napi_struct *napi, int budget)
3424{
3425 struct ibmvnic_sub_crq_queue *rx_scrq;
3426 struct ibmvnic_adapter *adapter;
3427 struct net_device *netdev;
3428 int frames_processed;
3429 int scrq_num;
3430
3431 netdev = napi->dev;
3432 adapter = netdev_priv(dev: netdev);
3433 scrq_num = (int)(napi - adapter->napi);
3434 frames_processed = 0;
3435 rx_scrq = adapter->rx_scrq[scrq_num];
3436
3437restart_poll:
3438 while (frames_processed < budget) {
3439 struct sk_buff *skb;
3440 struct ibmvnic_rx_buff *rx_buff;
3441 union sub_crq *next;
3442 u32 length;
3443 u16 offset;
3444 u8 flags = 0;
3445
3446 if (unlikely(test_bit(0, &adapter->resetting) &&
3447 adapter->reset_reason != VNIC_RESET_NON_FATAL)) {
3448 enable_scrq_irq(adapter, rx_scrq);
3449 napi_complete_done(n: napi, work_done: frames_processed);
3450 return frames_processed;
3451 }
3452
3453 if (!pending_scrq(adapter, rx_scrq))
3454 break;
3455 next = ibmvnic_next_scrq(adapter, rx_scrq);
3456 rx_buff = (struct ibmvnic_rx_buff *)
3457 be64_to_cpu(next->rx_comp.correlator);
3458 /* do error checking */
3459 if (next->rx_comp.rc) {
3460 netdev_dbg(netdev, "rx buffer returned with rc %x\n",
3461 be16_to_cpu(next->rx_comp.rc));
3462 /* free the entry */
3463 next->rx_comp.first = 0;
3464 dev_kfree_skb_any(skb: rx_buff->skb);
3465 remove_buff_from_pool(adapter, rx_buff);
3466 continue;
3467 } else if (!rx_buff->skb) {
3468 /* free the entry */
3469 next->rx_comp.first = 0;
3470 remove_buff_from_pool(adapter, rx_buff);
3471 continue;
3472 }
3473
3474 length = be32_to_cpu(next->rx_comp.len);
3475 offset = be16_to_cpu(next->rx_comp.off_frame_data);
3476 flags = next->rx_comp.flags;
3477 skb = rx_buff->skb;
3478 /* load long_term_buff before copying to skb */
3479 dma_rmb();
3480 skb_copy_to_linear_data(skb, from: rx_buff->data + offset,
3481 len: length);
3482
3483 /* VLAN Header has been stripped by the system firmware and
3484 * needs to be inserted by the driver
3485 */
3486 if (adapter->rx_vlan_header_insertion &&
3487 (flags & IBMVNIC_VLAN_STRIPPED))
3488 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
3489 ntohs(next->rx_comp.vlan_tci));
3490
3491 /* free the entry */
3492 next->rx_comp.first = 0;
3493 remove_buff_from_pool(adapter, rx_buff);
3494
3495 skb_put(skb, len: length);
3496 skb->protocol = eth_type_trans(skb, dev: netdev);
3497 skb_record_rx_queue(skb, rx_queue: scrq_num);
3498
3499 if (flags & IBMVNIC_IP_CHKSUM_GOOD &&
3500 flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) {
3501 skb->ip_summed = CHECKSUM_UNNECESSARY;
3502 }
3503
3504 length = skb->len;
3505 napi_gro_receive(napi, skb); /* send it up */
3506 netdev->stats.rx_packets++;
3507 netdev->stats.rx_bytes += length;
3508 adapter->rx_stats_buffers[scrq_num].packets++;
3509 adapter->rx_stats_buffers[scrq_num].bytes += length;
3510 frames_processed++;
3511 }
3512
3513 if (adapter->state != VNIC_CLOSING &&
3514 ((atomic_read(v: &adapter->rx_pool[scrq_num].available) <
3515 adapter->req_rx_add_entries_per_subcrq / 2) ||
3516 frames_processed < budget))
3517 replenish_rx_pool(adapter, pool: &adapter->rx_pool[scrq_num]);
3518 if (frames_processed < budget) {
3519 if (napi_complete_done(n: napi, work_done: frames_processed)) {
3520 enable_scrq_irq(adapter, rx_scrq);
3521 if (pending_scrq(adapter, rx_scrq)) {
3522 if (napi_schedule(n: napi)) {
3523 disable_scrq_irq(adapter, rx_scrq);
3524 goto restart_poll;
3525 }
3526 }
3527 }
3528 }
3529 return frames_processed;
3530}
3531
3532static int wait_for_reset(struct ibmvnic_adapter *adapter)
3533{
3534 int rc, ret;
3535
3536 adapter->fallback.mtu = adapter->req_mtu;
3537 adapter->fallback.rx_queues = adapter->req_rx_queues;
3538 adapter->fallback.tx_queues = adapter->req_tx_queues;
3539 adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq;
3540 adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq;
3541
3542 reinit_completion(x: &adapter->reset_done);
3543 adapter->wait_for_reset = true;
3544 rc = ibmvnic_reset(adapter, reason: VNIC_RESET_CHANGE_PARAM);
3545
3546 if (rc) {
3547 ret = rc;
3548 goto out;
3549 }
3550 rc = ibmvnic_wait_for_completion(adapter, comp_done: &adapter->reset_done, timeout: 60000);
3551 if (rc) {
3552 ret = -ENODEV;
3553 goto out;
3554 }
3555
3556 ret = 0;
3557 if (adapter->reset_done_rc) {
3558 ret = -EIO;
3559 adapter->desired.mtu = adapter->fallback.mtu;
3560 adapter->desired.rx_queues = adapter->fallback.rx_queues;
3561 adapter->desired.tx_queues = adapter->fallback.tx_queues;
3562 adapter->desired.rx_entries = adapter->fallback.rx_entries;
3563 adapter->desired.tx_entries = adapter->fallback.tx_entries;
3564
3565 reinit_completion(x: &adapter->reset_done);
3566 adapter->wait_for_reset = true;
3567 rc = ibmvnic_reset(adapter, reason: VNIC_RESET_CHANGE_PARAM);
3568 if (rc) {
3569 ret = rc;
3570 goto out;
3571 }
3572 rc = ibmvnic_wait_for_completion(adapter, comp_done: &adapter->reset_done,
3573 timeout: 60000);
3574 if (rc) {
3575 ret = -ENODEV;
3576 goto out;
3577 }
3578 }
3579out:
3580 adapter->wait_for_reset = false;
3581
3582 return ret;
3583}
3584
3585static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu)
3586{
3587 struct ibmvnic_adapter *adapter = netdev_priv(dev: netdev);
3588
3589 adapter->desired.mtu = new_mtu + ETH_HLEN;
3590
3591 return wait_for_reset(adapter);
3592}
3593
3594static netdev_features_t ibmvnic_features_check(struct sk_buff *skb,
3595 struct net_device *dev,
3596 netdev_features_t features)
3597{
3598 /* Some backing hardware adapters can not
3599 * handle packets with a MSS less than 224
3600 * or with only one segment.
3601 */
3602 if (skb_is_gso(skb)) {
3603 if (skb_shinfo(skb)->gso_size < 224 ||
3604 skb_shinfo(skb)->gso_segs == 1)
3605 features &= ~NETIF_F_GSO_MASK;
3606 }
3607
3608 return features;
3609}
3610
3611static const struct net_device_ops ibmvnic_netdev_ops = {
3612 .ndo_open = ibmvnic_open,
3613 .ndo_stop = ibmvnic_close,
3614 .ndo_start_xmit = ibmvnic_xmit,
3615 .ndo_set_rx_mode = ibmvnic_set_multi,
3616 .ndo_set_mac_address = ibmvnic_set_mac,
3617 .ndo_validate_addr = eth_validate_addr,
3618 .ndo_tx_timeout = ibmvnic_tx_timeout,
3619 .ndo_change_mtu = ibmvnic_change_mtu,
3620 .ndo_features_check = ibmvnic_features_check,
3621};
3622
3623/* ethtool functions */
3624
3625static int ibmvnic_get_link_ksettings(struct net_device *netdev,
3626 struct ethtool_link_ksettings *cmd)
3627{
3628 struct ibmvnic_adapter *adapter = netdev_priv(dev: netdev);
3629 int rc;
3630
3631 rc = send_query_phys_parms(adapter);
3632 if (rc) {
3633 adapter->speed = SPEED_UNKNOWN;
3634 adapter->duplex = DUPLEX_UNKNOWN;
3635 }
3636 cmd->base.speed = adapter->speed;
3637 cmd->base.duplex = adapter->duplex;
3638 cmd->base.port = PORT_FIBRE;
3639 cmd->base.phy_address = 0;
3640 cmd->base.autoneg = AUTONEG_ENABLE;
3641
3642 return 0;
3643}
3644
3645static void ibmvnic_get_drvinfo(struct net_device *netdev,
3646 struct ethtool_drvinfo *info)
3647{
3648 struct ibmvnic_adapter *adapter = netdev_priv(dev: netdev);
3649
3650 strscpy(p: info->driver, q: ibmvnic_driver_name, size: sizeof(info->driver));
3651 strscpy(p: info->version, IBMVNIC_DRIVER_VERSION, size: sizeof(info->version));
3652 strscpy(p: info->fw_version, q: adapter->fw_version,
3653 size: sizeof(info->fw_version));
3654}
3655
3656static u32 ibmvnic_get_msglevel(struct net_device *netdev)
3657{
3658 struct ibmvnic_adapter *adapter = netdev_priv(dev: netdev);
3659
3660 return adapter->msg_enable;
3661}
3662
3663static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data)
3664{
3665 struct ibmvnic_adapter *adapter = netdev_priv(dev: netdev);
3666
3667 adapter->msg_enable = data;
3668}
3669
3670static u32 ibmvnic_get_link(struct net_device *netdev)
3671{
3672 struct ibmvnic_adapter *adapter = netdev_priv(dev: netdev);
3673
3674 /* Don't need to send a query because we request a logical link up at
3675 * init and then we wait for link state indications
3676 */
3677 return adapter->logical_link_state;
3678}
3679
3680static void ibmvnic_get_ringparam(struct net_device *netdev,
3681 struct ethtool_ringparam *ring,
3682 struct kernel_ethtool_ringparam *kernel_ring,
3683 struct netlink_ext_ack *extack)
3684{
3685 struct ibmvnic_adapter *adapter = netdev_priv(dev: netdev);
3686
3687 ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq;
3688 ring->tx_max_pending = adapter->max_tx_entries_per_subcrq;
3689 ring->rx_mini_max_pending = 0;
3690 ring->rx_jumbo_max_pending = 0;
3691 ring->rx_pending = adapter->req_rx_add_entries_per_subcrq;
3692 ring->tx_pending = adapter->req_tx_entries_per_subcrq;
3693 ring->rx_mini_pending = 0;
3694 ring->rx_jumbo_pending = 0;
3695}
3696
3697static int ibmvnic_set_ringparam(struct net_device *netdev,
3698 struct ethtool_ringparam *ring,
3699 struct kernel_ethtool_ringparam *kernel_ring,
3700 struct netlink_ext_ack *extack)
3701{
3702 struct ibmvnic_adapter *adapter = netdev_priv(dev: netdev);
3703
3704 if (ring->rx_pending > adapter->max_rx_add_entries_per_subcrq ||
3705 ring->tx_pending > adapter->max_tx_entries_per_subcrq) {
3706 netdev_err(dev: netdev, format: "Invalid request.\n");
3707 netdev_err(dev: netdev, format: "Max tx buffers = %llu\n",
3708 adapter->max_rx_add_entries_per_subcrq);
3709 netdev_err(dev: netdev, format: "Max rx buffers = %llu\n",
3710 adapter->max_tx_entries_per_subcrq);
3711 return -EINVAL;
3712 }
3713
3714 adapter->desired.rx_entries = ring->rx_pending;
3715 adapter->desired.tx_entries = ring->tx_pending;
3716
3717 return wait_for_reset(adapter);
3718}
3719
3720static void ibmvnic_get_channels(struct net_device *netdev,
3721 struct ethtool_channels *channels)
3722{
3723 struct ibmvnic_adapter *adapter = netdev_priv(dev: netdev);
3724
3725 channels->max_rx = adapter->max_rx_queues;
3726 channels->max_tx = adapter->max_tx_queues;
3727 channels->max_other = 0;
3728 channels->max_combined = 0;
3729 channels->rx_count = adapter->req_rx_queues;
3730 channels->tx_count = adapter->req_tx_queues;
3731 channels->other_count = 0;
3732 channels->combined_count = 0;
3733}
3734
3735static int ibmvnic_set_channels(struct net_device *netdev,
3736 struct ethtool_channels *channels)
3737{
3738 struct ibmvnic_adapter *adapter = netdev_priv(dev: netdev);
3739
3740 adapter->desired.rx_queues = channels->rx_count;
3741 adapter->desired.tx_queues = channels->tx_count;
3742
3743 return wait_for_reset(adapter);
3744}
3745
3746static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data)
3747{
3748 struct ibmvnic_adapter *adapter = netdev_priv(dev);
3749 int i;
3750
3751 if (stringset != ETH_SS_STATS)
3752 return;
3753
3754 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++, data += ETH_GSTRING_LEN)
3755 memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN);
3756
3757 for (i = 0; i < adapter->req_tx_queues; i++) {
3758 snprintf(buf: data, ETH_GSTRING_LEN, fmt: "tx%d_packets", i);
3759 data += ETH_GSTRING_LEN;
3760
3761 snprintf(buf: data, ETH_GSTRING_LEN, fmt: "tx%d_bytes", i);
3762 data += ETH_GSTRING_LEN;
3763
3764 snprintf(buf: data, ETH_GSTRING_LEN, fmt: "tx%d_dropped_packets", i);
3765 data += ETH_GSTRING_LEN;
3766 }
3767
3768 for (i = 0; i < adapter->req_rx_queues; i++) {
3769 snprintf(buf: data, ETH_GSTRING_LEN, fmt: "rx%d_packets", i);
3770 data += ETH_GSTRING_LEN;
3771
3772 snprintf(buf: data, ETH_GSTRING_LEN, fmt: "rx%d_bytes", i);
3773 data += ETH_GSTRING_LEN;
3774
3775 snprintf(buf: data, ETH_GSTRING_LEN, fmt: "rx%d_interrupts", i);
3776 data += ETH_GSTRING_LEN;
3777 }
3778}
3779
3780static int ibmvnic_get_sset_count(struct net_device *dev, int sset)
3781{
3782 struct ibmvnic_adapter *adapter = netdev_priv(dev);
3783
3784 switch (sset) {
3785 case ETH_SS_STATS:
3786 return ARRAY_SIZE(ibmvnic_stats) +
3787 adapter->req_tx_queues * NUM_TX_STATS +
3788 adapter->req_rx_queues * NUM_RX_STATS;
3789 default:
3790 return -EOPNOTSUPP;
3791 }
3792}
3793
3794static void ibmvnic_get_ethtool_stats(struct net_device *dev,
3795 struct ethtool_stats *stats, u64 *data)
3796{
3797 struct ibmvnic_adapter *adapter = netdev_priv(dev);
3798 union ibmvnic_crq crq;
3799 int i, j;
3800 int rc;
3801
3802 memset(&crq, 0, sizeof(crq));
3803 crq.request_statistics.first = IBMVNIC_CRQ_CMD;
3804 crq.request_statistics.cmd = REQUEST_STATISTICS;
3805 crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token);
3806 crq.request_statistics.len =
3807 cpu_to_be32(sizeof(struct ibmvnic_statistics));
3808
3809 /* Wait for data to be written */
3810 reinit_completion(x: &adapter->stats_done);
3811 rc = ibmvnic_send_crq(adapter, &crq);
3812 if (rc)
3813 return;
3814 rc = ibmvnic_wait_for_completion(adapter, comp_done: &adapter->stats_done, timeout: 10000);
3815 if (rc)
3816 return;
3817
3818 for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++)
