xref: /openbmc/linux/drivers/net/ethernet/ibm/ibmvnic.c (revision e8614313)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /**************************************************************************/
3 /*                                                                        */
4 /*  IBM System i and System p Virtual NIC Device Driver                   */
5 /*  Copyright (C) 2014 IBM Corp.                                          */
6 /*  Santiago Leon (santi_leon@yahoo.com)                                  */
7 /*  Thomas Falcon (tlfalcon@linux.vnet.ibm.com)                           */
8 /*  John Allen (jallen@linux.vnet.ibm.com)                                */
9 /*                                                                        */
10 /*                                                                        */
11 /* This module contains the implementation of a virtual ethernet device   */
12 /* for use with IBM i/p Series LPAR Linux. It utilizes the logical LAN    */
13 /* option of the RS/6000 Platform Architecture to interface with virtual  */
14 /* ethernet NICs that are presented to the partition by the hypervisor.   */
15 /*									   */
16 /* Messages are passed between the VNIC driver and the VNIC server using  */
17 /* Command/Response Queues (CRQs) and sub CRQs (sCRQs). CRQs are used to  */
18 /* issue and receive commands that initiate communication with the server */
19 /* on driver initialization. Sub CRQs (sCRQs) are similar to CRQs, but    */
20 /* are used by the driver to notify the server that a packet is           */
21 /* ready for transmission or that a buffer has been added to receive a    */
22 /* packet. Subsequently, sCRQs are used by the server to notify the       */
23 /* driver that a packet transmission has been completed or that a packet  */
24 /* has been received and placed in a waiting buffer.                      */
25 /*                                                                        */
26 /* In lieu of a more conventional "on-the-fly" DMA mapping strategy in    */
27 /* which skbs are DMA mapped and immediately unmapped when the transmit   */
28 /* or receive has been completed, the VNIC driver is required to use      */
29 /* "long term mapping". This entails that large, continuous DMA mapped    */
30 /* buffers are allocated on driver initialization and these buffers are   */
31 /* then continuously reused to pass skbs to and from the VNIC server.     */
32 /*                                                                        */
33 /**************************************************************************/
34 
35 #include <linux/module.h>
36 #include <linux/moduleparam.h>
37 #include <linux/types.h>
38 #include <linux/errno.h>
39 #include <linux/completion.h>
40 #include <linux/ioport.h>
41 #include <linux/dma-mapping.h>
42 #include <linux/kernel.h>
43 #include <linux/netdevice.h>
44 #include <linux/etherdevice.h>
45 #include <linux/skbuff.h>
46 #include <linux/init.h>
47 #include <linux/delay.h>
48 #include <linux/mm.h>
49 #include <linux/ethtool.h>
50 #include <linux/proc_fs.h>
51 #include <linux/if_arp.h>
52 #include <linux/in.h>
53 #include <linux/ip.h>
54 #include <linux/ipv6.h>
55 #include <linux/irq.h>
56 #include <linux/kthread.h>
57 #include <linux/seq_file.h>
58 #include <linux/interrupt.h>
59 #include <net/net_namespace.h>
60 #include <asm/hvcall.h>
61 #include <linux/atomic.h>
62 #include <asm/vio.h>
63 #include <asm/iommu.h>
64 #include <linux/uaccess.h>
65 #include <asm/firmware.h>
66 #include <linux/workqueue.h>
67 #include <linux/if_vlan.h>
68 #include <linux/utsname.h>
69 
70 #include "ibmvnic.h"
71 
72 static const char ibmvnic_driver_name[] = "ibmvnic";
73 static const char ibmvnic_driver_string[] = "IBM System i/p Virtual NIC Driver";
74 
75 MODULE_AUTHOR("Santiago Leon");
76 MODULE_DESCRIPTION("IBM System i/p Virtual NIC Driver");
77 MODULE_LICENSE("GPL");
78 MODULE_VERSION(IBMVNIC_DRIVER_VERSION);
79 
80 static int ibmvnic_version = IBMVNIC_INITIAL_VERSION;
81 static void release_sub_crqs(struct ibmvnic_adapter *, bool);
82 static int ibmvnic_reset_crq(struct ibmvnic_adapter *);
83 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *);
84 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *);
85 static int ibmvnic_send_crq(struct ibmvnic_adapter *, union ibmvnic_crq *);
86 static int send_subcrq_indirect(struct ibmvnic_adapter *, u64, u64, u64);
87 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance);
88 static int enable_scrq_irq(struct ibmvnic_adapter *,
89 			   struct ibmvnic_sub_crq_queue *);
90 static int disable_scrq_irq(struct ibmvnic_adapter *,
91 			    struct ibmvnic_sub_crq_queue *);
92 static int pending_scrq(struct ibmvnic_adapter *,
93 			struct ibmvnic_sub_crq_queue *);
94 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *,
95 					struct ibmvnic_sub_crq_queue *);
96 static int ibmvnic_poll(struct napi_struct *napi, int data);
97 static void send_query_map(struct ibmvnic_adapter *adapter);
98 static int send_request_map(struct ibmvnic_adapter *, dma_addr_t, __be32, u8);
99 static int send_request_unmap(struct ibmvnic_adapter *, u8);
100 static int send_login(struct ibmvnic_adapter *adapter);
101 static void send_query_cap(struct ibmvnic_adapter *adapter);
102 static int init_sub_crqs(struct ibmvnic_adapter *);
103 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter);
104 static int ibmvnic_reset_init(struct ibmvnic_adapter *, bool reset);
105 static void release_crq_queue(struct ibmvnic_adapter *);
106 static int __ibmvnic_set_mac(struct net_device *, u8 *);
107 static int init_crq_queue(struct ibmvnic_adapter *adapter);
108 static int send_query_phys_parms(struct ibmvnic_adapter *adapter);
109 
110 struct ibmvnic_stat {
111 	char name[ETH_GSTRING_LEN];
112 	int offset;
113 };
114 
115 #define IBMVNIC_STAT_OFF(stat) (offsetof(struct ibmvnic_adapter, stats) + \
116 			     offsetof(struct ibmvnic_statistics, stat))
117 #define IBMVNIC_GET_STAT(a, off) (*((u64 *)(((unsigned long)(a)) + (off))))
118 
119 static const struct ibmvnic_stat ibmvnic_stats[] = {
120 	{"rx_packets", IBMVNIC_STAT_OFF(rx_packets)},
121 	{"rx_bytes", IBMVNIC_STAT_OFF(rx_bytes)},
122 	{"tx_packets", IBMVNIC_STAT_OFF(tx_packets)},
123 	{"tx_bytes", IBMVNIC_STAT_OFF(tx_bytes)},
124 	{"ucast_tx_packets", IBMVNIC_STAT_OFF(ucast_tx_packets)},
125 	{"ucast_rx_packets", IBMVNIC_STAT_OFF(ucast_rx_packets)},
126 	{"mcast_tx_packets", IBMVNIC_STAT_OFF(mcast_tx_packets)},
127 	{"mcast_rx_packets", IBMVNIC_STAT_OFF(mcast_rx_packets)},
128 	{"bcast_tx_packets", IBMVNIC_STAT_OFF(bcast_tx_packets)},
129 	{"bcast_rx_packets", IBMVNIC_STAT_OFF(bcast_rx_packets)},
130 	{"align_errors", IBMVNIC_STAT_OFF(align_errors)},
131 	{"fcs_errors", IBMVNIC_STAT_OFF(fcs_errors)},
132 	{"single_collision_frames", IBMVNIC_STAT_OFF(single_collision_frames)},
133 	{"multi_collision_frames", IBMVNIC_STAT_OFF(multi_collision_frames)},
134 	{"sqe_test_errors", IBMVNIC_STAT_OFF(sqe_test_errors)},
135 	{"deferred_tx", IBMVNIC_STAT_OFF(deferred_tx)},
136 	{"late_collisions", IBMVNIC_STAT_OFF(late_collisions)},
137 	{"excess_collisions", IBMVNIC_STAT_OFF(excess_collisions)},
138 	{"internal_mac_tx_errors", IBMVNIC_STAT_OFF(internal_mac_tx_errors)},
139 	{"carrier_sense", IBMVNIC_STAT_OFF(carrier_sense)},
140 	{"too_long_frames", IBMVNIC_STAT_OFF(too_long_frames)},
141 	{"internal_mac_rx_errors", IBMVNIC_STAT_OFF(internal_mac_rx_errors)},
142 };
143 
144 static long h_reg_sub_crq(unsigned long unit_address, unsigned long token,
145 			  unsigned long length, unsigned long *number,
146 			  unsigned long *irq)
147 {
148 	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
149 	long rc;
150 
151 	rc = plpar_hcall(H_REG_SUB_CRQ, retbuf, unit_address, token, length);
152 	*number = retbuf[0];
153 	*irq = retbuf[1];
154 
155 	return rc;
156 }
157 
158 /**
159  * ibmvnic_wait_for_completion - Check device state and wait for completion
160  * @adapter: private device data
161  * @comp_done: completion structure to wait for
162  * @timeout: time to wait in milliseconds
163  *
164  * Wait for a completion signal or until the timeout limit is reached
165  * while checking that the device is still active.