3819 data[i] = be64_to_cpu(IBMVNIC_GET_STAT
3820 (adapter, ibmvnic_stats[i].offset));
3821
3822 for (j = 0; j < adapter->req_tx_queues; j++) {
3823 data[i] = adapter->tx_stats_buffers[j].packets;
3824 i++;
3825 data[i] = adapter->tx_stats_buffers[j].bytes;
3826 i++;
3827 data[i] = adapter->tx_stats_buffers[j].dropped_packets;
3828 i++;
3829 }
3830
3831 for (j = 0; j < adapter->req_rx_queues; j++) {
3832 data[i] = adapter->rx_stats_buffers[j].packets;
3833 i++;
3834 data[i] = adapter->rx_stats_buffers[j].bytes;
3835 i++;
3836 data[i] = adapter->rx_stats_buffers[j].interrupts;
3837 i++;
3838 }
3839}
3840
3841static const struct ethtool_ops ibmvnic_ethtool_ops = {
3842 .get_drvinfo = ibmvnic_get_drvinfo,
3843 .get_msglevel = ibmvnic_get_msglevel,
3844 .set_msglevel = ibmvnic_set_msglevel,
3845 .get_link = ibmvnic_get_link,
3846 .get_ringparam = ibmvnic_get_ringparam,
3847 .set_ringparam = ibmvnic_set_ringparam,
3848 .get_channels = ibmvnic_get_channels,
3849 .set_channels = ibmvnic_set_channels,
3850 .get_strings = ibmvnic_get_strings,
3851 .get_sset_count = ibmvnic_get_sset_count,
3852 .get_ethtool_stats = ibmvnic_get_ethtool_stats,
3853 .get_link_ksettings = ibmvnic_get_link_ksettings,
3854};
3855
3856/* Routines for managing CRQs/sCRQs */
3857
3858static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter,
3859 struct ibmvnic_sub_crq_queue *scrq)
3860{
3861 int rc;
3862
3863 if (!scrq) {
3864 netdev_dbg(adapter->netdev, "Invalid scrq reset.\n");
3865 return -EINVAL;
3866 }
3867
3868 if (scrq->irq) {
3869 free_irq(scrq->irq, scrq);
3870 irq_dispose_mapping(virq: scrq->irq);
3871 scrq->irq = 0;
3872 }
3873
3874 if (scrq->msgs) {
3875 memset(scrq->msgs, 0, 4 * PAGE_SIZE);
3876 atomic_set(v: &scrq->used, i: 0);
3877 scrq->cur = 0;
3878 scrq->ind_buf.index = 0;
3879 } else {
3880 netdev_dbg(adapter->netdev, "Invalid scrq reset\n");
3881 return -EINVAL;
3882 }
3883
3884 rc = h_reg_sub_crq(unit_address: adapter->vdev->unit_address, token: scrq->msg_token,
3885 length: 4 * PAGE_SIZE, number: &scrq->crq_num, irq: &scrq->hw_irq);
3886 return rc;
3887}
3888
3889static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter)
3890{
3891 int i, rc;
3892
3893 if (!adapter->tx_scrq || !adapter->rx_scrq)
3894 return -EINVAL;
3895
3896 ibmvnic_clean_affinity(adapter);
3897
3898 for (i = 0; i < adapter->req_tx_queues; i++) {
3899 netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i);
3900 rc = reset_one_sub_crq_queue(adapter, scrq: adapter->tx_scrq[i]);
3901 if (rc)
3902 return rc;
3903 }
3904
3905 for (i = 0; i < adapter->req_rx_queues; i++) {
3906 netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i);
3907 rc = reset_one_sub_crq_queue(adapter, scrq: adapter->rx_scrq[i]);
3908 if (rc)
3909 return rc;
3910 }
3911
3912 return rc;
3913}
3914
3915static void release_sub_crq_queue(struct ibmvnic_adapter *adapter,
3916 struct ibmvnic_sub_crq_queue *scrq,
3917 bool do_h_free)
3918{
3919 struct device *dev = &adapter->vdev->dev;
3920 long rc;
3921
3922 netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n");
3923
3924 if (do_h_free) {
3925 /* Close the sub-crqs */
3926 do {
3927 rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3928 adapter->vdev->unit_address,
3929 scrq->crq_num);
3930 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
3931
3932 if (rc) {
3933 netdev_err(dev: adapter->netdev,
3934 format: "Failed to release sub-CRQ %16lx, rc = %ld\n",
3935 scrq->crq_num, rc);
3936 }
3937 }
3938
3939 dma_free_coherent(dev,
3940 IBMVNIC_IND_ARR_SZ,
3941 cpu_addr: scrq->ind_buf.indir_arr,
3942 dma_handle: scrq->ind_buf.indir_dma);
3943
3944 dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3945 DMA_BIDIRECTIONAL);
3946 free_pages(addr: (unsigned long)scrq->msgs, order: 2);
3947 free_cpumask_var(mask: scrq->affinity_mask);
3948 kfree(objp: scrq);
3949}
3950
3951static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter
3952 *adapter)
3953{
3954 struct device *dev = &adapter->vdev->dev;
3955 struct ibmvnic_sub_crq_queue *scrq;
3956 int rc;
3957
3958 scrq = kzalloc(size: sizeof(*scrq), GFP_KERNEL);
3959 if (!scrq)
3960 return NULL;
3961
3962 scrq->msgs =
3963 (union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, order: 2);
3964 if (!scrq->msgs) {
3965 dev_warn(dev, "Couldn't allocate crq queue messages page\n");
3966 goto zero_page_failed;
3967 }
3968 if (!zalloc_cpumask_var(mask: &scrq->affinity_mask, GFP_KERNEL))
3969 goto cpumask_alloc_failed;
3970
3971 scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE,
3972 DMA_BIDIRECTIONAL);
3973 if (dma_mapping_error(dev, dma_addr: scrq->msg_token)) {
3974 dev_warn(dev, "Couldn't map crq queue messages page\n");
3975 goto map_failed;
3976 }
3977
3978 rc = h_reg_sub_crq(unit_address: adapter->vdev->unit_address, token: scrq->msg_token,
3979 length: 4 * PAGE_SIZE, number: &scrq->crq_num, irq: &scrq->hw_irq);
3980
3981 if (rc == H_RESOURCE)
3982 rc = ibmvnic_reset_crq(adapter);
3983
3984 if (rc == H_CLOSED) {
3985 dev_warn(dev, "Partner adapter not ready, waiting.\n");
3986 } else if (rc) {
3987 dev_warn(dev, "Error %d registering sub-crq\n", rc);
3988 goto reg_failed;
3989 }
3990
3991 scrq->adapter = adapter;
3992 scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs);
3993 scrq->ind_buf.index = 0;
3994
3995 scrq->ind_buf.indir_arr =
3996 dma_alloc_coherent(dev,
3997 IBMVNIC_IND_ARR_SZ,
3998 dma_handle: &scrq->ind_buf.indir_dma,
3999 GFP_KERNEL);
4000
4001 if (!scrq->ind_buf.indir_arr)
4002 goto indir_failed;
4003
4004 spin_lock_init(&scrq->lock);
4005
4006 netdev_dbg(adapter->netdev,
4007 "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n",
4008 scrq->crq_num, scrq->hw_irq, scrq->irq);
4009
4010 return scrq;
4011
4012indir_failed:
4013 do {
4014 rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
4015 adapter->vdev->unit_address,
4016 scrq->crq_num);
4017 } while (rc == H_BUSY || rc == H_IS_LONG_BUSY(rc));
4018reg_failed:
4019 dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
4020 DMA_BIDIRECTIONAL);
4021map_failed:
4022 free_cpumask_var(mask: scrq->affinity_mask);
4023cpumask_alloc_failed:
4024 free_pages(addr: (unsigned long)scrq->msgs, order: 2);
4025zero_page_failed:
4026 kfree(objp: scrq);
4027
4028 return NULL;
4029}
4030
4031static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free)
4032{
4033 int i;
4034
4035 ibmvnic_clean_affinity(adapter);
4036 if (adapter->tx_scrq) {
4037 for (i = 0; i < adapter->num_active_tx_scrqs; i++) {
4038 if (!adapter->tx_scrq[i])
4039 continue;
4040
4041 netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n",
4042 i);
4043 ibmvnic_tx_scrq_clean_buffer(adapter, tx_scrq: adapter->tx_scrq[i]);
4044 if (adapter->tx_scrq[i]->irq) {
4045 free_irq(adapter->tx_scrq[i]->irq,
4046 adapter->tx_scrq[i]);
4047 irq_dispose_mapping(virq: adapter->tx_scrq[i]->irq);
4048 adapter->tx_scrq[i]->irq = 0;
4049 }
4050
4051 release_sub_crq_queue(adapter, scrq: adapter->tx_scrq[i],
4052 do_h_free);
4053 }
4054
4055 kfree(objp: adapter->tx_scrq);
4056 adapter->tx_scrq = NULL;
4057 adapter->num_active_tx_scrqs = 0;
4058 }
4059
4060 if (adapter->rx_scrq) {
4061 for (i = 0; i < adapter->num_active_rx_scrqs; i++) {
4062 if (!adapter->rx_scrq[i])
4063 continue;
4064
4065 netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n",
4066 i);
4067 if (adapter->rx_scrq[i]->irq) {
4068 free_irq(adapter->rx_scrq[i]->irq,
4069 adapter->rx_scrq[i]);
4070 irq_dispose_mapping(virq: adapter->rx_scrq[i]->irq);
4071 adapter->rx_scrq[i]->irq = 0;
4072 }
4073
4074 release_sub_crq_queue(adapter, scrq: adapter->rx_scrq[i],
4075 do_h_free);
4076 }
4077
4078 kfree(objp: adapter->rx_scrq);
4079 adapter->rx_scrq = NULL;
4080 adapter->num_active_rx_scrqs = 0;
4081 }
4082}
4083
4084static int disable_scrq_irq(struct ibmvnic_adapter *adapter,
4085 struct ibmvnic_sub_crq_queue *scrq)
4086{
4087 struct device *dev = &adapter->vdev->dev;
4088 unsigned long rc;
4089
4090 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
4091 H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
4092 if (rc)
4093 dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n",
4094 scrq->hw_irq, rc);
4095 return rc;
4096}
4097
4098/* We can not use the IRQ chip EOI handler because that has the
4099 * unintended effect of changing the interrupt priority.
4100 */
4101static void ibmvnic_xics_eoi(struct device *dev, struct ibmvnic_sub_crq_queue *scrq)
4102{
4103 u64 val = 0xff000000 | scrq->hw_irq;
4104 unsigned long rc;
4105
4106 rc = plpar_hcall_norets(H_EOI, val);
4107 if (rc)
4108 dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n", val, rc);
4109}
4110
4111/* Due to a firmware bug, the hypervisor can send an interrupt to a
4112 * transmit or receive queue just prior to a partition migration.
4113 * Force an EOI after migration.
4114 */
4115static void ibmvnic_clear_pending_interrupt(struct device *dev,
4116 struct ibmvnic_sub_crq_queue *scrq)
4117{
4118 if (!xive_enabled())
4119 ibmvnic_xics_eoi(dev, scrq);
4120}
4121
4122static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
4123 struct ibmvnic_sub_crq_queue *scrq)
4124{
4125 struct device *dev = &adapter->vdev->dev;
4126 unsigned long rc;
4127
4128 if (scrq->hw_irq > 0x100000000ULL) {
4129 dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq);
4130 return 1;
4131 }
4132
4133 if (test_bit(0, &adapter->resetting) &&
4134 adapter->reset_reason == VNIC_RESET_MOBILITY) {
4135 ibmvnic_clear_pending_interrupt(dev, scrq);
4136 }
4137
4138 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
4139 H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
4140 if (rc)
4141 dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n",
4142 scrq->hw_irq, rc);
4143 return rc;
4144}
4145
4146static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter,
4147 struct ibmvnic_sub_crq_queue *scrq)
4148{
4149 struct device *dev = &adapter->vdev->dev;
4150 struct ibmvnic_tx_pool *tx_pool;
4151 struct ibmvnic_tx_buff *txbuff;
4152 struct netdev_queue *txq;
4153 union sub_crq *next;
4154 int index;
4155 int i;
4156
4157restart_loop:
4158 while (pending_scrq(adapter, scrq)) {
4159 unsigned int pool = scrq->pool_index;
4160 int num_entries = 0;
4161 int total_bytes = 0;
4162 int num_packets = 0;
4163
4164 next = ibmvnic_next_scrq(adapter, scrq);
4165 for (i = 0; i < next->tx_comp.num_comps; i++) {
4166 index = be32_to_cpu(next->tx_comp.correlators[i]);
4167 if (index & IBMVNIC_TSO_POOL_MASK) {
4168 tx_pool = &adapter->tso_pool[pool];
4169 index &= ~IBMVNIC_TSO_POOL_MASK;
4170 } else {
4171 tx_pool = &adapter->tx_pool[pool];
4172 }
4173
4174 txbuff = &tx_pool->tx_buff[index];
4175 num_packets++;
4176 num_entries += txbuff->num_entries;
4177 if (txbuff->skb) {
4178 total_bytes += txbuff->skb->len;
4179 if (next->tx_comp.rcs[i]) {
4180 dev_err(dev, "tx error %x\n",
4181 next->tx_comp.rcs[i]);
4182 dev_kfree_skb_irq(skb: txbuff->skb);
4183 } else {
4184 dev_consume_skb_irq(skb: txbuff->skb);
4185 }
4186 txbuff->skb = NULL;
4187 } else {
4188 netdev_warn(dev: adapter->netdev,
4189 format: "TX completion received with NULL socket buffer\n");
4190 }
4191 tx_pool->free_map[tx_pool->producer_index] = index;
4192 tx_pool->producer_index =
4193 (tx_pool->producer_index + 1) %
4194 tx_pool->num_buffers;
4195 }
4196 /* remove tx_comp scrq*/
4197 next->tx_comp.first = 0;
4198
4199 txq = netdev_get_tx_queue(dev: adapter->netdev, index: scrq->pool_index);
4200 netdev_tx_completed_queue(dev_queue: txq, pkts: num_packets, bytes: total_bytes);
4201
4202 if (atomic_sub_return(i: num_entries, v: &scrq->used) <=
4203 (adapter->req_tx_entries_per_subcrq / 2) &&
4204 __netif_subqueue_stopped(dev: adapter->netdev,
4205 queue_index: scrq->pool_index)) {
4206 rcu_read_lock();
4207 if (adapter->tx_queues_active) {
4208 netif_wake_subqueue(dev: adapter->netdev,
4209 queue_index: scrq->pool_index);
4210 netdev_dbg(adapter->netdev,
4211 "Started queue %d\n",
4212 scrq->pool_index);
4213 }
4214 rcu_read_unlock();
4215 }
4216 }
4217
4218 enable_scrq_irq(adapter, scrq);
4219
4220 if (pending_scrq(adapter, scrq)) {
4221 disable_scrq_irq(adapter, scrq);
4222 goto restart_loop;
4223 }
4224
4225 return 0;
4226}
4227
4228static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance)
4229{
4230 struct ibmvnic_sub_crq_queue *scrq = instance;
4231 struct ibmvnic_adapter *adapter = scrq->adapter;
4232
4233 disable_scrq_irq(adapter, scrq);
4234 ibmvnic_complete_tx(adapter, scrq);
4235
4236 return IRQ_HANDLED;
4237}
4238
4239static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance)
4240{
4241 struct ibmvnic_sub_crq_queue *scrq = instance;
4242 struct ibmvnic_adapter *adapter = scrq->adapter;
4243
4244 /* When booting a kdump kernel we can hit pending interrupts
4245 * prior to completing driver initialization.
4246 */
4247 if (unlikely(adapter->state != VNIC_OPEN))
4248 return IRQ_NONE;
4249
4250 adapter->rx_stats_buffers[scrq->scrq_num].interrupts++;
4251
4252 if (napi_schedule_prep(n: &adapter->napi[scrq->scrq_num])) {
4253 disable_scrq_irq(adapter, scrq);
4254 __napi_schedule(n: &adapter->napi[scrq->scrq_num]);
4255 }
4256
4257 return IRQ_HANDLED;
4258}
4259
4260static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter)
4261{
4262 struct device *dev = &adapter->vdev->dev;
4263 struct ibmvnic_sub_crq_queue *scrq;
4264 int i = 0, j = 0;
4265 int rc = 0;
4266
4267 for (i = 0; i < adapter->req_tx_queues; i++) {
4268 netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n",
4269 i);
4270 scrq = adapter->tx_scrq[i];
4271 scrq->irq = irq_create_mapping(NULL, hwirq: scrq->hw_irq);
4272
4273 if (!scrq->irq) {
4274 rc = -EINVAL;
4275 dev_err(dev, "Error mapping irq\n");
4276 goto req_tx_irq_failed;
4277 }
4278
4279 snprintf(buf: scrq->name, size: sizeof(scrq->name), fmt: "ibmvnic-%x-tx%d",
4280 adapter->vdev->unit_address, i);
4281 rc = request_irq(irq: scrq->irq, handler: ibmvnic_interrupt_tx,
4282 flags: 0, name: scrq->name, dev: scrq);
4283
4284 if (rc) {
4285 dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n",
4286 scrq->irq, rc);
4287 irq_dispose_mapping(virq: scrq->irq);
4288 goto req_tx_irq_failed;
4289 }
4290 }
4291
4292 for (i = 0; i < adapter->req_rx_queues; i++) {
4293 netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n",
4294 i);
4295 scrq = adapter->rx_scrq[i];
4296 scrq->irq = irq_create_mapping(NULL, hwirq: scrq->hw_irq);
4297 if (!scrq->irq) {
4298 rc = -EINVAL;
4299 dev_err(dev, "Error mapping irq\n");
4300 goto req_rx_irq_failed;
4301 }
4302 snprintf(buf: scrq->name, size: sizeof(scrq->name), fmt: "ibmvnic-%x-rx%d",
4303 adapter->vdev->unit_address, i);
4304 rc = request_irq(irq: scrq->irq, handler: ibmvnic_interrupt_rx,
4305 flags: 0, name: scrq->name, dev: scrq);
4306 if (rc) {
4307 dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n",
4308 scrq->irq, rc);
4309 irq_dispose_mapping(virq: scrq->irq);
4310 goto req_rx_irq_failed;
4311 }
4312 }
4313
4314 cpus_read_lock();
4315 ibmvnic_set_affinity(adapter);
4316 cpus_read_unlock();
4317
4318 return rc;
4319
4320req_rx_irq_failed:
4321 for (j = 0; j < i; j++) {
4322 free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]);
4323 irq_dispose_mapping(virq: adapter->rx_scrq[j]->irq);
4324 }
4325 i = adapter->req_tx_queues;
4326req_tx_irq_failed:
4327 for (j = 0; j < i; j++) {
4328 free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]);
4329 irq_dispose_mapping(virq: adapter->tx_scrq[j]->irq);
4330 }
4331 release_sub_crqs(adapter, do_h_free: 1);
4332 return rc;
4333}
4334
4335static int init_sub_crqs(struct ibmvnic_adapter *adapter)
4336{
4337 struct device *dev = &adapter->vdev->dev;
4338 struct ibmvnic_sub_crq_queue **allqueues;
4339 int registered_queues = 0;
4340 int total_queues;
4341 int more = 0;
4342 int i;
4343
4344 total_queues = adapter->req_tx_queues + adapter->req_rx_queues;
4345
4346 allqueues = kcalloc(n: total_queues, size: sizeof(*allqueues), GFP_KERNEL);
4347 if (!allqueues)
4348 return -ENOMEM;
4349
4350 for (i = 0; i < total_queues; i++) {
4351 allqueues[i] = init_sub_crq_queue(adapter);
4352 if (!allqueues[i]) {
4353 dev_warn(dev, "Couldn't allocate all sub-crqs\n");
4354 break;
4355 }
4356 registered_queues++;
4357 }
4358
4359 /* Make sure we were able to register the minimum number of queues */
4360 if (registered_queues <
4361 adapter->min_tx_queues + adapter->min_rx_queues) {
4362 dev_err(dev, "Fatal: Couldn't init min number of sub-crqs\n");
4363 goto tx_failed;
4364 }
4365
4366 /* Distribute the failed allocated queues*/
4367 for (i = 0; i < total_queues - registered_queues + more ; i++) {
4368 netdev_dbg(adapter->netdev, "Reducing number of queues\n");
4369 switch (i % 3) {
4370 case 0:
4371 if (adapter->req_rx_queues > adapter->min_rx_queues)
4372 adapter->req_rx_queues--;
4373 else
4374 more++;
4375 break;
4376 case 1:
4377 if (adapter->req_tx_queues > adapter->min_tx_queues)
4378 adapter->req_tx_queues--;
4379 else
4380 more++;
4381 break;
4382 }
4383 }
4384
4385 adapter->tx_scrq = kcalloc(n: adapter->req_tx_queues,
4386 size: sizeof(*adapter->tx_scrq), GFP_KERNEL);
4387 if (!adapter->tx_scrq)
4388 goto tx_failed;
4389
4390 for (i = 0; i < adapter->req_tx_queues; i++) {
4391 adapter->tx_scrq[i] = allqueues[i];
4392 adapter->tx_scrq[i]->pool_index = i;
4393 adapter->num_active_tx_scrqs++;
4394 }
4395
4396 adapter->rx_scrq = kcalloc(n: adapter->req_rx_queues,
4397 size: sizeof(*adapter->rx_scrq), GFP_KERNEL);
4398 if (!adapter->rx_scrq)
4399 goto rx_failed;
4400
4401 for (i = 0; i < adapter->req_rx_queues; i++) {
4402 adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues];
4403 adapter->rx_scrq[i]->scrq_num = i;
4404 adapter->num_active_rx_scrqs++;
4405 }
4406
4407 kfree(objp: allqueues);
4408 return 0;
4409
4410rx_failed:
4411 kfree(objp: adapter->tx_scrq);
4412 adapter->tx_scrq = NULL;
4413tx_failed:
4414 for (i = 0; i < registered_queues; i++)
4415 release_sub_crq_queue(adapter, scrq: allqueues[i], do_h_free: 1);
4416 kfree(objp: allqueues);
4417 return -ENOMEM;
4418}
4419
4420static void send_request_cap(struct ibmvnic_adapter *adapter, int retry)
4421{
4422 struct device *dev = &adapter->vdev->dev;
4423 union ibmvnic_crq crq;
4424 int max_entries;
4425 int cap_reqs;
4426
4427 /* We send out 6 or 7 REQUEST_CAPABILITY CRQs below (depending on
4428 * the PROMISC flag). Initialize this count upfront. When the tasklet
4429 * receives a response to all of these, it will send the next protocol
4430 * message (QUERY_IP_OFFLOAD).
4431 */
4432 if (!(adapter->netdev->flags & IFF_PROMISC) ||
4433 adapter->promisc_supported)
4434 cap_reqs = 7;
4435 else
4436 cap_reqs = 6;
4437
4438 if (!retry) {
4439 /* Sub-CRQ entries are 32 byte long */
4440 int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4);
4441
4442 atomic_set(v: &adapter->running_cap_crqs, i: cap_reqs);
4443
4444 if (adapter->min_tx_entries_per_subcrq > entries_page ||
4445 adapter->min_rx_add_entries_per_subcrq > entries_page) {
4446 dev_err(dev, "Fatal, invalid entries per sub-crq\n");
4447 return;
4448 }
4449
4450 if (adapter->desired.mtu)
4451 adapter->req_mtu = adapter->desired.mtu;
4452 else
4453 adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN;
4454
4455 if (!adapter->desired.tx_entries)
4456 adapter->desired.tx_entries =
4457 adapter->max_tx_entries_per_subcrq;
4458 if (!adapter->desired.rx_entries)
4459 adapter->desired.rx_entries =
4460 adapter->max_rx_add_entries_per_subcrq;
4461
4462 max_entries = IBMVNIC_LTB_SET_SIZE /
4463 (adapter->req_mtu + IBMVNIC_BUFFER_HLEN);
4464
4465 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
4466 adapter->desired.tx_entries > IBMVNIC_LTB_SET_SIZE) {
4467 adapter->desired.tx_entries = max_entries;
4468 }
4469
4470 if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
4471 adapter->desired.rx_entries > IBMVNIC_LTB_SET_SIZE) {
4472 adapter->desired.rx_entries = max_entries;
4473 }
4474
4475 if (adapter->desired.tx_entries)
4476 adapter->req_tx_entries_per_subcrq =
4477 adapter->desired.tx_entries;
4478 else
4479 adapter->req_tx_entries_per_subcrq =
4480 adapter->max_tx_entries_per_subcrq;
4481
4482 if (adapter->desired.rx_entries)
4483 adapter->req_rx_add_entries_per_subcrq =
4484 adapter->desired.rx_entries;
4485 else
4486 adapter->req_rx_add_entries_per_subcrq =
4487 adapter->max_rx_add_entries_per_subcrq;
4488
4489 if (adapter->desired.tx_queues)
4490 adapter->req_tx_queues =
4491 adapter->desired.tx_queues;
4492 else
4493 adapter->req_tx_queues =
4494 adapter->opt_tx_comp_sub_queues;
4495
4496 if (adapter->desired.rx_queues)
4497 adapter->req_rx_queues =
4498 adapter->desired.rx_queues;
4499 else
4500 adapter->req_rx_queues =
4501 adapter->opt_rx_comp_queues;
4502
4503 adapter->req_rx_add_queues = adapter->max_rx_add_queues;
4504 } else {
4505 atomic_add(i: cap_reqs, v: &adapter->running_cap_crqs);
4506 }
4507 memset(&crq, 0, sizeof(crq));
4508 crq.request_capability.first = IBMVNIC_CRQ_CMD;
4509 crq.request_capability.cmd = REQUEST_CAPABILITY;
4510
4511 crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES);
4512 crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues);
4513 cap_reqs--;
4514 ibmvnic_send_crq(adapter, &crq);
4515
4516 crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES);
4517 crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues);
4518 cap_reqs--;
4519 ibmvnic_send_crq(adapter, &crq);
4520
4521 crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES);
4522 crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues);
4523 cap_reqs--;
4524 ibmvnic_send_crq(adapter, &crq);
4525
4526 crq.request_capability.capability =
4527 cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ);
4528 crq.request_capability.number =
4529 cpu_to_be64(adapter->req_tx_entries_per_subcrq);
4530 cap_reqs--;
4531 ibmvnic_send_crq(adapter, &crq);
4532
4533 crq.request_capability.capability =
4534 cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ);
4535 crq.request_capability.number =
4536 cpu_to_be64(adapter->req_rx_add_entries_per_subcrq);
4537 cap_reqs--;
4538 ibmvnic_send_crq(adapter, &crq);
4539
4540 crq.request_capability.capability = cpu_to_be16(REQ_MTU);
4541 crq.request_capability.number = cpu_to_be64(adapter->req_mtu);
4542 cap_reqs--;
4543 ibmvnic_send_crq(adapter, &crq);
4544
4545 if (adapter->netdev->flags & IFF_PROMISC) {
4546 if (adapter->promisc_supported) {
4547 crq.request_capability.capability =
4548 cpu_to_be16(PROMISC_REQUESTED);
4549 crq.request_capability.number = cpu_to_be64(1);
4550 cap_reqs--;
4551 ibmvnic_send_crq(adapter, &crq);
4552 }
4553 } else {
4554 crq.request_capability.capability =
4555 cpu_to_be16(PROMISC_REQUESTED);
4556 crq.request_capability.number = cpu_to_be64(0);
4557 cap_reqs--;
4558 ibmvnic_send_crq(adapter, &crq);
4559 }
4560
4561 /* Keep at end to catch any discrepancy between expected and actual
4562 * CRQs sent.
4563 */
4564 WARN_ON(cap_reqs != 0);
4565}
4566
4567static int pending_scrq(struct ibmvnic_adapter *adapter,
4568 struct ibmvnic_sub_crq_queue *scrq)
4569{
4570 union sub_crq *entry = &scrq->msgs[scrq->cur];
4571 int rc;
4572
4573 rc = !!(entry->generic.first & IBMVNIC_CRQ_CMD_RSP);
4574
4575 /* Ensure that the SCRQ valid flag is loaded prior to loading the
4576 * contents of the SCRQ descriptor
4577 */
4578 dma_rmb();
4579
4580 return rc;
4581}
4582
4583static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter,
4584 struct ibmvnic_sub_crq_queue *scrq)
4585{
4586 union sub_crq *entry;
4587 unsigned long flags;
4588
4589 spin_lock_irqsave(&scrq->lock, flags);
4590 entry = &scrq->msgs[scrq->cur];
4591 if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) {
4592 if (++scrq->cur == scrq->size)
4593 scrq->cur = 0;
4594 } else {
4595 entry = NULL;
4596 }
4597 spin_unlock_irqrestore(lock: &scrq->lock, flags);
4598
4599 /* Ensure that the SCRQ valid flag is loaded prior to loading the
4600 * contents of the SCRQ descriptor
4601 */
4602 dma_rmb();
4603
4604 return entry;
4605}
4606
4607static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter)
4608{
4609 struct ibmvnic_crq_queue *queue = &adapter->crq;
4610 union ibmvnic_crq *crq;
4611
4612 crq = &queue->msgs[queue->cur];
4613 if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) {
4614 if (++queue->cur == queue->size)
4615 queue->cur = 0;
4616 } else {
4617 crq = NULL;
4618 }
4619
4620 return crq;
4621}
4622
4623static void print_subcrq_error(struct device *dev, int rc, const char *func)
4624{
4625 switch (rc) {
4626 case H_PARAMETER:
4627 dev_warn_ratelimited(dev,
4628 "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n",
4629 func, rc);
4630 break;
4631 case H_CLOSED:
4632 dev_warn_ratelimited(dev,
4633 "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n",
4634 func, rc);
4635 break;
4636 default:
4637 dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc);
4638 break;
4639 }
4640}
4641
4642static int send_subcrq_indirect(struct ibmvnic_adapter *adapter,
4643 u64 remote_handle, u64 ioba, u64 num_entries)
4644{
4645 unsigned int ua = adapter->vdev->unit_address;
4646 struct device *dev = &adapter->vdev->dev;
4647 int rc;
4648
4649 /* Make sure the hypervisor sees the complete request */
4650 dma_wmb();
4651 rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua,
4652 cpu_to_be64(remote_handle),
4653 ioba, num_entries);
4654
4655 if (rc)
4656 print_subcrq_error(dev, rc, func: __func__);
4657
4658 return rc;
4659}
4660
4661static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter,
4662 union ibmvnic_crq *crq)
4663{
4664 unsigned int ua = adapter->vdev->unit_address;
4665 struct device *dev = &adapter->vdev->dev;
4666 u64 *u64_crq = (u64 *)crq;
4667 int rc;
4668
4669 netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n",
4670 (unsigned long)cpu_to_be64(u64_crq[0]),
4671 (unsigned long)cpu_to_be64(u64_crq[1]));
4672
4673 if (!adapter->crq.active &&
4674 crq->generic.first != IBMVNIC_CRQ_INIT_CMD) {
4675 dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n");
4676 return -EINVAL;
4677 }
4678
4679 /* Make sure the hypervisor sees the complete request */
4680 dma_wmb();
4681
4682 rc = plpar_hcall_norets(H_SEND_CRQ, ua,
4683 cpu_to_be64(u64_crq[0]),
4684 cpu_to_be64(u64_crq[1]));
4685
4686 if (rc) {
4687 if (rc == H_CLOSED) {
4688 dev_warn(dev, "CRQ Queue closed\n");
4689 /* do not reset, report the fail, wait for passive init from server */
4690 }
4691
4692 dev_warn(dev, "Send error (rc=%d)\n", rc);
4693 }
4694
4695 return rc;
4696}
4697
4698static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter)
4699{
4700 struct device *dev = &adapter->vdev->dev;
4701 union ibmvnic_crq crq;
4702 int retries = 100;
4703 int rc;
4704
4705 memset(&crq, 0, sizeof(crq));
4706 crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
4707 crq.generic.cmd = IBMVNIC_CRQ_INIT;
4708 netdev_dbg(adapter->netdev, "Sending CRQ init\n");
4709
4710 do {
4711 rc = ibmvnic_send_crq(adapter, crq: &crq);
4712 if (rc != H_CLOSED)
4713 break;
4714 retries--;
4715 msleep(msecs: 50);
4716
4717 } while (retries > 0);
4718
4719 if (rc) {
4720 dev_err(dev, "Failed to send init request, rc = %d\n", rc);
4721 return rc;
4722 }
4723
4724 return 0;
4725}
4726
4727struct vnic_login_client_data {
4728 u8 type;
4729 __be16 len;
4730 char name[];
4731} __packed;
4732
4733static int vnic_client_data_len(struct ibmvnic_adapter *adapter)
4734{
4735 int len;
4736
4737 /* Calculate the amount of buffer space needed for the
4738 * vnic client data in the login buffer. There are four entries,
4739 * OS name, LPAR name, device name, and a null last entry.