166  */
167 static int ibmvnic_wait_for_completion(struct ibmvnic_adapter *adapter,
168 				       struct completion *comp_done,
169 				       unsigned long timeout)
170 {
171 	struct net_device *netdev;
172 	unsigned long div_timeout;
173 	u8 retry;
174 
175 	netdev = adapter->netdev;
176 	retry = 5;
177 	div_timeout = msecs_to_jiffies(timeout / retry);
178 	while (true) {
179 		if (!adapter->crq.active) {
180 			netdev_err(netdev, "Device down!\n");
181 			return -ENODEV;
182 		}
183 		if (!retry--)
184 			break;
185 		if (wait_for_completion_timeout(comp_done, div_timeout))
186 			return 0;
187 	}
188 	netdev_err(netdev, "Operation timed out.\n");
189 	return -ETIMEDOUT;
190 }
191 
192 static int alloc_long_term_buff(struct ibmvnic_adapter *adapter,
193 				struct ibmvnic_long_term_buff *ltb, int size)
194 {
195 	struct device *dev = &adapter->vdev->dev;
196 	int rc;
197 
198 	ltb->size = size;
199 	ltb->buff = dma_alloc_coherent(dev, ltb->size, &ltb->addr,
200 				       GFP_KERNEL);
201 
202 	if (!ltb->buff) {
203 		dev_err(dev, "Couldn't alloc long term buffer\n");
204 		return -ENOMEM;
205 	}
206 	ltb->map_id = adapter->map_id;
207 	adapter->map_id++;
208 
209 	mutex_lock(&adapter->fw_lock);
210 	adapter->fw_done_rc = 0;
211 	reinit_completion(&adapter->fw_done);
212 	rc = send_request_map(adapter, ltb->addr,
213 			      ltb->size, ltb->map_id);
214 	if (rc) {
215 		dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
216 		mutex_unlock(&adapter->fw_lock);
217 		return rc;
218 	}
219 
220 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
221 	if (rc) {
222 		dev_err(dev,
223 			"Long term map request aborted or timed out,rc = %d\n",
224 			rc);
225 		dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
226 		mutex_unlock(&adapter->fw_lock);
227 		return rc;
228 	}
229 
230 	if (adapter->fw_done_rc) {
231 		dev_err(dev, "Couldn't map long term buffer,rc = %d\n",
232 			adapter->fw_done_rc);
233 		dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
234 		mutex_unlock(&adapter->fw_lock);
235 		return -1;
236 	}
237 	mutex_unlock(&adapter->fw_lock);
238 	return 0;
239 }
240 
241 static void free_long_term_buff(struct ibmvnic_adapter *adapter,
242 				struct ibmvnic_long_term_buff *ltb)
243 {
244 	struct device *dev = &adapter->vdev->dev;
245 
246 	if (!ltb->buff)
247 		return;
248 
249 	/* VIOS automatically unmaps the long term buffer at remote
250 	 * end for the following resets:
251 	 * FAILOVER, MOBILITY, TIMEOUT.
252 	 */
253 	if (adapter->reset_reason != VNIC_RESET_FAILOVER &&
254 	    adapter->reset_reason != VNIC_RESET_MOBILITY &&
255 	    adapter->reset_reason != VNIC_RESET_TIMEOUT)
256 		send_request_unmap(adapter, ltb->map_id);
257 	dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
258 }
259 
260 static int reset_long_term_buff(struct ibmvnic_long_term_buff *ltb)
261 {
262 	if (!ltb->buff)
263 		return -EINVAL;
264 
265 	memset(ltb->buff, 0, ltb->size);
266 	return 0;
267 }
268 
269 static void deactivate_rx_pools(struct ibmvnic_adapter *adapter)
270 {
271 	int i;
272 
273 	for (i = 0; i < adapter->num_active_rx_pools; i++)
274 		adapter->rx_pool[i].active = 0;
275 }
276 
277 static void replenish_rx_pool(struct ibmvnic_adapter *adapter,
278 			      struct ibmvnic_rx_pool *pool)
279 {
280 	int count = pool->size - atomic_read(&pool->available);
281 	u64 handle = adapter->rx_scrq[pool->index]->handle;
282 	struct device *dev = &adapter->vdev->dev;
283 	struct ibmvnic_ind_xmit_queue *ind_bufp;
284 	struct ibmvnic_sub_crq_queue *rx_scrq;
285 	union sub_crq *sub_crq;
286 	int buffers_added = 0;
287 	unsigned long lpar_rc;
288 	struct sk_buff *skb;
289 	unsigned int offset;
290 	dma_addr_t dma_addr;
291 	unsigned char *dst;
292 	int shift = 0;
293 	int index;
294 	int i;
295 
296 	if (!pool->active)
297 		return;
298 
299 	rx_scrq = adapter->rx_scrq[pool->index];
300 	ind_bufp = &rx_scrq->ind_buf;
301 	for (i = 0; i < count; ++i) {
302 		skb = netdev_alloc_skb(adapter->netdev, pool->buff_size);
303 		if (!skb) {
304 			dev_err(dev, "Couldn't replenish rx buff\n");
305 			adapter->replenish_no_mem++;
306 			break;
307 		}
308 
309 		index = pool->free_map[pool->next_free];
310 
311 		if (pool->rx_buff[index].skb)
312 			dev_err(dev, "Inconsistent free_map!\n");
313 
314 		/* Copy the skb to the long term mapped DMA buffer */
315 		offset = index * pool->buff_size;
316 		dst = pool->long_term_buff.buff + offset;
317 		memset(dst, 0, pool->buff_size);
318 		dma_addr = pool->long_term_buff.addr + offset;
319 		pool->rx_buff[index].data = dst;
320 
321 		pool->free_map[pool->next_free] = IBMVNIC_INVALID_MAP;
322 		pool->rx_buff[index].dma = dma_addr;
323 		pool->rx_buff[index].skb = skb;
324 		pool->rx_buff[index].pool_index = pool->index;
325 		pool->rx_buff[index].size = pool->buff_size;
326 
327 		sub_crq = &ind_bufp->indir_arr[ind_bufp->index++];
328 		memset(sub_crq, 0, sizeof(*sub_crq));
329 		sub_crq->rx_add.first = IBMVNIC_CRQ_CMD;
330 		sub_crq->rx_add.correlator =
331 		    cpu_to_be64((u64)&pool->rx_buff[index]);
332 		sub_crq->rx_add.ioba = cpu_to_be32(dma_addr);
333 		sub_crq->rx_add.map_id = pool->long_term_buff.map_id;
334 
335 		/* The length field of the sCRQ is defined to be 24 bits so the
336 		 * buffer size needs to be left shifted by a byte before it is
337 		 * converted to big endian to prevent the last byte from being
338 		 * truncated.