4740 */
4741 len = 4 * sizeof(struct vnic_login_client_data);
4742 len += 6; /* "Linux" plus NULL */
4743 len += strlen(utsname()->nodename) + 1;
4744 len += strlen(adapter->netdev->name) + 1;
4745
4746 return len;
4747}
4748
4749static void vnic_add_client_data(struct ibmvnic_adapter *adapter,
4750 struct vnic_login_client_data *vlcd)
4751{
4752 const char *os_name = "Linux";
4753 int len;
4754
4755 /* Type 1 - LPAR OS */
4756 vlcd->type = 1;
4757 len = strlen(os_name) + 1;
4758 vlcd->len = cpu_to_be16(len);
4759 strscpy(p: vlcd->name, q: os_name, size: len);
4760 vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
4761
4762 /* Type 2 - LPAR name */
4763 vlcd->type = 2;
4764 len = strlen(utsname()->nodename) + 1;
4765 vlcd->len = cpu_to_be16(len);
4766 strscpy(p: vlcd->name, q: utsname()->nodename, size: len);
4767 vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
4768
4769 /* Type 3 - device name */
4770 vlcd->type = 3;
4771 len = strlen(adapter->netdev->name) + 1;
4772 vlcd->len = cpu_to_be16(len);
4773 strscpy(p: vlcd->name, q: adapter->netdev->name, size: len);
4774}
4775
4776static int send_login(struct ibmvnic_adapter *adapter)
4777{
4778 struct ibmvnic_login_rsp_buffer *login_rsp_buffer;
4779 struct ibmvnic_login_buffer *login_buffer;
4780 struct device *dev = &adapter->vdev->dev;
4781 struct vnic_login_client_data *vlcd;
4782 dma_addr_t rsp_buffer_token;
4783 dma_addr_t buffer_token;
4784 size_t rsp_buffer_size;
4785 union ibmvnic_crq crq;
4786 int client_data_len;
4787 size_t buffer_size;
4788 __be64 *tx_list_p;
4789 __be64 *rx_list_p;
4790 int rc;
4791 int i;
4792
4793 if (!adapter->tx_scrq || !adapter->rx_scrq) {
4794 netdev_err(dev: adapter->netdev,
4795 format: "RX or TX queues are not allocated, device login failed\n");
4796 return -ENOMEM;
4797 }
4798
4799 release_login_buffer(adapter);
4800 release_login_rsp_buffer(adapter);
4801
4802 client_data_len = vnic_client_data_len(adapter);
4803
4804 buffer_size =
4805 sizeof(struct ibmvnic_login_buffer) +
4806 sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) +
4807 client_data_len;
4808
4809 login_buffer = kzalloc(size: buffer_size, GFP_ATOMIC);
4810 if (!login_buffer)
4811 goto buf_alloc_failed;
4812
4813 buffer_token = dma_map_single(dev, login_buffer, buffer_size,
4814 DMA_TO_DEVICE);
4815 if (dma_mapping_error(dev, dma_addr: buffer_token)) {
4816 dev_err(dev, "Couldn't map login buffer\n");
4817 goto buf_map_failed;
4818 }
4819
4820 rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) +
4821 sizeof(u64) * adapter->req_tx_queues +
4822 sizeof(u64) * adapter->req_rx_queues +
4823 sizeof(u64) * adapter->req_rx_queues +
4824 sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS;
4825
4826 login_rsp_buffer = kmalloc(size: rsp_buffer_size, GFP_ATOMIC);
4827 if (!login_rsp_buffer)
4828 goto buf_rsp_alloc_failed;
4829
4830 rsp_buffer_token = dma_map_single(dev, login_rsp_buffer,
4831 rsp_buffer_size, DMA_FROM_DEVICE);
4832 if (dma_mapping_error(dev, dma_addr: rsp_buffer_token)) {
4833 dev_err(dev, "Couldn't map login rsp buffer\n");
4834 goto buf_rsp_map_failed;
4835 }
4836
4837 adapter->login_buf = login_buffer;
4838 adapter->login_buf_token = buffer_token;
4839 adapter->login_buf_sz = buffer_size;
4840 adapter->login_rsp_buf = login_rsp_buffer;
4841 adapter->login_rsp_buf_token = rsp_buffer_token;
4842 adapter->login_rsp_buf_sz = rsp_buffer_size;
4843
4844 login_buffer->len = cpu_to_be32(buffer_size);
4845 login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB);
4846 login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues);
4847 login_buffer->off_txcomp_subcrqs =
4848 cpu_to_be32(sizeof(struct ibmvnic_login_buffer));
4849 login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues);
4850 login_buffer->off_rxcomp_subcrqs =
4851 cpu_to_be32(sizeof(struct ibmvnic_login_buffer) +
4852 sizeof(u64) * adapter->req_tx_queues);
4853 login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token);
4854 login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size);
4855
4856 tx_list_p = (__be64 *)((char *)login_buffer +
4857 sizeof(struct ibmvnic_login_buffer));
4858 rx_list_p = (__be64 *)((char *)login_buffer +
4859 sizeof(struct ibmvnic_login_buffer) +
4860 sizeof(u64) * adapter->req_tx_queues);
4861
4862 for (i = 0; i < adapter->req_tx_queues; i++) {
4863 if (adapter->tx_scrq[i]) {
4864 tx_list_p[i] =
4865 cpu_to_be64(adapter->tx_scrq[i]->crq_num);
4866 }
4867 }
4868
4869 for (i = 0; i < adapter->req_rx_queues; i++) {
4870 if (adapter->rx_scrq[i]) {
4871 rx_list_p[i] =
4872 cpu_to_be64(adapter->rx_scrq[i]->crq_num);
4873 }
4874 }
4875
4876 /* Insert vNIC login client data */
4877 vlcd = (struct vnic_login_client_data *)
4878 ((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues));
4879 login_buffer->client_data_offset =
4880 cpu_to_be32((char *)vlcd - (char *)login_buffer);
4881 login_buffer->client_data_len = cpu_to_be32(client_data_len);
4882
4883 vnic_add_client_data(adapter, vlcd);
4884
4885 netdev_dbg(adapter->netdev, "Login Buffer:\n");
4886 for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) {
4887 netdev_dbg(adapter->netdev, "%016lx\n",
4888 ((unsigned long *)(adapter->login_buf))[i]);
4889 }
4890
4891 memset(&crq, 0, sizeof(crq));
4892 crq.login.first = IBMVNIC_CRQ_CMD;
4893 crq.login.cmd = LOGIN;
4894 crq.login.ioba = cpu_to_be32(buffer_token);
4895 crq.login.len = cpu_to_be32(buffer_size);
4896
4897 adapter->login_pending = true;
4898 rc = ibmvnic_send_crq(adapter, crq: &crq);
4899 if (rc) {
4900 adapter->login_pending = false;
4901 netdev_err(dev: adapter->netdev, format: "Failed to send login, rc=%d\n", rc);
4902 goto buf_send_failed;
4903 }
4904
4905 return 0;
4906
4907buf_send_failed:
4908 dma_unmap_single(dev, rsp_buffer_token, rsp_buffer_size,
4909 DMA_FROM_DEVICE);
4910buf_rsp_map_failed:
4911 kfree(objp: login_rsp_buffer);
4912 adapter->login_rsp_buf = NULL;
4913buf_rsp_alloc_failed:
4914 dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE);
4915buf_map_failed:
4916 kfree(objp: login_buffer);
4917 adapter->login_buf = NULL;
4918buf_alloc_failed:
4919 return -ENOMEM;
4920}
4921
4922static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr,
4923 u32 len, u8 map_id)
4924{
4925 union ibmvnic_crq crq;
4926
4927 memset(&crq, 0, sizeof(crq));
4928 crq.request_map.first = IBMVNIC_CRQ_CMD;
4929 crq.request_map.cmd = REQUEST_MAP;
4930 crq.request_map.map_id = map_id;
4931 crq.request_map.ioba = cpu_to_be32(addr);
4932 crq.request_map.len = cpu_to_be32(len);
4933 return ibmvnic_send_crq(adapter, crq: &crq);
4934}
4935
4936static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id)
4937{
4938 union ibmvnic_crq crq;
4939
4940 memset(&crq, 0, sizeof(crq));
4941 crq.request_unmap.first = IBMVNIC_CRQ_CMD;
4942 crq.request_unmap.cmd = REQUEST_UNMAP;
4943 crq.request_unmap.map_id = map_id;
4944 return ibmvnic_send_crq(adapter, crq: &crq);
4945}
4946
4947static void send_query_map(struct ibmvnic_adapter *adapter)
4948{
4949 union ibmvnic_crq crq;
4950
4951 memset(&crq, 0, sizeof(crq));
4952 crq.query_map.first = IBMVNIC_CRQ_CMD;
4953 crq.query_map.cmd = QUERY_MAP;
4954 ibmvnic_send_crq(adapter, crq: &crq);
4955}
4956
4957/* Send a series of CRQs requesting various capabilities of the VNIC server */
4958static void send_query_cap(struct ibmvnic_adapter *adapter)
4959{
4960 union ibmvnic_crq crq;
4961 int cap_reqs;
4962
4963 /* We send out 25 QUERY_CAPABILITY CRQs below. Initialize this count
4964 * upfront. When the tasklet receives a response to all of these, it
4965 * can send out the next protocol messaage (REQUEST_CAPABILITY).
4966 */
4967 cap_reqs = 25;
4968
4969 atomic_set(v: &adapter->running_cap_crqs, i: cap_reqs);
4970
4971 memset(&crq, 0, sizeof(crq));
4972 crq.query_capability.first = IBMVNIC_CRQ_CMD;
4973 crq.query_capability.cmd = QUERY_CAPABILITY;
4974
4975 crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES);
4976 ibmvnic_send_crq(adapter, crq: &crq);
4977 cap_reqs--;
4978
4979 crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES);
4980 ibmvnic_send_crq(adapter, crq: &crq);
4981 cap_reqs--;
4982
4983 crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES);
4984 ibmvnic_send_crq(adapter, crq: &crq);
4985 cap_reqs--;
4986
4987 crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES);
4988 ibmvnic_send_crq(adapter, crq: &crq);
4989 cap_reqs--;
4990
4991 crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES);
4992 ibmvnic_send_crq(adapter, crq: &crq);
4993 cap_reqs--;
4994
4995 crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES);
4996 ibmvnic_send_crq(adapter, crq: &crq);
4997 cap_reqs--;
4998
4999 crq.query_capability.capability =
5000 cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ);
5001 ibmvnic_send_crq(adapter, crq: &crq);
5002 cap_reqs--;
5003
5004 crq.query_capability.capability =
5005 cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ);
5006 ibmvnic_send_crq(adapter, crq: &crq);
5007 cap_reqs--;
5008
5009 crq.query_capability.capability =
5010 cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ);
5011 ibmvnic_send_crq(adapter, crq: &crq);
5012 cap_reqs--;
5013
5014 crq.query_capability.capability =
5015 cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ);
5016 ibmvnic_send_crq(adapter, crq: &crq);
5017 cap_reqs--;
5018
5019 crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD);
5020 ibmvnic_send_crq(adapter, crq: &crq);
5021 cap_reqs--;
5022
5023 crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED);
5024 ibmvnic_send_crq(adapter, crq: &crq);
5025 cap_reqs--;
5026
5027 crq.query_capability.capability = cpu_to_be16(MIN_MTU);
5028 ibmvnic_send_crq(adapter, crq: &crq);
5029 cap_reqs--;
5030
5031 crq.query_capability.capability = cpu_to_be16(MAX_MTU);
5032 ibmvnic_send_crq(adapter, crq: &crq);
5033 cap_reqs--;
5034
5035 crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS);
5036 ibmvnic_send_crq(adapter, crq: &crq);
5037 cap_reqs--;
5038
5039 crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION);
5040 ibmvnic_send_crq(adapter, crq: &crq);
5041 cap_reqs--;
5042
5043 crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION);
5044 ibmvnic_send_crq(adapter, crq: &crq);
5045 cap_reqs--;
5046
5047 crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES);
5048 ibmvnic_send_crq(adapter, crq: &crq);
5049 cap_reqs--;
5050
5051 crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED);
5052 ibmvnic_send_crq(adapter, crq: &crq);
5053 cap_reqs--;
5054
5055 crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES);
5056 ibmvnic_send_crq(adapter, crq: &crq);
5057 cap_reqs--;
5058
5059 crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES);
5060 ibmvnic_send_crq(adapter, crq: &crq);
5061 cap_reqs--;
5062
5063 crq.query_capability.capability =
5064 cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q);
5065 ibmvnic_send_crq(adapter, crq: &crq);
5066 cap_reqs--;
5067
5068 crq.query_capability.capability =
5069 cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ);
5070 ibmvnic_send_crq(adapter, crq: &crq);
5071 cap_reqs--;
5072
5073 crq.query_capability.capability =
5074 cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ);
5075 ibmvnic_send_crq(adapter, crq: &crq);
5076 cap_reqs--;
5077
5078 crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ);
5079
5080 ibmvnic_send_crq(adapter, crq: &crq);
5081 cap_reqs--;
5082
5083 /* Keep at end to catch any discrepancy between expected and actual
5084 * CRQs sent.