339 		 */
340 #ifdef __LITTLE_ENDIAN__
341 		shift = 8;
342 #endif
343 		sub_crq->rx_add.len = cpu_to_be32(pool->buff_size << shift);
344 		pool->next_free = (pool->next_free + 1) % pool->size;
345 		if (ind_bufp->index == IBMVNIC_MAX_IND_DESCS ||
346 		    i == count - 1) {
347 			lpar_rc =
348 				send_subcrq_indirect(adapter, handle,
349 						     (u64)ind_bufp->indir_dma,
350 						     (u64)ind_bufp->index);
351 			if (lpar_rc != H_SUCCESS)
352 				goto failure;
353 			buffers_added += ind_bufp->index;
354 			adapter->replenish_add_buff_success += ind_bufp->index;
355 			ind_bufp->index = 0;
356 		}
357 	}
358 	atomic_add(buffers_added, &pool->available);
359 	return;
360 
361 failure:
362 	if (lpar_rc != H_PARAMETER && lpar_rc != H_CLOSED)
363 		dev_err_ratelimited(dev, "rx: replenish packet buffer failed\n");
364 	for (i = ind_bufp->index - 1; i >= 0; --i) {
365 		struct ibmvnic_rx_buff *rx_buff;
366 
367 		pool->next_free = pool->next_free == 0 ?
368 				  pool->size - 1 : pool->next_free - 1;
369 		sub_crq = &ind_bufp->indir_arr[i];
370 		rx_buff = (struct ibmvnic_rx_buff *)
371 				be64_to_cpu(sub_crq->rx_add.correlator);
372 		index = (int)(rx_buff - pool->rx_buff);
373 		pool->free_map[pool->next_free] = index;
374 		dev_kfree_skb_any(pool->rx_buff[index].skb);
375 		pool->rx_buff[index].skb = NULL;
376 	}
377 	adapter->replenish_add_buff_failure += ind_bufp->index;
378 	atomic_add(buffers_added, &pool->available);
379 	ind_bufp->index = 0;
380 	if (lpar_rc == H_CLOSED || adapter->failover_pending) {
381 		/* Disable buffer pool replenishment and report carrier off if
382 		 * queue is closed or pending failover.
383 		 * Firmware guarantees that a signal will be sent to the
384 		 * driver, triggering a reset.
385 		 */
386 		deactivate_rx_pools(adapter);
387 		netif_carrier_off(adapter->netdev);
388 	}
389 }
390 
391 static void replenish_pools(struct ibmvnic_adapter *adapter)
392 {
393 	int i;
394 
395 	adapter->replenish_task_cycles++;
396 	for (i = 0; i < adapter->num_active_rx_pools; i++) {
397 		if (adapter->rx_pool[i].active)
398 			replenish_rx_pool(adapter, &adapter->rx_pool[i]);
399 	}
400 
401 	netdev_dbg(adapter->netdev, "Replenished %d pools\n", i);
402 }
403 
404 static void release_stats_buffers(struct ibmvnic_adapter *adapter)
405 {
406 	kfree(adapter->tx_stats_buffers);
407 	kfree(adapter->rx_stats_buffers);
408 	adapter->tx_stats_buffers = NULL;
409 	adapter->rx_stats_buffers = NULL;
410 }
411 
412 static int init_stats_buffers(struct ibmvnic_adapter *adapter)
413 {
414 	adapter->tx_stats_buffers =
415 				kcalloc(IBMVNIC_MAX_QUEUES,
416 					sizeof(struct ibmvnic_tx_queue_stats),
417 					GFP_KERNEL);
418 	if (!adapter->tx_stats_buffers)
419 		return -ENOMEM;
420 
421 	adapter->rx_stats_buffers =
422 				kcalloc(IBMVNIC_MAX_QUEUES,
423 					sizeof(struct ibmvnic_rx_queue_stats),
424 					GFP_KERNEL);
425 	if (!adapter->rx_stats_buffers)
426 		return -ENOMEM;
427 
428 	return 0;
429 }
430 
431 static void release_stats_token(struct ibmvnic_adapter *adapter)
432 {
433 	struct device *dev = &adapter->vdev->dev;
434 
435 	if (!adapter->stats_token)
436 		return;
437 
438 	dma_unmap_single(dev, adapter->stats_token,
439 			 sizeof(struct ibmvnic_statistics),
440 			 DMA_FROM_DEVICE);
441 	adapter->stats_token = 0;
442 }
443 
444 static int init_stats_token(struct ibmvnic_adapter *adapter)
445 {
446 	struct device *dev = &adapter->vdev->dev;
447 	dma_addr_t stok;
448 
449 	stok = dma_map_single(dev, &adapter->stats,
450 			      sizeof(struct ibmvnic_statistics),
451 			      DMA_FROM_DEVICE);
452 	if (dma_mapping_error(dev, stok)) {
453 		dev_err(dev, "Couldn't map stats buffer\n");
454 		return -1;
455 	}
456 
457 	adapter->stats_token = stok;
458 	netdev_dbg(adapter->netdev, "Stats token initialized (%llx)\n", stok);
459 	return 0;
460 }
461 
462 static int reset_rx_pools(struct ibmvnic_adapter *adapter)
463 {
464 	struct ibmvnic_rx_pool *rx_pool;
465 	u64 buff_size;
466 	int rx_scrqs;
467 	int i, j, rc;
468 
469 	if (!adapter->rx_pool)
470 		return -1;
471 
472 	buff_size = adapter->cur_rx_buf_sz;
473 	rx_scrqs = adapter->num_active_rx_pools;
474 	for (i = 0; i < rx_scrqs; i++) {
475 		rx_pool = &adapter->rx_pool[i];
476 
477 		netdev_dbg(adapter->netdev, "Re-setting rx_pool[%d]\n", i);
478 
479 		if (rx_pool->buff_size != buff_size) {
480 			free_long_term_buff(adapter, &rx_pool->long_term_buff);
481 			rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
482 			rc = alloc_long_term_buff(adapter,
483 						  &rx_pool->long_term_buff,
484 						  rx_pool->size *
485 						  rx_pool->buff_size);
486 		} else {
487 			rc = reset_long_term_buff(&rx_pool->long_term_buff);
488 		}
489 
490 		if (rc)
491 			return rc;
492 
493 		for (j = 0; j < rx_pool->size; j++)
494 			rx_pool->free_map[j] = j;
495 
496 		memset(rx_pool->rx_buff, 0,
497 		       rx_pool->size * sizeof(struct ibmvnic_rx_buff));
498 
499 		atomic_set(&rx_pool->available, 0);
500 		rx_pool->next_alloc = 0;
501 		rx_pool->next_free = 0;
502 		rx_pool->active = 1;
503 	}
504 
505 	return 0;
506 }
507 
508 static void release_rx_pools(struct ibmvnic_adapter *adapter)
509 {
510 	struct ibmvnic_rx_pool *rx_pool;
511 	int i, j;
512 
513 	if (!adapter->rx_pool)
514 		return;
515 
516 	for (i = 0; i < adapter->num_active_rx_pools; i++) {
517 		rx_pool = &adapter->rx_pool[i];
518 
519 		netdev_dbg(adapter->netdev, "Releasing rx_pool[%d]\n", i);
520 
521 		kfree(rx_pool->free_map);
522 		free_long_term_buff(adapter, &rx_pool->long_term_buff);
523 
524 		if (!rx_pool->rx_buff)
525 			continue;
526 
527 		for (j = 0; j < rx_pool->size; j++) {
528 			if (rx_pool->rx_buff[j].skb) {
529 				dev_kfree_skb_any(rx_pool->rx_buff[j].skb);
530 				rx_pool->rx_buff[j].skb = NULL;
531 			}
532 		}
533 
534 		kfree(rx_pool->rx_buff);
535 	}
536 