5085 */
5086 WARN_ON(cap_reqs != 0);
5087}
5088
5089static void send_query_ip_offload(struct ibmvnic_adapter *adapter)
5090{
5091 int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer);
5092 struct device *dev = &adapter->vdev->dev;
5093 union ibmvnic_crq crq;
5094
5095 adapter->ip_offload_tok =
5096 dma_map_single(dev,
5097 &adapter->ip_offload_buf,
5098 buf_sz,
5099 DMA_FROM_DEVICE);
5100
5101 if (dma_mapping_error(dev, dma_addr: adapter->ip_offload_tok)) {
5102 if (!firmware_has_feature(FW_FEATURE_CMO))
5103 dev_err(dev, "Couldn't map offload buffer\n");
5104 return;
5105 }
5106
5107 memset(&crq, 0, sizeof(crq));
5108 crq.query_ip_offload.first = IBMVNIC_CRQ_CMD;
5109 crq.query_ip_offload.cmd = QUERY_IP_OFFLOAD;
5110 crq.query_ip_offload.len = cpu_to_be32(buf_sz);
5111 crq.query_ip_offload.ioba =
5112 cpu_to_be32(adapter->ip_offload_tok);
5113
5114 ibmvnic_send_crq(adapter, crq: &crq);
5115}
5116
5117static void send_control_ip_offload(struct ibmvnic_adapter *adapter)
5118{
5119 struct ibmvnic_control_ip_offload_buffer *ctrl_buf = &adapter->ip_offload_ctrl;
5120 struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
5121 struct device *dev = &adapter->vdev->dev;
5122 netdev_features_t old_hw_features = 0;
5123 union ibmvnic_crq crq;
5124
5125 adapter->ip_offload_ctrl_tok =
5126 dma_map_single(dev,
5127 ctrl_buf,
5128 sizeof(adapter->ip_offload_ctrl),
5129 DMA_TO_DEVICE);
5130
5131 if (dma_mapping_error(dev, dma_addr: adapter->ip_offload_ctrl_tok)) {
5132 dev_err(dev, "Couldn't map ip offload control buffer\n");
5133 return;
5134 }
5135
5136 ctrl_buf->len = cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
5137 ctrl_buf->version = cpu_to_be32(INITIAL_VERSION_IOB);
5138 ctrl_buf->ipv4_chksum = buf->ipv4_chksum;
5139 ctrl_buf->ipv6_chksum = buf->ipv6_chksum;
5140 ctrl_buf->tcp_ipv4_chksum = buf->tcp_ipv4_chksum;
5141 ctrl_buf->udp_ipv4_chksum = buf->udp_ipv4_chksum;
5142 ctrl_buf->tcp_ipv6_chksum = buf->tcp_ipv6_chksum;
5143 ctrl_buf->udp_ipv6_chksum = buf->udp_ipv6_chksum;
5144 ctrl_buf->large_tx_ipv4 = buf->large_tx_ipv4;
5145 ctrl_buf->large_tx_ipv6 = buf->large_tx_ipv6;
5146
5147 /* large_rx disabled for now, additional features needed */
5148 ctrl_buf->large_rx_ipv4 = 0;
5149 ctrl_buf->large_rx_ipv6 = 0;
5150
5151 if (adapter->state != VNIC_PROBING) {
5152 old_hw_features = adapter->netdev->hw_features;
5153 adapter->netdev->hw_features = 0;
5154 }
5155
5156 adapter->netdev->hw_features = NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
5157
5158 if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum)
5159 adapter->netdev->hw_features |= NETIF_F_IP_CSUM;
5160
5161 if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum)
5162 adapter->netdev->hw_features |= NETIF_F_IPV6_CSUM;
5163
5164 if ((adapter->netdev->features &
5165 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
5166 adapter->netdev->hw_features |= NETIF_F_RXCSUM;
5167
5168 if (buf->large_tx_ipv4)
5169 adapter->netdev->hw_features |= NETIF_F_TSO;
5170 if (buf->large_tx_ipv6)
5171 adapter->netdev->hw_features |= NETIF_F_TSO6;
5172
5173 if (adapter->state == VNIC_PROBING) {
5174 adapter->netdev->features |= adapter->netdev->hw_features;
5175 } else if (old_hw_features != adapter->netdev->hw_features) {
5176 netdev_features_t tmp = 0;
5177
5178 /* disable features no longer supported */
5179 adapter->netdev->features &= adapter->netdev->hw_features;
5180 /* turn on features now supported if previously enabled */
5181 tmp = (old_hw_features ^ adapter->netdev->hw_features) &
5182 adapter->netdev->hw_features;
5183 adapter->netdev->features |=
5184 tmp & adapter->netdev->wanted_features;
5185 }
5186
5187 memset(&crq, 0, sizeof(crq));
5188 crq.control_ip_offload.first = IBMVNIC_CRQ_CMD;
5189 crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD;
5190 crq.control_ip_offload.len =
5191 cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
5192 crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok);
5193 ibmvnic_send_crq(adapter, crq: &crq);
5194}
5195
5196static void handle_vpd_size_rsp(union ibmvnic_crq *crq,
5197 struct ibmvnic_adapter *adapter)
5198{
5199 struct device *dev = &adapter->vdev->dev;
5200
5201 if (crq->get_vpd_size_rsp.rc.code) {
5202 dev_err(dev, "Error retrieving VPD size, rc=%x\n",
5203 crq->get_vpd_size_rsp.rc.code);
5204 complete(&adapter->fw_done);
5205 return;
5206 }
5207
5208 adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len);
5209 complete(&adapter->fw_done);
5210}
5211
5212static void handle_vpd_rsp(union ibmvnic_crq *crq,
5213 struct ibmvnic_adapter *adapter)
5214{
5215 struct device *dev = &adapter->vdev->dev;
5216 unsigned char *substr = NULL;
5217 u8 fw_level_len = 0;
5218
5219 memset(adapter->fw_version, 0, 32);
5220
5221 dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len,
5222 DMA_FROM_DEVICE);
5223
5224 if (crq->get_vpd_rsp.rc.code) {
5225 dev_err(dev, "Error retrieving VPD from device, rc=%x\n",
5226 crq->get_vpd_rsp.rc.code);
5227 goto complete;
5228 }
5229
5230 /* get the position of the firmware version info
5231 * located after the ASCII 'RM' substring in the buffer
5232 */
5233 substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len);
5234 if (!substr) {
5235 dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n");
5236 goto complete;
5237 }
5238
5239 /* get length of firmware level ASCII substring */
5240 if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) {
5241 fw_level_len = *(substr + 2);
5242 } else {
5243 dev_info(dev, "Length of FW substr extrapolated VDP buff\n");
5244 goto complete;
5245 }
5246
5247 /* copy firmware version string from vpd into adapter */
5248 if ((substr + 3 + fw_level_len) <
5249 (adapter->vpd->buff + adapter->vpd->len)) {
5250 strscpy(p: adapter->fw_version, q: substr + 3,
5251 size: sizeof(adapter->fw_version));
5252 } else {
5253 dev_info(dev, "FW substr extrapolated VPD buff\n");
5254 }
5255
5256complete:
5257 if (adapter->fw_version[0] == '\0')
5258 strscpy(p: (char *)adapter->fw_version, q: "N/A", size: sizeof(adapter->fw_version));
5259 complete(&adapter->fw_done);
5260}
5261
5262static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter)
5263{
5264 struct device *dev = &adapter->vdev->dev;
5265 struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
5266 int i;
5267
5268 dma_unmap_single(dev, adapter->ip_offload_tok,
5269 sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE);
5270
5271 netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n");
5272 for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++)
5273 netdev_dbg(adapter->netdev, "%016lx\n",
5274 ((unsigned long *)(buf))[i]);
5275
5276 netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum);
5277 netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum);
5278 netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n",
5279 buf->tcp_ipv4_chksum);
5280 netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n",
5281 buf->tcp_ipv6_chksum);
5282 netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n",
5283 buf->udp_ipv4_chksum);
5284 netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n",
5285 buf->udp_ipv6_chksum);
5286 netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n",
5287 buf->large_tx_ipv4);
5288 netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n",
5289 buf->large_tx_ipv6);
5290 netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n",
5291 buf->large_rx_ipv4);
5292 netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n",
5293 buf->large_rx_ipv6);
5294 netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n",
5295 buf->max_ipv4_header_size);
5296 netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n",
5297 buf->max_ipv6_header_size);
5298 netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n",
5299 buf->max_tcp_header_size);
5300 netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n",
5301 buf->max_udp_header_size);
5302 netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n",
5303 buf->max_large_tx_size);
5304 netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n",
5305 buf->max_large_rx_size);
5306 netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n",
5307 buf->ipv6_extension_header);
5308 netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n",
5309 buf->tcp_pseudosum_req);
5310 netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n",
5311 buf->num_ipv6_ext_headers);
5312 netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n",
5313 buf->off_ipv6_ext_headers);
5314
5315 send_control_ip_offload(adapter);
5316}
5317
5318static const char *ibmvnic_fw_err_cause(u16 cause)
5319{
5320 switch (cause) {
5321 case ADAPTER_PROBLEM:
5322 return "adapter problem";
5323 case BUS_PROBLEM:
5324 return "bus problem";
5325 case FW_PROBLEM:
5326 return "firmware problem";
5327 case DD_PROBLEM:
5328 return "device driver problem";
5329 case EEH_RECOVERY:
5330 return "EEH recovery";
5331 case FW_UPDATED:
5332 return "firmware updated";
5333 case LOW_MEMORY:
5334 return "low Memory";
5335 default:
5336 return "unknown";
5337 }
5338}
5339
5340static void handle_error_indication(union ibmvnic_crq *crq,
5341 struct ibmvnic_adapter *adapter)
5342{
5343 struct device *dev = &adapter->vdev->dev;
5344 u16 cause;
5345
5346 cause = be16_to_cpu(crq->error_indication.error_cause);
5347
5348 dev_warn_ratelimited(dev,
5349 "Firmware reports %serror, cause: %s. Starting recovery...\n",
5350 crq->error_indication.flags
5351 & IBMVNIC_FATAL_ERROR ? "FATAL " : "",
5352 ibmvnic_fw_err_cause(cause));
5353
5354 if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR)
5355 ibmvnic_reset(adapter, reason: VNIC_RESET_FATAL);
5356 else
5357 ibmvnic_reset(adapter, reason: VNIC_RESET_NON_FATAL);
5358}
5359
5360static int handle_change_mac_rsp(union ibmvnic_crq *crq,
5361 struct ibmvnic_adapter *adapter)
5362{
5363 struct net_device *netdev = adapter->netdev;
5364 struct device *dev = &adapter->vdev->dev;
5365 long rc;
5366
5367 rc = crq->change_mac_addr_rsp.rc.code;
5368 if (rc) {
5369 dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc);
5370 goto out;
5371 }
5372 /* crq->change_mac_addr.mac_addr is the requested one
5373 * crq->change_mac_addr_rsp.mac_addr is the returned valid one.
5374 */
5375 eth_hw_addr_set(dev: netdev, addr: &crq->change_mac_addr_rsp.mac_addr[0]);
5376 ether_addr_copy(dst: adapter->mac_addr,
5377 src: &crq->change_mac_addr_rsp.mac_addr[0]);
5378out:
5379 complete(&adapter->fw_done);
5380 return rc;
5381}
5382
5383static void handle_request_cap_rsp(union ibmvnic_crq *crq,
5384 struct ibmvnic_adapter *adapter)
5385{
5386 struct device *dev = &adapter->vdev->dev;
5387 u64 *req_value;
5388 char *name;
5389
5390 atomic_dec(v: &adapter->running_cap_crqs);
5391 netdev_dbg(adapter->netdev, "Outstanding request-caps: %d\n",
5392 atomic_read(&adapter->running_cap_crqs));
5393 switch (be16_to_cpu(crq->request_capability_rsp.capability)) {
5394 case REQ_TX_QUEUES:
5395 req_value = &adapter->req_tx_queues;
5396 name = "tx";
5397 break;
5398 case REQ_RX_QUEUES:
5399 req_value = &adapter->req_rx_queues;
5400 name = "rx";
5401 break;
5402 case REQ_RX_ADD_QUEUES:
5403 req_value = &adapter->req_rx_add_queues;
5404 name = "rx_add";
5405 break;
5406 case REQ_TX_ENTRIES_PER_SUBCRQ:
5407 req_value = &adapter->req_tx_entries_per_subcrq;
5408 name = "tx_entries_per_subcrq";
5409 break;
5410 case REQ_RX_ADD_ENTRIES_PER_SUBCRQ:
5411 req_value = &adapter->req_rx_add_entries_per_subcrq;
5412 name = "rx_add_entries_per_subcrq";
5413 break;
5414 case REQ_MTU:
5415 req_value = &adapter->req_mtu;
5416 name = "mtu";
5417 break;
5418 case PROMISC_REQUESTED:
5419 req_value = &adapter->promisc;
5420 name = "promisc";
5421 break;
5422 default:
5423 dev_err(dev, "Got invalid cap request rsp %d\n",
5424 crq->request_capability.capability);
5425 return;
5426 }
5427
5428 switch (crq->request_capability_rsp.rc.code) {
5429 case SUCCESS:
5430 break;
5431 case PARTIALSUCCESS:
5432 dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n",
5433 *req_value,
5434 (long)be64_to_cpu(crq->request_capability_rsp.number),
5435 name);
5436
5437 if (be16_to_cpu(crq->request_capability_rsp.capability) ==
5438 REQ_MTU) {
5439 pr_err("mtu of %llu is not supported. Reverting.\n",
5440 *req_value);
5441 *req_value = adapter->fallback.mtu;
5442 } else {
5443 *req_value =
5444 be64_to_cpu(crq->request_capability_rsp.number);
5445 }
5446
5447 send_request_cap(adapter, retry: 1);
5448 return;
5449 default:
5450 dev_err(dev, "Error %d in request cap rsp\n",
5451 crq->request_capability_rsp.rc.code);
5452 return;
5453 }
5454
5455 /* Done receiving requested capabilities, query IP offload support */
5456 if (atomic_read(v: &adapter->running_cap_crqs) == 0)
5457 send_query_ip_offload(adapter);
5458}
5459
5460static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq,
5461 struct ibmvnic_adapter *adapter)
5462{
5463 struct device *dev = &adapter->vdev->dev;
5464 struct net_device *netdev = adapter->netdev;
5465 struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf;
5466 struct ibmvnic_login_buffer *login = adapter->login_buf;
5467 u64 *tx_handle_array;
5468 u64 *rx_handle_array;
5469 int num_tx_pools;
5470 int num_rx_pools;
5471 u64 *size_array;
5472 u32 rsp_len;
5473 int i;
5474
5475 /* CHECK: Test/set of login_pending does not need to be atomic
5476 * because only ibmvnic_tasklet tests/clears this.
5477 */
5478 if (!adapter->login_pending) {
5479 netdev_warn(dev: netdev, format: "Ignoring unexpected login response\n");
5480 return 0;
5481 }
5482 adapter->login_pending = false;
5483
5484 /* If the number of queues requested can't be allocated by the
5485 * server, the login response will return with code 1. We will need
5486 * to resend the login buffer with fewer queues requested.