537 	kfree(adapter->rx_pool);
538 	adapter->rx_pool = NULL;
539 	adapter->num_active_rx_pools = 0;
540 }
541 
542 static int init_rx_pools(struct net_device *netdev)
543 {
544 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
545 	struct device *dev = &adapter->vdev->dev;
546 	struct ibmvnic_rx_pool *rx_pool;
547 	int rxadd_subcrqs;
548 	u64 buff_size;
549 	int i, j;
550 
551 	rxadd_subcrqs = adapter->num_active_rx_scrqs;
552 	buff_size = adapter->cur_rx_buf_sz;
553 
554 	adapter->rx_pool = kcalloc(rxadd_subcrqs,
555 				   sizeof(struct ibmvnic_rx_pool),
556 				   GFP_KERNEL);
557 	if (!adapter->rx_pool) {
558 		dev_err(dev, "Failed to allocate rx pools\n");
559 		return -1;
560 	}
561 
562 	adapter->num_active_rx_pools = rxadd_subcrqs;
563 
564 	for (i = 0; i < rxadd_subcrqs; i++) {
565 		rx_pool = &adapter->rx_pool[i];
566 
567 		netdev_dbg(adapter->netdev,
568 			   "Initializing rx_pool[%d], %lld buffs, %lld bytes each\n",
569 			   i, adapter->req_rx_add_entries_per_subcrq,
570 			   buff_size);
571 
572 		rx_pool->size = adapter->req_rx_add_entries_per_subcrq;
573 		rx_pool->index = i;
574 		rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
575 		rx_pool->active = 1;
576 
577 		rx_pool->free_map = kcalloc(rx_pool->size, sizeof(int),
578 					    GFP_KERNEL);
579 		if (!rx_pool->free_map) {
580 			release_rx_pools(adapter);
581 			return -1;
582 		}
583 
584 		rx_pool->rx_buff = kcalloc(rx_pool->size,
585 					   sizeof(struct ibmvnic_rx_buff),
586 					   GFP_KERNEL);
587 		if (!rx_pool->rx_buff) {
588 			dev_err(dev, "Couldn't alloc rx buffers\n");
589 			release_rx_pools(adapter);
590 			return -1;
591 		}
592 
593 		if (alloc_long_term_buff(adapter, &rx_pool->long_term_buff,
594 					 rx_pool->size * rx_pool->buff_size)) {
595 			release_rx_pools(adapter);
596 			return -1;
597 		}
598 
599 		for (j = 0; j < rx_pool->size; ++j)
600 			rx_pool->free_map[j] = j;
601 
602 		atomic_set(&rx_pool->available, 0);
603 		rx_pool->next_alloc = 0;
604 		rx_pool->next_free = 0;
605 	}
606 
607 	return 0;
608 }
609 
610 static int reset_one_tx_pool(struct ibmvnic_tx_pool *tx_pool)
611 {
612 	int rc, i;
613 
614 	rc = reset_long_term_buff(&tx_pool->long_term_buff);
615 	if (rc)
616 		return rc;
617 
618 	memset(tx_pool->tx_buff, 0,
619 	       tx_pool->num_buffers *
620 	       sizeof(struct ibmvnic_tx_buff));
621 
622 	for (i = 0; i < tx_pool->num_buffers; i++)
623 		tx_pool->free_map[i] = i;
624 
625 	tx_pool->consumer_index = 0;
626 	tx_pool->producer_index = 0;
627 
628 	return 0;
629 }
630 
631 static int reset_tx_pools(struct ibmvnic_adapter *adapter)
632 {
633 	int tx_scrqs;
634 	int i, rc;
635 
636 	if (!adapter->tx_pool)
637 		return -1;
638 
639 	tx_scrqs = adapter->num_active_tx_pools;
640 	for (i = 0; i < tx_scrqs; i++) {
641 		rc = reset_one_tx_pool(&adapter->tso_pool[i]);
642 		if (rc)
643 			return rc;
644 		rc = reset_one_tx_pool(&adapter->tx_pool[i]);
645 		if (rc)
646 			return rc;
647 	}
648 
649 	return 0;
650 }
651 
652 static void release_vpd_data(struct ibmvnic_adapter *adapter)
653 {
654 	if (!adapter->vpd)
655 		return;
656 
657 	kfree(adapter->vpd->buff);
658 	kfree(adapter->vpd);
659 
660 	adapter->vpd = NULL;
661 }
662 
663 static void release_one_tx_pool(struct ibmvnic_adapter *adapter,
664 				struct ibmvnic_tx_pool *tx_pool)
665 {
666 	kfree(tx_pool->tx_buff);
667 	kfree(tx_pool->free_map);
668 	free_long_term_buff(adapter, &tx_pool->long_term_buff);
669 }
670 
671 static void release_tx_pools(struct ibmvnic_adapter *adapter)
672 {
673 	int i;
674 
675 	if (!adapter->tx_pool)
676 		return;
677 
678 	for (i = 0; i < adapter->num_active_tx_pools; i++) {
679 		release_one_tx_pool(adapter, &adapter->tx_pool[i]);
680 		release_one_tx_pool(adapter, &adapter->tso_pool[i]);
681 	}
682 
683 	kfree(adapter->tx_pool);
684 	adapter->tx_pool = NULL;
685 	kfree(adapter->tso_pool);
686 	adapter->tso_pool = NULL;
687 	adapter->num_active_tx_pools = 0;
688 }
689 
690 static int init_one_tx_pool(struct net_device *netdev,
691 			    struct ibmvnic_tx_pool *tx_pool,
692 			    int num_entries, int buf_size)
693 {
694 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
695 	int i;
696 
697 	tx_pool->tx_buff = kcalloc(num_entries,
698 				   sizeof(struct ibmvnic_tx_buff),
699 				   GFP_KERNEL);
700 	if (!tx_pool->tx_buff)
701 		return -1;
702 
703 	if (alloc_long_term_buff(adapter, &tx_pool->long_term_buff,
704 				 num_entries * buf_size))
705 		return -1;
706 
707 	tx_pool->free_map = kcalloc(num_entries, sizeof(int), GFP_KERNEL);
708 	if (!tx_pool->free_map)
709 		return -1;
710 
711 	for (i = 0; i < num_entries; i++)
712 		tx_pool->free_map[i] = i;
713 
714 	tx_pool->consumer_index = 0;
715 	tx_pool->producer_index = 0;
716 	tx_pool->num_buffers = num_entries;
717 	tx_pool->buf_size = buf_size;
718 
719 	return 0;
720 }
721 
722 static int init_tx_pools(struct net_device *netdev)
723 {
724 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
725 	int tx_subcrqs;
726 	u64 buff_size;
727 	int i, rc;
728 
729 	tx_subcrqs = adapter->num_active_tx_scrqs;
730 	adapter->tx_pool = kcalloc(tx_subcrqs,
731 				   sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
732 	if (!adapter->tx_pool)
733 		return -1;
734 
735 	adapter->tso_pool = kcalloc(tx_subcrqs,
736 				    sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
737 	if (!adapter->tso_pool)
738 		return -1;
739 
740 	adapter->num_active_tx_pools = tx_subcrqs;
741 
742 	for (i = 0; i < tx_subcrqs; i++) {
743 		buff_size = adapter->req_mtu + VLAN_HLEN;
744 		buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
745 		rc = init_one_tx_pool(netdev, &adapter->tx_pool[i],
746 				      adapter->req_tx_entries_per_subcrq,
747 				      buff_size);
748 		if (rc) {
749 			release_tx_pools(adapter);
750 			return rc;
751 		}
752 
753 		rc = init_one_tx_pool(netdev, &adapter->tso_pool[i],
754 				      IBMVNIC_TSO_BUFS,
755 				      IBMVNIC_TSO_BUF_SZ);
756 		if (rc) {
757 			release_tx_pools(adapter);
758 			return rc;
759 		}
760 	}
761 
762 	return 0;
763 }
764 
765 static void ibmvnic_napi_enable(struct ibmvnic_adapter *adapter)
766 {
767 	int i;
768 
769 	if (adapter->napi_enabled)
770 		return;
771 
772 	for (i = 0; i < adapter->req_rx_queues; i++)
773 		napi_enable(&adapter->napi[i]);
774 
775 	adapter->napi_enabled = true;
776 }
777 
778 static void ibmvnic_napi_disable(struct ibmvnic_adapter *adapter)
779 {
780 	int i;
781 
782 	if (!adapter->napi_enabled)
783 		return;
784 
785 	for (i = 0; i < adapter->req_rx_queues; i++) {
786 		netdev_dbg(adapter->netdev, "Disabling napi[%d]\n", i);
787 		napi_disable(&adapter->napi[i]);
788 	}
789 
790 	adapter->napi_enabled = false;
791 }
792 
793 static int init_napi(struct ibmvnic_adapter *adapter)
794 {
795 	int i;
796 
797 	adapter->napi = kcalloc(adapter->req_rx_queues,
798 				sizeof(struct napi_struct), GFP_KERNEL);
799 	if (!adapter->napi)
800 		return -ENOMEM;
801 
802 	for (i = 0; i < adapter->req_rx_queues; i++) {
803 		netdev_dbg(adapter->netdev, "Adding napi[%d]\n", i);
804 		netif_napi_add(adapter->netdev, &adapter->napi[i],
805 			       ibmvnic_poll, NAPI_POLL_WEIGHT);
806 	}
807 
808 	adapter->num_active_rx_napi = adapter->req_rx_queues;
809 	return 0;
810 }
811 
812 static void release_napi(struct ibmvnic_adapter *adapter)
813 {
814 	int i;
815 
816 	if (!adapter->napi)
817 		return;
818 
819 	for (i = 0; i < adapter->num_active_rx_napi; i++) {
820 		netdev_dbg(adapter->netdev, "Releasing napi[%d]\n", i);
821 		netif_napi_del(&adapter->napi[i]);
822 	}
823 
824 	kfree(adapter->napi);
825 	adapter->napi = NULL;
826 	adapter->num_active_rx_napi = 0;
827 	adapter->napi_enabled = false;
828 }
829 
830 static int ibmvnic_login(struct net_device *netdev)
831 {
832 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
833 	unsigned long timeout = msecs_to_jiffies(20000);
834 	int retry_count = 0;
835 	int retries = 10;
836 	bool retry;
837 	int rc;
838 
839 	do {
840 		retry = false;
841 		if (retry_count > retries) {
842 			netdev_warn(netdev, "Login attempts exceeded\n");
843 			return -1;
844 		}
845 
846 		adapter->init_done_rc = 0;
847 		reinit_completion(&adapter->init_done);
848 		rc = send_login(adapter);
849 		if (rc)
850 			return rc;
851 
852 		if (!wait_for_completion_timeout(&adapter->init_done,
853 						 timeout)) {
854 			netdev_warn(netdev, "Login timed out, retrying...\n");
855 			retry = true;
856 			adapter->init_done_rc = 0;
857 			retry_count++;
858 			continue;
859 		}
860 
861 		if (adapter->init_done_rc == ABORTED) {
862 			netdev_warn(netdev, "Login aborted, retrying...\n");
863 			retry = true;
864 			adapter->init_done_rc = 0;
865 			retry_count++;
866 			/* FW or device may be busy, so
867 			 * wait a bit before retrying login
868 			 */
869 			msleep(500);
870 		} else if (adapter->init_done_rc == PARTIALSUCCESS) {
871 			retry_count++;
872 			release_sub_crqs(adapter, 1);
873 
874 			retry = true;
875 			netdev_dbg(netdev,
876 				   "Received partial success, retrying...\n");
877 			adapter->init_done_rc = 0;
878 			reinit_completion(&adapter->init_done);
879 			send_query_cap(adapter);
880 			if (!wait_for_completion_timeout(&adapter->init_done,
881 							 timeout)) {
882 				netdev_warn(netdev,
883 					    "Capabilities query timed out\n");
884 				return -1;
885 			}
886 
887 			rc = init_sub_crqs(adapter);
888 			if (rc) {
889 				netdev_warn(netdev,
890 					    "SCRQ initialization failed\n");
891 				return -1;
892 			}
893 
894 			rc = init_sub_crq_irqs(adapter);
895 			if (rc) {
896 				netdev_warn(netdev,
897 					    "SCRQ irq initialization failed\n");
898 				return -1;
899 			}
900 		} else if (adapter->init_done_rc) {
901 			netdev_warn(netdev, "Adapter login failed\n");
902 			return -1;
903 		}
904 	} while (retry);
905 
906 	__ibmvnic_set_mac(netdev, adapter->mac_addr);
907 
908 	netdev_dbg(netdev, "[S:%d] Login succeeded\n", adapter->state);
909 	return 0;
910 }
911 
912 static void release_login_buffer(struct ibmvnic_adapter *adapter)
913 {
914 	kfree(adapter->login_buf);
915 	adapter->login_buf = NULL;
916 }
917 
918 static void release_login_rsp_buffer(struct ibmvnic_adapter *adapter)
919 {
920 	kfree(adapter->login_rsp_buf);
921 	adapter->login_rsp_buf = NULL;
922 }
923 
924 static void release_resources(struct ibmvnic_adapter *adapter)
925 {
926 	release_vpd_data(adapter);
927 
928 	release_tx_pools(adapter);
929 	release_rx_pools(adapter);
930 
931 	release_napi(adapter);
932 	release_login_buffer(adapter);
933 	release_login_rsp_buffer(adapter);
934 }
935 
936 static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state)
937 {
938 	struct net_device *netdev = adapter->netdev;
939 	unsigned long timeout = msecs_to_jiffies(20000);
940 	union ibmvnic_crq crq;
941 	bool resend;
942 	int rc;
943 
944 	netdev_dbg(netdev, "setting link state %d\n", link_state);
945 
946 	memset(&crq, 0, sizeof(crq));
947 	crq.logical_link_state.first = IBMVNIC_CRQ_CMD;
948 	crq.logical_link_state.cmd = LOGICAL_LINK_STATE;
949 	crq.logical_link_state.link_state = link_state;
950 
951 	do {
952 		resend = false;
953 
954 		reinit_completion(&adapter->init_done);
955 		rc = ibmvnic_send_crq(adapter, &crq);
956 		if (rc) {
957 			netdev_err(netdev, "Failed to set link state\n");
958 			return rc;
959 		}
960 
961 		if (!wait_for_completion_timeout(&adapter->init_done,
962 						 timeout)) {
963 			netdev_err(netdev, "timeout setting link state\n");
964 			return -1;
965 		}
966 
967 		if (adapter->init_done_rc == PARTIALSUCCESS) {
968 			/* Partuial success, delay and re-send */
969 			mdelay(1000);
970 			resend = true;
971 		} else if (adapter->init_done_rc) {