5487 */
5488 if (login_rsp_crq->generic.rc.code) {
5489 adapter->init_done_rc = login_rsp_crq->generic.rc.code;
5490 complete(&adapter->init_done);
5491 return 0;
5492 }
5493
5494 if (adapter->failover_pending) {
5495 adapter->init_done_rc = -EAGAIN;
5496 netdev_dbg(netdev, "Failover pending, ignoring login response\n");
5497 complete(&adapter->init_done);
5498 /* login response buffer will be released on reset */
5499 return 0;
5500 }
5501
5502 netdev->mtu = adapter->req_mtu - ETH_HLEN;
5503
5504 netdev_dbg(adapter->netdev, "Login Response Buffer:\n");
5505 for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) {
5506 netdev_dbg(adapter->netdev, "%016lx\n",
5507 ((unsigned long *)(adapter->login_rsp_buf))[i]);
5508 }
5509
5510 /* Sanity checks */
5511 if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs ||
5512 (be32_to_cpu(login->num_rxcomp_subcrqs) *
5513 adapter->req_rx_add_queues !=
5514 be32_to_cpu(login_rsp->num_rxadd_subcrqs))) {
5515 dev_err(dev, "FATAL: Inconsistent login and login rsp\n");
5516 ibmvnic_reset(adapter, reason: VNIC_RESET_FATAL);
5517 return -EIO;
5518 }
5519
5520 rsp_len = be32_to_cpu(login_rsp->len);
5521 if (be32_to_cpu(login->login_rsp_len) < rsp_len ||
5522 rsp_len <= be32_to_cpu(login_rsp->off_txsubm_subcrqs) ||
5523 rsp_len <= be32_to_cpu(login_rsp->off_rxadd_subcrqs) ||
5524 rsp_len <= be32_to_cpu(login_rsp->off_rxadd_buff_size) ||
5525 rsp_len <= be32_to_cpu(login_rsp->off_supp_tx_desc)) {
5526 /* This can happen if a login request times out and there are
5527 * 2 outstanding login requests sent, the LOGIN_RSP crq
5528 * could have been for the older login request. So we are
5529 * parsing the newer response buffer which may be incomplete
5530 */
5531 dev_err(dev, "FATAL: Login rsp offsets/lengths invalid\n");
5532 ibmvnic_reset(adapter, reason: VNIC_RESET_FATAL);
5533 return -EIO;
5534 }
5535
5536 size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
5537 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size));
5538 /* variable buffer sizes are not supported, so just read the
5539 * first entry.
5540 */
5541 adapter->cur_rx_buf_sz = be64_to_cpu(size_array[0]);
5542
5543 num_tx_pools = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs);
5544 num_rx_pools = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
5545
5546 tx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
5547 be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs));
5548 rx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
5549 be32_to_cpu(adapter->login_rsp_buf->off_rxadd_subcrqs));
5550
5551 for (i = 0; i < num_tx_pools; i++)
5552 adapter->tx_scrq[i]->handle = tx_handle_array[i];
5553
5554 for (i = 0; i < num_rx_pools; i++)
5555 adapter->rx_scrq[i]->handle = rx_handle_array[i];
5556
5557 adapter->num_active_tx_scrqs = num_tx_pools;
5558 adapter->num_active_rx_scrqs = num_rx_pools;
5559 release_login_rsp_buffer(adapter);
5560 release_login_buffer(adapter);
5561 complete(&adapter->init_done);
5562
5563 return 0;
5564}
5565
5566static void handle_request_unmap_rsp(union ibmvnic_crq *crq,
5567 struct ibmvnic_adapter *adapter)
5568{
5569 struct device *dev = &adapter->vdev->dev;
5570 long rc;
5571
5572 rc = crq->request_unmap_rsp.rc.code;
5573 if (rc)
5574 dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc);
5575}
5576
5577static void handle_query_map_rsp(union ibmvnic_crq *crq,
5578 struct ibmvnic_adapter *adapter)
5579{
5580 struct net_device *netdev = adapter->netdev;
5581 struct device *dev = &adapter->vdev->dev;
5582 long rc;
5583
5584 rc = crq->query_map_rsp.rc.code;
5585 if (rc) {
5586 dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc);
5587 return;
5588 }
5589 netdev_dbg(netdev, "page_size = %d\ntot_pages = %u\nfree_pages = %u\n",
5590 crq->query_map_rsp.page_size,
5591 __be32_to_cpu(crq->query_map_rsp.tot_pages),
5592 __be32_to_cpu(crq->query_map_rsp.free_pages));
5593}
5594
5595static void handle_query_cap_rsp(union ibmvnic_crq *crq,
5596 struct ibmvnic_adapter *adapter)
5597{
5598 struct net_device *netdev = adapter->netdev;
5599 struct device *dev = &adapter->vdev->dev;
5600 long rc;
5601
5602 atomic_dec(v: &adapter->running_cap_crqs);
5603 netdev_dbg(netdev, "Outstanding queries: %d\n",
5604 atomic_read(&adapter->running_cap_crqs));
5605 rc = crq->query_capability.rc.code;
5606 if (rc) {
5607 dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc);
5608 goto out;
5609 }
5610
5611 switch (be16_to_cpu(crq->query_capability.capability)) {
5612 case MIN_TX_QUEUES:
5613 adapter->min_tx_queues =
5614 be64_to_cpu(crq->query_capability.number);
5615 netdev_dbg(netdev, "min_tx_queues = %lld\n",
5616 adapter->min_tx_queues);
5617 break;
5618 case MIN_RX_QUEUES:
5619 adapter->min_rx_queues =
5620 be64_to_cpu(crq->query_capability.number);
5621 netdev_dbg(netdev, "min_rx_queues = %lld\n",
5622 adapter->min_rx_queues);
5623 break;
5624 case MIN_RX_ADD_QUEUES:
5625 adapter->min_rx_add_queues =
5626 be64_to_cpu(crq->query_capability.number);
5627 netdev_dbg(netdev, "min_rx_add_queues = %lld\n",
5628 adapter->min_rx_add_queues);
5629 break;
5630 case MAX_TX_QUEUES:
5631 adapter->max_tx_queues =
5632 be64_to_cpu(crq->query_capability.number);
5633 netdev_dbg(netdev, "max_tx_queues = %lld\n",
5634 adapter->max_tx_queues);
5635 break;
5636 case MAX_RX_QUEUES:
5637 adapter->max_rx_queues =
5638 be64_to_cpu(crq->query_capability.number);
5639 netdev_dbg(netdev, "max_rx_queues = %lld\n",
5640 adapter->max_rx_queues);
5641 break;
5642 case MAX_RX_ADD_QUEUES:
5643 adapter->max_rx_add_queues =
5644 be64_to_cpu(crq->query_capability.number);
5645 netdev_dbg(netdev, "max_rx_add_queues = %lld\n",
5646 adapter->max_rx_add_queues);
5647 break;
5648 case MIN_TX_ENTRIES_PER_SUBCRQ:
5649 adapter->min_tx_entries_per_subcrq =
5650 be64_to_cpu(crq->query_capability.number);
5651 netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n",
5652 adapter->min_tx_entries_per_subcrq);
5653 break;
5654 case MIN_RX_ADD_ENTRIES_PER_SUBCRQ:
5655 adapter->min_rx_add_entries_per_subcrq =
5656 be64_to_cpu(crq->query_capability.number);
5657 netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n",
5658 adapter->min_rx_add_entries_per_subcrq);
5659 break;
5660 case MAX_TX_ENTRIES_PER_SUBCRQ:
5661 adapter->max_tx_entries_per_subcrq =
5662 be64_to_cpu(crq->query_capability.number);
5663 netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n",
5664 adapter->max_tx_entries_per_subcrq);
5665 break;
5666 case MAX_RX_ADD_ENTRIES_PER_SUBCRQ:
5667 adapter->max_rx_add_entries_per_subcrq =
5668 be64_to_cpu(crq->query_capability.number);
5669 netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n",
5670 adapter->max_rx_add_entries_per_subcrq);
5671 break;
5672 case TCP_IP_OFFLOAD:
5673 adapter->tcp_ip_offload =
5674 be64_to_cpu(crq->query_capability.number);
5675 netdev_dbg(netdev, "tcp_ip_offload = %lld\n",
5676 adapter->tcp_ip_offload);
5677 break;
5678 case PROMISC_SUPPORTED:
5679 adapter->promisc_supported =
5680 be64_to_cpu(crq->query_capability.number);
5681 netdev_dbg(netdev, "promisc_supported = %lld\n",
5682 adapter->promisc_supported);
5683 break;
5684 case MIN_MTU:
5685 adapter->min_mtu = be64_to_cpu(crq->query_capability.number);
5686 netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
5687 netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu);
5688 break;
5689 case MAX_MTU:
5690 adapter->max_mtu = be64_to_cpu(crq->query_capability.number);
5691 netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
5692 netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu);
5693 break;
5694 case MAX_MULTICAST_FILTERS:
5695 adapter->max_multicast_filters =
5696 be64_to_cpu(crq->query_capability.number);
5697 netdev_dbg(netdev, "max_multicast_filters = %lld\n",
5698 adapter->max_multicast_filters);
5699 break;
5700 case VLAN_HEADER_INSERTION:
5701 adapter->vlan_header_insertion =
5702 be64_to_cpu(crq->query_capability.number);
5703 if (adapter->vlan_header_insertion)
5704 netdev->features |= NETIF_F_HW_VLAN_STAG_TX;
5705 netdev_dbg(netdev, "vlan_header_insertion = %lld\n",
5706 adapter->vlan_header_insertion);
5707 break;
5708 case RX_VLAN_HEADER_INSERTION:
5709 adapter->rx_vlan_header_insertion =
5710 be64_to_cpu(crq->query_capability.number);
5711 netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n",
5712 adapter->rx_vlan_header_insertion);
5713 break;
5714 case MAX_TX_SG_ENTRIES:
5715 adapter->max_tx_sg_entries =
5716 be64_to_cpu(crq->query_capability.number);
5717 netdev_dbg(netdev, "max_tx_sg_entries = %lld\n",
5718 adapter->max_tx_sg_entries);
5719 break;
5720 case RX_SG_SUPPORTED:
5721 adapter->rx_sg_supported =
5722 be64_to_cpu(crq->query_capability.number);
5723 netdev_dbg(netdev, "rx_sg_supported = %lld\n",
5724 adapter->rx_sg_supported);
5725 break;
5726 case OPT_TX_COMP_SUB_QUEUES:
5727 adapter->opt_tx_comp_sub_queues =
5728 be64_to_cpu(crq->query_capability.number);
5729 netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n",
5730 adapter->opt_tx_comp_sub_queues);
5731 break;
5732 case OPT_RX_COMP_QUEUES:
5733 adapter->opt_rx_comp_queues =
5734 be64_to_cpu(crq->query_capability.number);
5735 netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n",
5736 adapter->opt_rx_comp_queues);
5737 break;
5738 case OPT_RX_BUFADD_Q_PER_RX_COMP_Q:
5739 adapter->opt_rx_bufadd_q_per_rx_comp_q =
5740 be64_to_cpu(crq->query_capability.number);
5741 netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n",
5742 adapter->opt_rx_bufadd_q_per_rx_comp_q);
5743 break;
5744 case OPT_TX_ENTRIES_PER_SUBCRQ:
5745 adapter->opt_tx_entries_per_subcrq =
5746 be64_to_cpu(crq->query_capability.number);
5747 netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n",
5748 adapter->opt_tx_entries_per_subcrq);
5749 break;
5750 case OPT_RXBA_ENTRIES_PER_SUBCRQ:
5751 adapter->opt_rxba_entries_per_subcrq =
5752 be64_to_cpu(crq->query_capability.number);
5753 netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n",
5754 adapter->opt_rxba_entries_per_subcrq);
5755 break;
5756 case TX_RX_DESC_REQ:
5757 adapter->tx_rx_desc_req = crq->query_capability.number;
5758 netdev_dbg(netdev, "tx_rx_desc_req = %llx\n",
5759 adapter->tx_rx_desc_req);
5760 break;
5761
5762 default:
5763 netdev_err(dev: netdev, format: "Got invalid cap rsp %d\n",
5764 crq->query_capability.capability);
5765 }
5766
5767out:
5768 if (atomic_read(v: &adapter->running_cap_crqs) == 0)
5769 send_request_cap(adapter, retry: 0);
5770}
5771
5772static int send_query_phys_parms(struct ibmvnic_adapter *adapter)
5773{
5774 union ibmvnic_crq crq;
5775 int rc;
5776
5777 memset(&crq, 0, sizeof(crq));
5778 crq.query_phys_parms.first = IBMVNIC_CRQ_CMD;
5779 crq.query_phys_parms.cmd = QUERY_PHYS_PARMS;
5780
5781 mutex_lock(&adapter->fw_lock);
5782 adapter->fw_done_rc = 0;
5783 reinit_completion(x: &adapter->fw_done);
5784
5785 rc = ibmvnic_send_crq(adapter, crq: &crq);
5786 if (rc) {
5787 mutex_unlock(lock: &adapter->fw_lock);
5788 return rc;
5789 }
5790
5791 rc = ibmvnic_wait_for_completion(adapter, comp_done: &adapter->fw_done, timeout: 10000);
5792 if (rc) {
5793 mutex_unlock(lock: &adapter->fw_lock);
5794 return rc;
5795 }
5796
5797 mutex_unlock(lock: &adapter->fw_lock);
5798 return adapter->fw_done_rc ? -EIO : 0;
5799}
5800
5801static int handle_query_phys_parms_rsp(union ibmvnic_crq *crq,
5802 struct ibmvnic_adapter *adapter)
5803{
5804 struct net_device *netdev = adapter->netdev;
5805 int rc;
5806 __be32 rspeed = cpu_to_be32(crq->query_phys_parms_rsp.speed);
5807
5808 rc = crq->query_phys_parms_rsp.rc.code;
5809 if (rc) {
5810 netdev_err(dev: netdev, format: "Error %d in QUERY_PHYS_PARMS\n", rc);
5811 return rc;
5812 }
5813 switch (rspeed) {
5814 case IBMVNIC_10MBPS:
5815 adapter->speed = SPEED_10;
5816 break;
5817 case IBMVNIC_100MBPS:
5818 adapter->speed = SPEED_100;
5819 break;
5820 case IBMVNIC_1GBPS:
5821 adapter->speed = SPEED_1000;
5822 break;
5823 case IBMVNIC_10GBPS:
5824 adapter->speed = SPEED_10000;
5825 break;
5826 case IBMVNIC_25GBPS:
5827 adapter->speed = SPEED_25000;
5828 break;
5829 case IBMVNIC_40GBPS:
5830 adapter->speed = SPEED_40000;
5831 break;
5832 case IBMVNIC_50GBPS:
5833 adapter->speed = SPEED_50000;
5834 break;
5835 case IBMVNIC_100GBPS:
5836 adapter->speed = SPEED_100000;
5837 break;
5838 case IBMVNIC_200GBPS:
5839 adapter->speed = SPEED_200000;
5840 break;
5841 default:
5842 if (netif_carrier_ok(dev: netdev))
5843 netdev_warn(dev: netdev, format: "Unknown speed 0x%08x\n", rspeed);
5844 adapter->speed = SPEED_UNKNOWN;
5845 }
5846 if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_FULL_DUPLEX)
5847 adapter->duplex = DUPLEX_FULL;
5848 else if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_HALF_DUPLEX)
5849 adapter->duplex = DUPLEX_HALF;
5850 else
5851 adapter->duplex = DUPLEX_UNKNOWN;
5852
5853 return rc;
5854}
5855
5856static void ibmvnic_handle_crq(union ibmvnic_crq *crq,
5857 struct ibmvnic_adapter *adapter)
5858{
5859 struct ibmvnic_generic_crq *gen_crq = &crq->generic;
5860 struct net_device *netdev = adapter->netdev;
5861 struct device *dev = &adapter->vdev->dev;
5862 u64 *u64_crq = (u64 *)crq;
5863 long rc;
5864
5865 netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n",
5866 (unsigned long)cpu_to_be64(u64_crq[0]),
5867 (unsigned long)cpu_to_be64(u64_crq[1]));
5868 switch (gen_crq->first) {
5869 case IBMVNIC_CRQ_INIT_RSP:
5870 switch (gen_crq->cmd) {
5871 case IBMVNIC_CRQ_INIT:
5872 dev_info(dev, "Partner initialized\n");
5873 adapter->from_passive_init = true;
5874 /* Discard any stale login responses from prev reset.