972 			netdev_warn(netdev, "Unable to set link state, rc=%d\n",
973 				    adapter->init_done_rc);
974 			return adapter->init_done_rc;
975 		}
976 	} while (resend);
977 
978 	return 0;
979 }
980 
981 static int set_real_num_queues(struct net_device *netdev)
982 {
983 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
984 	int rc;
985 
986 	netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n",
987 		   adapter->req_tx_queues, adapter->req_rx_queues);
988 
989 	rc = netif_set_real_num_tx_queues(netdev, adapter->req_tx_queues);
990 	if (rc) {
991 		netdev_err(netdev, "failed to set the number of tx queues\n");
992 		return rc;
993 	}
994 
995 	rc = netif_set_real_num_rx_queues(netdev, adapter->req_rx_queues);
996 	if (rc)
997 		netdev_err(netdev, "failed to set the number of rx queues\n");
998 
999 	return rc;
1000 }
1001 
1002 static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter)
1003 {
1004 	struct device *dev = &adapter->vdev->dev;
1005 	union ibmvnic_crq crq;
1006 	int len = 0;
1007 	int rc;
1008 
1009 	if (adapter->vpd->buff)
1010 		len = adapter->vpd->len;
1011 
1012 	mutex_lock(&adapter->fw_lock);
1013 	adapter->fw_done_rc = 0;
1014 	reinit_completion(&adapter->fw_done);
1015 
1016 	crq.get_vpd_size.first = IBMVNIC_CRQ_CMD;
1017 	crq.get_vpd_size.cmd = GET_VPD_SIZE;
1018 	rc = ibmvnic_send_crq(adapter, &crq);
1019 	if (rc) {
1020 		mutex_unlock(&adapter->fw_lock);
1021 		return rc;
1022 	}
1023 
1024 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1025 	if (rc) {
1026 		dev_err(dev, "Could not retrieve VPD size, rc = %d\n", rc);
1027 		mutex_unlock(&adapter->fw_lock);
1028 		return rc;
1029 	}
1030 	mutex_unlock(&adapter->fw_lock);
1031 
1032 	if (!adapter->vpd->len)
1033 		return -ENODATA;
1034 
1035 	if (!adapter->vpd->buff)
1036 		adapter->vpd->buff = kzalloc(adapter->vpd->len, GFP_KERNEL);
1037 	else if (adapter->vpd->len != len)
1038 		adapter->vpd->buff =
1039 			krealloc(adapter->vpd->buff,
1040 				 adapter->vpd->len, GFP_KERNEL);
1041 
1042 	if (!adapter->vpd->buff) {
1043 		dev_err(dev, "Could allocate VPD buffer\n");
1044 		return -ENOMEM;
1045 	}
1046 
1047 	adapter->vpd->dma_addr =
1048 		dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len,
1049 			       DMA_FROM_DEVICE);
1050 	if (dma_mapping_error(dev, adapter->vpd->dma_addr)) {
1051 		dev_err(dev, "Could not map VPD buffer\n");
1052 		kfree(adapter->vpd->buff);
1053 		adapter->vpd->buff = NULL;
1054 		return -ENOMEM;
1055 	}
1056 
1057 	mutex_lock(&adapter->fw_lock);
1058 	adapter->fw_done_rc = 0;
1059 	reinit_completion(&adapter->fw_done);
1060 
1061 	crq.get_vpd.first = IBMVNIC_CRQ_CMD;
1062 	crq.get_vpd.cmd = GET_VPD;
1063 	crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr);
1064 	crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len);
1065 	rc = ibmvnic_send_crq(adapter, &crq);
1066 	if (rc) {
1067 		kfree(adapter->vpd->buff);
1068 		adapter->vpd->buff = NULL;
1069 		mutex_unlock(&adapter->fw_lock);
1070 		return rc;
1071 	}
1072 
1073 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1074 	if (rc) {
1075 		dev_err(dev, "Unable to retrieve VPD, rc = %d\n", rc);
1076 		kfree(adapter->vpd->buff);
1077 		adapter->vpd->buff = NULL;
1078 		mutex_unlock(&adapter->fw_lock);
1079 		return rc;
1080 	}
1081 
1082 	mutex_unlock(&adapter->fw_lock);
1083 	return 0;
1084 }
1085 
1086 static int init_resources(struct ibmvnic_adapter *adapter)
1087 {
1088 	struct net_device *netdev = adapter->netdev;
1089 	int rc;
1090 
1091 	rc = set_real_num_queues(netdev);
1092 	if (rc)
1093 		return rc;
1094 
1095 	adapter->vpd = kzalloc(sizeof(*adapter->vpd), GFP_KERNEL);
1096 	if (!adapter->vpd)
1097 		return -ENOMEM;
1098 
1099 	/* Vital Product Data (VPD) */
1100 	rc = ibmvnic_get_vpd(adapter);
1101 	if (rc) {
1102 		netdev_err(netdev, "failed to initialize Vital Product Data (VPD)\n");
1103 		return rc;
1104 	}
1105 
1106 	adapter->map_id = 1;
1107 
1108 	rc = init_napi(adapter);
1109 	if (rc)
1110 		return rc;
1111 
1112 	send_query_map(adapter);
1113 
1114 	rc = init_rx_pools(netdev);
1115 	if (rc)
1116 		return rc;
1117 
1118 	rc = init_tx_pools(netdev);
1119 	return rc;
1120 }
1121 
1122 static int __ibmvnic_open(struct net_device *netdev)
1123 {
1124 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1125 	enum vnic_state prev_state = adapter->state;
1126 	int i, rc;
1127 
1128 	adapter->state = VNIC_OPENING;
1129 	replenish_pools(adapter);
1130 	ibmvnic_napi_enable(adapter);
1131 
1132 	/* We're ready to receive frames, enable the sub-crq interrupts and
1133 	 * set the logical link state to up
1134 	 */
1135 	for (i = 0; i < adapter->req_rx_queues; i++) {
1136 		netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i);
1137 		if (prev_state == VNIC_CLOSED)
1138 			enable_irq(adapter->rx_scrq[i]->irq);
1139 		enable_scrq_irq(adapter, adapter->rx_scrq[i]);
1140 	}
1141 
1142 	for (i = 0; i < adapter->req_tx_queues; i++) {
1143 		netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i);
1144 		if (prev_state == VNIC_CLOSED)
1145 			enable_irq(adapter->tx_scrq[i]->irq);
1146 		enable_scrq_irq(adapter, adapter->tx_scrq[i]);
1147 		netdev_tx_reset_queue(netdev_get_tx_queue(netdev, i));
1148 	}
1149 
1150 	rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP);
1151 	if (rc) {
1152 		ibmvnic_napi_disable(adapter);
1153 		release_resources(adapter);
1154 		return rc;
1155 	}
1156 
1157 	netif_tx_start_all_queues(netdev);
1158 
1159 	adapter->state = VNIC_OPEN;
1160 	return rc;
1161 }
1162 
1163 static int ibmvnic_open(struct net_device *netdev)
1164 {
1165 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1166 	int rc;
1167 
1168 	ASSERT_RTNL();
1169 
1170 	/* If device failover is pending or we are about to reset, just set
1171 	 * device state and return. Device operation will be handled by reset
1172 	 * routine.