5875 * CHECK: should we clear even on INIT_COMPLETE?
5876 */
5877 adapter->login_pending = false;
5878
5879 if (adapter->state == VNIC_DOWN)
5880 rc = ibmvnic_reset(adapter, reason: VNIC_RESET_PASSIVE_INIT);
5881 else
5882 rc = ibmvnic_reset(adapter, reason: VNIC_RESET_FAILOVER);
5883
5884 if (rc && rc != -EBUSY) {
5885 /* We were unable to schedule the failover
5886 * reset either because the adapter was still
5887 * probing (eg: during kexec) or we could not
5888 * allocate memory. Clear the failover_pending
5889 * flag since no one else will. We ignore
5890 * EBUSY because it means either FAILOVER reset
5891 * is already scheduled or the adapter is
5892 * being removed.
5893 */
5894 netdev_err(dev: netdev,
5895 format: "Error %ld scheduling failover reset\n",
5896 rc);
5897 adapter->failover_pending = false;
5898 }
5899
5900 if (!completion_done(x: &adapter->init_done)) {
5901 if (!adapter->init_done_rc)
5902 adapter->init_done_rc = -EAGAIN;
5903 complete(&adapter->init_done);
5904 }
5905
5906 break;
5907 case IBMVNIC_CRQ_INIT_COMPLETE:
5908 dev_info(dev, "Partner initialization complete\n");
5909 adapter->crq.active = true;
5910 send_version_xchg(adapter);
5911 break;
5912 default:
5913 dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd);
5914 }
5915 return;
5916 case IBMVNIC_CRQ_XPORT_EVENT:
5917 netif_carrier_off(dev: netdev);
5918 adapter->crq.active = false;
5919 /* terminate any thread waiting for a response
5920 * from the device
5921 */
5922 if (!completion_done(x: &adapter->fw_done)) {
5923 adapter->fw_done_rc = -EIO;
5924 complete(&adapter->fw_done);
5925 }
5926
5927 /* if we got here during crq-init, retry crq-init */
5928 if (!completion_done(x: &adapter->init_done)) {
5929 adapter->init_done_rc = -EAGAIN;
5930 complete(&adapter->init_done);
5931 }
5932
5933 if (!completion_done(x: &adapter->stats_done))
5934 complete(&adapter->stats_done);
5935 if (test_bit(0, &adapter->resetting))
5936 adapter->force_reset_recovery = true;
5937 if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) {
5938 dev_info(dev, "Migrated, re-enabling adapter\n");
5939 ibmvnic_reset(adapter, reason: VNIC_RESET_MOBILITY);
5940 } else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) {
5941 dev_info(dev, "Backing device failover detected\n");
5942 adapter->failover_pending = true;
5943 } else {
5944 /* The adapter lost the connection */
5945 dev_err(dev, "Virtual Adapter failed (rc=%d)\n",
5946 gen_crq->cmd);
5947 ibmvnic_reset(adapter, reason: VNIC_RESET_FATAL);
5948 }
5949 return;
5950 case IBMVNIC_CRQ_CMD_RSP:
5951 break;
5952 default:
5953 dev_err(dev, "Got an invalid msg type 0x%02x\n",
5954 gen_crq->first);
5955 return;
5956 }
5957
5958 switch (gen_crq->cmd) {
5959 case VERSION_EXCHANGE_RSP:
5960 rc = crq->version_exchange_rsp.rc.code;
5961 if (rc) {
5962 dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc);
5963 break;
5964 }
5965 ibmvnic_version =
5966 be16_to_cpu(crq->version_exchange_rsp.version);
5967 dev_info(dev, "Partner protocol version is %d\n",
5968 ibmvnic_version);
5969 send_query_cap(adapter);
5970 break;
5971 case QUERY_CAPABILITY_RSP:
5972 handle_query_cap_rsp(crq, adapter);
5973 break;
5974 case QUERY_MAP_RSP:
5975 handle_query_map_rsp(crq, adapter);
5976 break;
5977 case REQUEST_MAP_RSP:
5978 adapter->fw_done_rc = crq->request_map_rsp.rc.code;
5979 complete(&adapter->fw_done);
5980 break;
5981 case REQUEST_UNMAP_RSP:
5982 handle_request_unmap_rsp(crq, adapter);
5983 break;
5984 case REQUEST_CAPABILITY_RSP:
5985 handle_request_cap_rsp(crq, adapter);
5986 break;
5987 case LOGIN_RSP:
5988 netdev_dbg(netdev, "Got Login Response\n");
5989 handle_login_rsp(login_rsp_crq: crq, adapter);
5990 break;
5991 case LOGICAL_LINK_STATE_RSP:
5992 netdev_dbg(netdev,
5993 "Got Logical Link State Response, state: %d rc: %d\n",
5994 crq->logical_link_state_rsp.link_state,
5995 crq->logical_link_state_rsp.rc.code);
5996 adapter->logical_link_state =
5997 crq->logical_link_state_rsp.link_state;
5998 adapter->init_done_rc = crq->logical_link_state_rsp.rc.code;
5999 complete(&adapter->init_done);
6000 break;
6001 case LINK_STATE_INDICATION:
6002 netdev_dbg(netdev, "Got Logical Link State Indication\n");
6003 adapter->phys_link_state =
6004 crq->link_state_indication.phys_link_state;
6005 adapter->logical_link_state =
6006 crq->link_state_indication.logical_link_state;
6007 if (adapter->phys_link_state && adapter->logical_link_state)
6008 netif_carrier_on(dev: netdev);
6009 else
6010 netif_carrier_off(dev: netdev);
6011 break;
6012 case CHANGE_MAC_ADDR_RSP:
6013 netdev_dbg(netdev, "Got MAC address change Response\n");
6014 adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter);
6015 break;
6016 case ERROR_INDICATION:
6017 netdev_dbg(netdev, "Got Error Indication\n");
6018 handle_error_indication(crq, adapter);
6019 break;
6020 case REQUEST_STATISTICS_RSP:
6021 netdev_dbg(netdev, "Got Statistics Response\n");
6022 complete(&adapter->stats_done);
6023 break;
6024 case QUERY_IP_OFFLOAD_RSP:
6025 netdev_dbg(netdev, "Got Query IP offload Response\n");
6026 handle_query_ip_offload_rsp(adapter);
6027 break;
6028 case MULTICAST_CTRL_RSP:
6029 netdev_dbg(netdev, "Got multicast control Response\n");
6030 break;
6031 case CONTROL_IP_OFFLOAD_RSP:
6032 netdev_dbg(netdev, "Got Control IP offload Response\n");
6033 dma_unmap_single(dev, adapter->ip_offload_ctrl_tok,
6034 sizeof(adapter->ip_offload_ctrl),
6035 DMA_TO_DEVICE);
6036 complete(&adapter->init_done);
6037 break;
6038 case COLLECT_FW_TRACE_RSP:
6039 netdev_dbg(netdev, "Got Collect firmware trace Response\n");
6040 complete(&adapter->fw_done);
6041 break;
6042 case GET_VPD_SIZE_RSP:
6043 handle_vpd_size_rsp(crq, adapter);
6044 break;
6045 case GET_VPD_RSP:
6046 handle_vpd_rsp(crq, adapter);
6047 break;
6048 case QUERY_PHYS_PARMS_RSP:
6049 adapter->fw_done_rc = handle_query_phys_parms_rsp(crq, adapter);
6050 complete(&adapter->fw_done);
6051 break;
6052 default:
6053 netdev_err(dev: netdev, format: "Got an invalid cmd type 0x%02x\n",
6054 gen_crq->cmd);
6055 }
6056}
6057
6058static irqreturn_t ibmvnic_interrupt(int irq, void *instance)
6059{
6060 struct ibmvnic_adapter *adapter = instance;
6061
6062 tasklet_schedule(t: &adapter->tasklet);
6063 return IRQ_HANDLED;
6064}
6065
6066static void ibmvnic_tasklet(struct tasklet_struct *t)
6067{
6068 struct ibmvnic_adapter *adapter = from_tasklet(adapter, t, tasklet);
6069 struct ibmvnic_crq_queue *queue = &adapter->crq;
6070 union ibmvnic_crq *crq;
6071 unsigned long flags;
6072
6073 spin_lock_irqsave(&queue->lock, flags);
6074
6075 /* Pull all the valid messages off the CRQ */
6076 while ((crq = ibmvnic_next_crq(adapter)) != NULL) {
6077 /* This barrier makes sure ibmvnic_next_crq()'s
6078 * crq->generic.first & IBMVNIC_CRQ_CMD_RSP is loaded
6079 * before ibmvnic_handle_crq()'s
6080 * switch(gen_crq->first) and switch(gen_crq->cmd).
6081 */
6082 dma_rmb();
6083 ibmvnic_handle_crq(crq, adapter);
6084 crq->generic.first = 0;
6085 }
6086
6087 spin_unlock_irqrestore(lock: &queue->lock, flags);
6088}
6089
6090static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter)
6091{
6092 struct vio_dev *vdev = adapter->vdev;
6093 int rc;
6094
6095 do {
6096 rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
6097 } while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc));
6098
6099 if (rc)
6100 dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc);
6101
6102 return rc;
6103}
6104
6105static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter)
6106{
6107 struct ibmvnic_crq_queue *crq = &adapter->crq;
6108 struct device *dev = &adapter->vdev->dev;
6109 struct vio_dev *vdev = adapter->vdev;
6110 int rc;
6111
6112 /* Close the CRQ */
6113 do {
6114 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
6115 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
6116
6117 /* Clean out the queue */
6118 if (!crq->msgs)
6119 return -EINVAL;
6120
6121 memset(crq->msgs, 0, PAGE_SIZE);
6122 crq->cur = 0;
6123 crq->active = false;
6124
6125 /* And re-open it again */
6126 rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
6127 crq->msg_token, PAGE_SIZE);
6128
6129 if (rc == H_CLOSED)
6130 /* Adapter is good, but other end is not ready */
6131 dev_warn(dev, "Partner adapter not ready\n");
6132 else if (rc != 0)
6133 dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc);
6134
6135 return rc;
6136}
6137
6138static void release_crq_queue(struct ibmvnic_adapter *adapter)
6139{
6140 struct ibmvnic_crq_queue *crq = &adapter->crq;
6141 struct vio_dev *vdev = adapter->vdev;
6142 long rc;
6143
6144 if (!crq->msgs)
6145 return;
6146
6147 netdev_dbg(adapter->netdev, "Releasing CRQ\n");
6148 free_irq(vdev->irq, adapter);
6149 tasklet_kill(t: &adapter->tasklet);
6150 do {
6151 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
6152 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
6153
6154 dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE,
6155 DMA_BIDIRECTIONAL);
6156 free_page((unsigned long)crq->msgs);
6157 crq->msgs = NULL;
6158 crq->active = false;
6159}
6160
6161static int init_crq_queue(struct ibmvnic_adapter *adapter)
6162{
6163 struct ibmvnic_crq_queue *crq = &adapter->crq;
6164 struct device *dev = &adapter->vdev->dev;
6165 struct vio_dev *vdev = adapter->vdev;
6166 int rc, retrc = -ENOMEM;
6167
6168 if (crq->msgs)
6169 return 0;
6170
6171 crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL);
6172 /* Should we allocate more than one page? */
6173
6174 if (!crq->msgs)
6175 return -ENOMEM;
6176
6177 crq->size = PAGE_SIZE / sizeof(*crq->msgs);
6178 crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE,
6179 DMA_BIDIRECTIONAL);
6180 if (dma_mapping_error(dev, dma_addr: crq->msg_token))
6181 goto map_failed;
6182
6183 rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
6184 crq->msg_token, PAGE_SIZE);
6185
6186 if (rc == H_RESOURCE)
6187 /* maybe kexecing and resource is busy. try a reset */
6188 rc = ibmvnic_reset_crq(adapter);
6189 retrc = rc;
6190
6191 if (rc == H_CLOSED) {
6192 dev_warn(dev, "Partner adapter not ready\n");
6193 } else if (rc) {
6194 dev_warn(dev, "Error %d opening adapter\n", rc);
6195 goto reg_crq_failed;
6196 }
6197
6198 retrc = 0;
6199
6200 tasklet_setup(t: &adapter->tasklet, callback: (void *)ibmvnic_tasklet);
6201
6202 netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq);
6203 snprintf(buf: crq->name, size: sizeof(crq->name), fmt: "ibmvnic-%x",
6204 adapter->vdev->unit_address);
6205 rc = request_irq(irq: vdev->irq, handler: ibmvnic_interrupt, flags: 0, name: crq->name, dev: adapter);
6206 if (rc) {
6207 dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n",
6208 vdev->irq, rc);
6209 goto req_irq_failed;
6210 }
6211
6212 rc = vio_enable_interrupts(vdev);
6213 if (rc) {
6214 dev_err(dev, "Error %d enabling interrupts\n", rc);
6215 goto req_irq_failed;
6216 }
6217
6218 crq->cur = 0;
6219 spin_lock_init(&crq->lock);
6220
6221 /* process any CRQs that were queued before we enabled interrupts */
6222 tasklet_schedule(t: &adapter->tasklet);
6223
6224 return retrc;
6225
6226req_irq_failed:
6227 tasklet_kill(t: &adapter->tasklet);
6228 do {
6229 rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
6230 } while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
6231reg_crq_failed:
6232 dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL);
6233map_failed:
6234 free_page((unsigned long)crq->msgs);
6235 crq->msgs = NULL;
6236 return retrc;
6237}
6238
6239static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset)
6240{
6241 struct device *dev = &adapter->vdev->dev;
6242 unsigned long timeout = msecs_to_jiffies(m: 20000);
6243 u64 old_num_rx_queues = adapter->req_rx_queues;
6244 u64 old_num_tx_queues = adapter->req_tx_queues;
6245 int rc;
6246
6247 adapter->from_passive_init = false;
6248
6249 rc = ibmvnic_send_crq_init(adapter);
6250 if (rc) {
6251 dev_err(dev, "Send crq init failed with error %d\n", rc);
6252 return rc;
6253 }
6254
6255 if (!wait_for_completion_timeout(x: &adapter->init_done, timeout)) {
6256 dev_err(dev, "Initialization sequence timed out\n");
6257 return -ETIMEDOUT;
6258 }
6259
6260 if (adapter->init_done_rc) {
6261 release_crq_queue(adapter);
6262 dev_err(dev, "CRQ-init failed, %d\n", adapter->init_done_rc);
6263 return adapter->init_done_rc;
6264 }
6265
6266 if (adapter->from_passive_init) {
6267 adapter->state = VNIC_OPEN;
6268 adapter->from_passive_init = false;
6269 dev_err(dev, "CRQ-init failed, passive-init\n");
6270 return -EINVAL;
6271 }
6272
6273 if (reset &&
6274 test_bit(0, &adapter->resetting) && !adapter->wait_for_reset &&
6275 adapter->reset_reason != VNIC_RESET_MOBILITY) {
6276 if (adapter->req_rx_queues != old_num_rx_queues ||
6277 adapter->req_tx_queues != old_num_tx_queues) {
6278 release_sub_crqs(adapter, do_h_free: 0);
6279 rc = init_sub_crqs(adapter);
6280 } else {
6281 /* no need to reinitialize completely, but we do
6282 * need to clean up transmits that were in flight
6283 * when we processed the reset. Failure to do so
6284 * will confound the upper layer, usually TCP, by
6285 * creating the illusion of transmits that are
6286 * awaiting completion.
6287 */
6288 clean_tx_pools(adapter);
6289
6290 rc = reset_sub_crq_queues(adapter);
6291 }
6292 } else {
6293 rc = init_sub_crqs(adapter);
6294 }
6295
6296 if (rc) {
6297 dev_err(dev, "Initialization of sub crqs failed\n");
6298 release_crq_queue(adapter);
6299 return rc;
6300 }
6301
6302 rc = init_sub_crq_irqs(adapter);
6303 if (rc) {
6304 dev_err(dev, "Failed to initialize sub crq irqs\n");
6305 release_crq_queue(adapter);
6306 }
6307
6308 return rc;
6309}
6310
6311static struct device_attribute dev_attr_failover;
6312
6313static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
6314{
6315 struct ibmvnic_adapter *adapter;
6316 struct net_device *netdev;
6317 unsigned char *mac_addr_p;
6318 unsigned long flags;
6319 bool init_success;
6320 int rc;
6321
6322 dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n",
6323 dev->unit_address);
6324
6325 mac_addr_p = (unsigned char *)vio_get_attribute(dev,
6326 VETH_MAC_ADDR, NULL);
6327 if (!mac_addr_p) {
6328 dev_err(&dev->dev,
6329 "(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n",
6330 __FILE__, __LINE__);
6331 return 0;
6332 }
6333
6334 netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter),
6335 IBMVNIC_MAX_QUEUES);
6336 if (!netdev)
6337 return -ENOMEM;
6338
6339 adapter = netdev_priv(dev: netdev);
6340 adapter->state = VNIC_PROBING;
6341 dev_set_drvdata(dev: &dev->dev, data: netdev);
6342 adapter->vdev = dev;
6343 adapter->netdev = netdev;
6344 adapter->login_pending = false;
6345 memset(&adapter->map_ids, 0, sizeof(adapter->map_ids));
6346 /* map_ids start at 1, so ensure map_id 0 is always "in-use" */
6347 bitmap_set(map: adapter->map_ids, start: 0, nbits: 1);
6348
6349 ether_addr_copy(dst: adapter->mac_addr, src: mac_addr_p);
6350 eth_hw_addr_set(dev: netdev, addr: adapter->mac_addr);
6351 netdev->irq = dev->irq;
6352 netdev->netdev_ops = &ibmvnic_netdev_ops;
6353 netdev->ethtool_ops = &ibmvnic_ethtool_ops;
6354 SET_NETDEV_DEV(netdev, &dev->dev);
6355
6356 INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset);
6357 INIT_DELAYED_WORK(&adapter->ibmvnic_delayed_reset,
6358 __ibmvnic_delayed_reset);
6359 INIT_LIST_HEAD(list: &adapter->rwi_list);
6360 spin_lock_init(&adapter->rwi_lock);
6361 spin_lock_init(&adapter->state_lock);
6362 mutex_init(&adapter->fw_lock);
6363 init_completion(x: &adapter->probe_done);
6364 init_completion(x: &adapter->init_done);
6365 init_completion(x: &adapter->fw_done);
6366 init_completion(x: &adapter->reset_done);
6367 init_completion(x: &adapter->stats_done);
6368 clear_bit(nr: 0, addr: &adapter->resetting);
6369 adapter->prev_rx_buf_sz = 0;
6370 adapter->prev_mtu = 0;
6371
6372 init_success = false;
6373 do {
6374 reinit_init_done(adapter);
6375
6376 /* clear any failovers we got in the previous pass
6377 * since we are reinitializing the CRQ
6378 */
6379 adapter->failover_pending = false;
6380
6381 /* If we had already initialized CRQ, we may have one or
6382 * more resets queued already. Discard those and release
6383 * the CRQ before initializing the CRQ again.
6384 */
6385 release_crq_queue(adapter);
6386
6387 /* Since we are still in PROBING state, __ibmvnic_reset()
6388 * will not access the ->rwi_list and since we released CRQ,
6389 * we won't get _new_ transport events. But there maybe an
6390 * ongoing ibmvnic_reset() call. So serialize access to
6391 * rwi_list. If we win the race, ibvmnic_reset() could add
6392 * a reset after we purged but thats ok - we just may end
6393 * up with an extra reset (i.e similar to having two or more
6394 * resets in the queue at once).
6395 * CHECK.
6396 */
6397 spin_lock_irqsave(&adapter->rwi_lock, flags);
6398 flush_reset_queue(adapter);
6399 spin_unlock_irqrestore(lock: &adapter->rwi_lock, flags);
6400
6401 rc = init_crq_queue(adapter);
6402 if (rc) {
6403 dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n",
6404 rc);
6405 goto ibmvnic_init_fail;
6406 }
6407
6408 rc = ibmvnic_reset_init(adapter, reset: false);
6409 } while (rc == -EAGAIN);
6410
6411 /* We are ignoring the error from ibmvnic_reset_init() assuming that the
6412 * partner is not ready. CRQ is not active. When the partner becomes
6413 * ready, we will do the passive init reset.
6414 */
6415
6416 if (!rc)
6417 init_success = true;
6418
6419 rc = init_stats_buffers(adapter);
6420 if (rc)
6421 goto ibmvnic_init_fail;
6422
6423 rc = init_stats_token(adapter);
6424 if (rc)
6425 goto ibmvnic_stats_fail;
6426
6427 rc = device_create_file(device: &dev->dev, entry: &dev_attr_failover);
6428 if (rc)
6429 goto ibmvnic_dev_file_err;
6430
6431 netif_carrier_off(dev: netdev);
6432
6433 if (init_success) {
6434 adapter->state = VNIC_PROBED;
6435 netdev->mtu = adapter->req_mtu - ETH_HLEN;
6436 netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
6437 netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
6438 } else {
6439 adapter->state = VNIC_DOWN;
6440 }
6441
6442 adapter->wait_for_reset = false;
6443 adapter->last_reset_time = jiffies;
6444
6445 rc = register_netdev(dev: netdev);
6446 if (rc) {
6447 dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc);
6448 goto ibmvnic_register_fail;
6449 }
6450 dev_info(&dev->dev, "ibmvnic registered\n");
6451
6452 rc = ibmvnic_cpu_notif_add(adapter);
6453 if (rc) {
6454 netdev_err(dev: netdev, format: "Registering cpu notifier failed\n");
6455 goto cpu_notif_add_failed;
6456 }
6457
6458 complete(&adapter->probe_done);
6459
6460 return 0;
6461
6462cpu_notif_add_failed:
6463 unregister_netdev(dev: netdev);
6464
6465ibmvnic_register_fail:
6466 device_remove_file(dev: &dev->dev, attr: &dev_attr_failover);
6467
6468ibmvnic_dev_file_err:
6469 release_stats_token(adapter);
6470
6471ibmvnic_stats_fail:
6472 release_stats_buffers(adapter);
6473
6474ibmvnic_init_fail:
6475 release_sub_crqs(adapter, do_h_free: 1);
6476 release_crq_queue(adapter);
6477
6478 /* cleanup worker thread after releasing CRQ so we don't get
6479 * transport events (i.e new work items for the worker thread).
6480 */
6481 adapter->state = VNIC_REMOVING;
6482 complete(&adapter->probe_done);
6483 flush_work(work: &adapter->ibmvnic_reset);
6484 flush_delayed_work(dwork: &adapter->ibmvnic_delayed_reset);
6485
6486 flush_reset_queue(adapter);
6487
6488 mutex_destroy(lock: &adapter->fw_lock);
6489 free_netdev(dev: netdev);
6490
6491 return rc;
6492}
6493
6494static void ibmvnic_remove(struct vio_dev *dev)
6495{
6496 struct net_device *netdev = dev_get_drvdata(dev: &dev->dev);
6497 struct ibmvnic_adapter *adapter = netdev_priv(dev: netdev);
6498 unsigned long flags;
6499
6500 spin_lock_irqsave(&adapter->state_lock, flags);
6501
6502 /* If ibmvnic_reset() is scheduling a reset, wait for it to
6503 * finish. Then, set the state to REMOVING to prevent it from
6504 * scheduling any more work and to have reset functions ignore
6505 * any resets that have already been scheduled. Drop the lock
6506 * after setting state, so __ibmvnic_reset() which is called
6507 * from the flush_work() below, can make progress.
6508 */
6509 spin_lock(lock: &adapter->rwi_lock);
6510 adapter->state = VNIC_REMOVING;
6511 spin_unlock(lock: &adapter->rwi_lock);
6512
6513 spin_unlock_irqrestore(lock: &adapter->state_lock, flags);
6514
6515 ibmvnic_cpu_notif_remove(adapter);
6516
6517 flush_work(work: &adapter->ibmvnic_reset);
6518 flush_delayed_work(dwork: &adapter->ibmvnic_delayed_reset);
6519
6520 rtnl_lock();
6521 unregister_netdevice(dev: netdev);
6522
6523 release_resources(adapter);
6524 release_rx_pools(adapter);
6525 release_tx_pools(adapter);
6526 release_sub_crqs(adapter, do_h_free: 1);
6527 release_crq_queue(adapter);
6528
6529 release_stats_token(adapter);
6530 release_stats_buffers(adapter);
6531
6532 adapter->state = VNIC_REMOVED;
6533
6534 rtnl_unlock();
6535 mutex_destroy(lock: &adapter->fw_lock);
6536 device_remove_file(dev: &dev->dev, attr: &dev_attr_failover);
6537 free_netdev(dev: netdev);
6538 dev_set_drvdata(dev: &dev->dev, NULL);
6539}
6540
6541static ssize_t failover_store(struct device *dev, struct device_attribute *attr,
6542 const char *buf, size_t count)
6543{
6544 struct net_device *netdev = dev_get_drvdata(dev);
6545 struct ibmvnic_adapter *adapter = netdev_priv(dev: netdev);
6546 unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
6547 __be64 session_token;
6548 long rc;
6549
6550 if (!sysfs_streq(s1: buf, s2: "1"))
6551 return -EINVAL;
6552
6553 rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address,
6554 H_GET_SESSION_TOKEN, 0, 0, 0);
6555 if (rc) {
6556 netdev_err(dev: netdev, format: "Couldn't retrieve session token, rc %ld\n",
6557 rc);
6558 goto last_resort;
6559 }
6560
6561 session_token = (__be64)retbuf[0];
6562 netdev_dbg(netdev, "Initiating client failover, session id %llx\n",
6563 be64_to_cpu(session_token));
6564 rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
6565 H_SESSION_ERR_DETECTED, session_token, 0, 0);
6566 if (rc) {
6567 netdev_err(dev: netdev,
6568 format: "H_VIOCTL initiated failover failed, rc %ld\n",
6569 rc);
6570 goto last_resort;
6571 }
6572
6573 return count;
6574
6575last_resort:
6576 netdev_dbg(netdev, "Trying to send CRQ_CMD, the last resort\n");
6577 ibmvnic_reset(adapter, reason: VNIC_RESET_FAILOVER);
6578
6579 return count;
6580}
6581static DEVICE_ATTR_WO(failover);
6582
6583static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev)
6584{
6585 struct net_device *netdev = dev_get_drvdata(dev: &vdev->dev);
6586 struct ibmvnic_adapter *adapter;
6587 struct iommu_table *tbl;
6588 unsigned long ret = 0;
6589 int i;
6590
6591 tbl = get_iommu_table_base(&vdev->dev);
6592
6593 /* netdev inits at probe time along with the structures we need below*/
6594 if (!netdev)
6595 return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl);
6596
6597 adapter = netdev_priv(dev: netdev);
6598
6599 ret += PAGE_SIZE; /* the crq message queue */
6600 ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl);
6601
6602 for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++)
6603 ret += 4 * PAGE_SIZE; /* the scrq message queue */
6604
6605 for (i = 0; i < adapter->num_active_rx_pools; i++)
6606 ret += adapter->rx_pool[i].size *
6607 IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl);
6608
6609 return ret;
6610}
6611
6612static int ibmvnic_resume(struct device *dev)
6613{
6614 struct net_device *netdev = dev_get_drvdata(dev);
6615 struct ibmvnic_adapter *adapter = netdev_priv(dev: netdev);
6616
6617 if (adapter->state != VNIC_OPEN)
6618 return 0;
6619
6620 tasklet_schedule(t: &adapter->tasklet);
6621
6622 return 0;
6623}
6624
6625static const struct vio_device_id ibmvnic_device_table[] = {
6626 {"network", "IBM,vnic"},
6627 {"", "" }
6628};
6629MODULE_DEVICE_TABLE(vio, ibmvnic_device_table);
6630
6631static const struct dev_pm_ops ibmvnic_pm_ops = {
6632 .resume = ibmvnic_resume
6633};
6634
6635static struct vio_driver ibmvnic_driver = {
6636 .id_table = ibmvnic_device_table,
6637 .probe = ibmvnic_probe,
6638 .remove = ibmvnic_remove,
6639 .get_desired_dma = ibmvnic_get_desired_dma,
6640 .name = ibmvnic_driver_name,
6641 .pm = &ibmvnic_pm_ops,
6642};
6643
6644/* module functions */
6645static int __init ibmvnic_module_init(void)
6646{
6647 int ret;
6648
6649 ret = cpuhp_setup_state_multi(state: CPUHP_AP_ONLINE_DYN, name: "net/ibmvnic:online",
6650 startup: ibmvnic_cpu_online,
6651 teardown: ibmvnic_cpu_down_prep);
6652 if (ret < 0)
6653 goto out;
6654 ibmvnic_online = ret;
6655 ret = cpuhp_setup_state_multi(state: CPUHP_IBMVNIC_DEAD, name: "net/ibmvnic:dead",
6656 NULL, teardown: ibmvnic_cpu_dead);
6657 if (ret)
6658 goto err_dead;
6659
6660 ret = vio_register_driver(&ibmvnic_driver);
6661 if (ret)
6662 goto err_vio_register;
6663
6664 pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string,
6665 IBMVNIC_DRIVER_VERSION);
6666
6667 return 0;
6668err_vio_register:
6669 cpuhp_remove_multi_state(state: CPUHP_IBMVNIC_DEAD);
6670err_dead:
6671 cpuhp_remove_multi_state(state: ibmvnic_online);
6672out:
6673 return ret;
6674}
6675
6676static void __exit ibmvnic_module_exit(void)
6677{
6678 vio_unregister_driver(&ibmvnic_driver);
6679 cpuhp_remove_multi_state(state: CPUHP_IBMVNIC_DEAD);
6680 cpuhp_remove_multi_state(state: ibmvnic_online);
6681}
6682
6683module_init(ibmvnic_module_init);
6684module_exit(ibmvnic_module_exit);
6685

source code of linux/drivers/net/ethernet/ibm/ibmvnic.c