1173 	 *
1174 	 * It should be safe to overwrite the adapter->state here. Since
1175 	 * we hold the rtnl, either the reset has not actually started or
1176 	 * the rtnl got dropped during the set_link_state() in do_reset().
1177 	 * In the former case, no one else is changing the state (again we
1178 	 * have the rtnl) and in the latter case, do_reset() will detect and
1179 	 * honor our setting below.
1180 	 */
1181 	if (adapter->failover_pending || (test_bit(0, &adapter->resetting))) {
1182 		netdev_dbg(netdev, "[S:%d FOP:%d] Resetting, deferring open\n",
1183 			   adapter->state, adapter->failover_pending);
1184 		adapter->state = VNIC_OPEN;
1185 		rc = 0;
1186 		goto out;
1187 	}
1188 
1189 	if (adapter->state != VNIC_CLOSED) {
1190 		rc = ibmvnic_login(netdev);
1191 		if (rc)
1192 			goto out;
1193 
1194 		rc = init_resources(adapter);
1195 		if (rc) {
1196 			netdev_err(netdev, "failed to initialize resources\n");
1197 			release_resources(adapter);
1198 			goto out;
1199 		}
1200 	}
1201 
1202 	rc = __ibmvnic_open(netdev);
1203 
1204 out:
1205 	/* If open failed and there is a pending failover or in-progress reset,
1206 	 * set device state and return. Device operation will be handled by
1207 	 * reset routine. See also comments above regarding rtnl.
1208 	 */
1209 	if (rc &&
1210 	    (adapter->failover_pending || (test_bit(0, &adapter->resetting)))) {
1211 		adapter->state = VNIC_OPEN;
1212 		rc = 0;
1213 	}
1214 	return rc;
1215 }
1216 
1217 static void clean_rx_pools(struct ibmvnic_adapter *adapter)
1218 {
1219 	struct ibmvnic_rx_pool *rx_pool;
1220 	struct ibmvnic_rx_buff *rx_buff;
1221 	u64 rx_entries;
1222 	int rx_scrqs;
1223 	int i, j;
1224 
1225 	if (!adapter->rx_pool)
1226 		return;
1227 
1228 	rx_scrqs = adapter->num_active_rx_pools;
1229 	rx_entries = adapter->req_rx_add_entries_per_subcrq;
1230 
1231 	/* Free any remaining skbs in the rx buffer pools */
1232 	for (i = 0; i < rx_scrqs; i++) {
1233 		rx_pool = &adapter->rx_pool[i];
1234 		if (!rx_pool || !rx_pool->rx_buff)
1235 			continue;
1236 
1237 		netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i);
1238 		for (j = 0; j < rx_entries; j++) {
1239 			rx_buff = &rx_pool->rx_buff[j];
1240 			if (rx_buff && rx_buff->skb) {
1241 				dev_kfree_skb_any(rx_buff->skb);
1242 				rx_buff->skb = NULL;
1243 			}
1244 		}
1245 	}
1246 }
1247 
1248 static void clean_one_tx_pool(struct ibmvnic_adapter *adapter,
1249 			      struct ibmvnic_tx_pool *tx_pool)
1250 {
1251 	struct ibmvnic_tx_buff *tx_buff;
1252 	u64 tx_entries;
1253 	int i;
1254 
1255 	if (!tx_pool || !tx_pool->tx_buff)
1256 		return;
1257 
1258 	tx_entries = tx_pool->num_buffers;
1259 
1260 	for (i = 0; i < tx_entries; i++) {
1261 		tx_buff = &tx_pool->tx_buff[i];
1262 		if (tx_buff && tx_buff->skb) {
1263 			dev_kfree_skb_any(tx_buff->skb);
1264 			tx_buff->skb = NULL;
1265 		}
1266 	}
1267 }
1268 
1269 static void clean_tx_pools(struct ibmvnic_adapter *adapter)
1270 {
1271 	int tx_scrqs;
1272 	int i;
1273 
1274 	if (!adapter->tx_pool || !adapter->tso_pool)
1275 		return;
1276 
1277 	tx_scrqs = adapter->num_active_tx_pools;
1278 
1279 	/* Free any remaining skbs in the tx buffer pools */
1280 	for (i = 0; i < tx_scrqs; i++) {
1281 		netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i);
1282 		clean_one_tx_pool(adapter, &adapter->tx_pool[i]);
1283 		clean_one_tx_pool(adapter, &adapter->tso_pool[i]);
1284 	}
1285 }
1286 
1287 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter)
1288 {
1289 	struct net_device *netdev = adapter->netdev;
1290 	int i;
1291 
1292 	if (adapter->tx_scrq) {
1293 		for (i = 0; i < adapter->req_tx_queues; i++)
1294 			if (adapter->tx_scrq[i]->irq) {
1295 				netdev_dbg(netdev,
1296 					   "Disabling tx_scrq[%d] irq\n", i);
1297 				disable_scrq_irq(adapter, adapter->tx_scrq[i]);
1298 				disable_irq(adapter->tx_scrq[i]->irq);
1299 			}
1300 	}
1301 
1302 	if (adapter->rx_scrq) {
1303 		for (i = 0; i < adapter->req_rx_queues; i++) {
1304 			if (adapter->rx_scrq[i]->irq) {
1305 				netdev_dbg(netdev,
1306 					   "Disabling rx_scrq[%d] irq\n", i);
1307 				disable_scrq_irq(adapter, adapter->rx_scrq[i]);
1308 				disable_irq(adapter->rx_scrq[i]->irq);
1309 			}
1310 		}
1311 	}
1312 }
1313 
1314 static void ibmvnic_cleanup(struct net_device *netdev)
1315 {
1316 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1317 
1318 	/* ensure that transmissions are stopped if called by do_reset */
1319 	if (test_bit(0, &adapter->resetting))
1320 		netif_tx_disable(netdev);
1321 	else
1322 		netif_tx_stop_all_queues(netdev);
1323 
1324 	ibmvnic_napi_disable(adapter);
1325 	ibmvnic_disable_irqs(adapter);
1326 
1327 	clean_rx_pools(adapter);
1328 	clean_tx_pools(adapter);
1329 }
1330 
1331 static int __ibmvnic_close(struct net_device *netdev)
1332 {
1333 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1334 	int rc = 0;
1335 
1336 	adapter->state = VNIC_CLOSING;
1337 	rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
1338 	adapter->state = VNIC_CLOSED;
1339 	return rc;
1340 }
1341 
1342 static int ibmvnic_close(struct net_device *netdev)
1343 {
1344 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1345 	int rc;
1346 
1347 	netdev_dbg(netdev, "[S:%d FOP:%d FRR:%d] Closing\n",
1348 		   adapter->state, adapter->failover_pending,
1349 		   adapter->force_reset_recovery);
1350 
1351 	/* If device failover is pending, just set device state and return.
1352 	 * Device operation will be handled by reset routine.
1353 	 */
1354 	if (adapter->failover_pending) {
1355 		adapter->state = VNIC_CLOSED;
1356 		return 0;
1357 	}
1358 
1359 	rc = __ibmvnic_close(netdev);
1360 	ibmvnic_cleanup(netdev);
1361 
1362 	return rc;
1363 }
1364 
1365 /**
1366  * build_hdr_data - creates L2/L3/L4 header data buffer
1367  * @hdr_field: bitfield determining needed headers
1368  * @skb: socket buffer
1369  * @hdr_len: array of header lengths
1370  * @hdr_data: buffer to write the header to
1371  *
1372  * Reads hdr_field to determine which headers are needed by firmware.
1373  * Builds a buffer containing these headers.  Saves individual header
1374  * lengths and total buffer length to be used to build descriptors.
1375  */
1376 static int build_hdr_data(u8 hdr_field, struct sk_buff *skb,
1377 			  int *hdr_len, u8 *hdr_data)
1378 {
1379 	int len = 0;
1380 	u8 *hdr;
1381 
1382 	if (skb_vlan_tagged(skb) && !skb_vlan_tag_present(skb))
1383 		hdr_len[0] = sizeof(struct vlan_ethhdr);
1384 	else
1385 		hdr_len[0] = sizeof(struct ethhdr);
1386 
1387 	if (skb->protocol == htons(ETH_P_IP)) {
1388 		hdr_len[1] = ip_hdr(skb)->ihl * 4;
1389 		if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1390 			hdr_len[2] = tcp_hdrlen(skb);
1391 		else if (ip_hdr(skb)->protocol == IPPROTO_UDP)
1392 			hdr_len[2] = sizeof(struct udphdr);
1393 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
1394 		hdr_len[1] = sizeof(struct ipv6hdr);
1395 		if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1396 			hdr_len[2] = tcp_hdrlen(skb);
1397 		else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
1398 			hdr_len[2] = sizeof(struct udphdr);
1399 	} else if (skb->protocol == htons(ETH_P_ARP)) {
1400 		hdr_len[1] = arp_hdr_len(skb->dev);
1401 		hdr_len[2] = 0;
1402 	}
1403 
1404 	memset(hdr_data, 0, 120);
1405 	if ((hdr_field >> 6) & 1) {
1406 		hdr = skb_mac_header(skb);
1407 		memcpy(hdr_data, hdr, hdr_len[0]);
1408 		len += hdr_len[0];
1409 	}
1410 
1411 	if ((hdr_field >> 5) & 1) {
1412 		hdr = skb_network_header(skb);
1413 		memcpy(hdr_data + len, hdr, hdr_len[1]);
1414 		len += hdr_len[1];
1415 	}
1416 
1417 	if ((hdr_field >> 4) & 1) {
1418 		hdr = skb_transport_header(skb);
1419 		memcpy(hdr_data + len, hdr, hdr_len[2]);
1420 		len += hdr_len[2];
1421 	}
1422 	return len;
1423 }
1424 
1425 /**
1426  * create_hdr_descs - create header and header extension descriptors
1427  * @hdr_field: bitfield determining needed headers
1428  * @hdr_data: buffer containing header data
1429  * @len: length of data buffer
1430  * @hdr_len: array of individual header lengths
1431  * @scrq_arr: descriptor array
1432  *
1433  * Creates header and, if needed, header extension descriptors and
1434  * places them in a descriptor array, scrq_arr
1435  */
1436 
1437 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len,
1438 			    union sub_crq *scrq_arr)
1439 {
1440 	union sub_crq hdr_desc;
1441 	int tmp_len = len;
1442 	int num_descs = 0;
1443 	u8 *data, *cur;
1444 	int tmp;
1445 
1446 	while (tmp_len > 0) {
1447 		cur = hdr_data + len - tmp_len;
1448 
1449 		memset(&hdr_desc, 0, sizeof(hdr_desc));
1450 		if (cur != hdr_data) {
1451 			data = hdr_desc.hdr_ext.data;
1452 			tmp = tmp_len > 29 ? 29 : tmp_len;
1453 			hdr_desc.hdr_ext.first = IBMVNIC_CRQ_CMD;
1454 			hdr_desc.hdr_ext.type = IBMVNIC_HDR_EXT_DESC;
1455 			hdr_desc.hdr_ext.len = tmp;
1456 		} else {
1457 			data = hdr_desc.hdr.data;
1458 			tmp = tmp_len > 24 ? 24 : tmp_len;
1459 			hdr_desc.hdr.first = IBMVNIC_CRQ_CMD;
1460 			hdr_desc.hdr.type = IBMVNIC_HDR_DESC;
1461 			hdr_desc.hdr.len = tmp;
1462 			hdr_desc.hdr.l2_len = (u8)hdr_len[0];
1463 			hdr_desc.hdr.l3_len = cpu_to_be16((u16)hdr_len[1]);
1464 			hdr_desc.hdr.l4_len = (u8)hdr_len[2];
1465 			hdr_desc.hdr.flag = hdr_field << 1;
1466 		}
1467 		memcpy(data, cur, tmp);
1468 		tmp_len -= tmp;
1469 		*scrq_arr = hdr_desc;
1470 		scrq_arr++;
1471 		num_descs++;
1472 	}
1473 
1474 	return num_descs;
1475 }
1476 
1477 /**
1478  * build_hdr_descs_arr - build a header descriptor array
1479  * @txbuff: tx buffer
1480  * @num_entries: number of descriptors to be sent
1481  * @hdr_field: bit field determining which headers will be sent
1482  *
1483  * This function will build a TX descriptor array with applicable
1484  * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect.
1485  */
1486 
1487 static void build_hdr_descs_arr(struct sk_buff *skb,
1488 				union sub_crq *indir_arr,
1489 				int *num_entries, u8 hdr_field)
1490 {
1491 	int hdr_len[3] = {0, 0, 0};
1492 	u8 hdr_data[140] = {0};
1493 	int tot_len;
1494 
1495 	tot_len = build_hdr_data(hdr_field, skb, hdr_len,
1496 				 hdr_data);
1497 	*num_entries += create_hdr_descs(hdr_field, hdr_data, tot_len, hdr_len,
1498 					 indir_arr + 1);
1499 }
1500 
1501 static int ibmvnic_xmit_workarounds(struct sk_buff *skb,
1502 				    struct net_device *netdev)
1503 {
1504 	/* For some backing devices, mishandling of small packets
1505 	 * can result in a loss of connection or TX stall. Device
1506 	 * architects recommend that no packet should be smaller
1507 	 * than the minimum MTU value provided to the driver, so
1508 	 * pad any packets to that length
1509 	 */
1510 	if (skb->len < netdev->min_mtu)
1511 		return skb_put_padto(skb, netdev->min_mtu);
1512 
1513 	return 0;
1514 }
1515 
1516 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
1517 					 struct ibmvnic_sub_crq_queue *tx_scrq)
1518 {
1519 	struct ibmvnic_ind_xmit_queue *ind_bufp;
1520 	struct ibmvnic_tx_buff *tx_buff;
1521 	struct ibmvnic_tx_pool *tx_pool;
1522 	union sub_crq tx_scrq_entry;
1523 	int queue_num;
1524 	int entries;
1525 	int index;
1526 	int i;
1527 
1528 	ind_bufp = &tx_scrq->ind_buf;
1529 	entries = (u64)ind_bufp->index;
1530 	queue_num = tx_scrq->pool_index;
1531 
1532 	for (i = entries - 1; i >= 0; --i) {
1533 		tx_scrq_entry = ind_bufp->indir_arr[i];
1534 		if (tx_scrq_entry.v1.type != IBMVNIC_TX_DESC)
1535 			continue;
1536 		index = be32_to_cpu(tx_scrq_entry.v1.correlator);
1537 		if (index & IBMVNIC_TSO_POOL_MASK) {
1538 			tx_pool = &adapter->tso_pool[queue_num];
1539 			index &= ~IBMVNIC_TSO_POOL_MASK;
1540 		} else {
1541 			tx_pool = &adapter->tx_pool[queue_num];
1542 		}
1543 		tx_pool->free_map[tx_pool->consumer_index] = index;
1544 		tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
1545 					  tx_pool->num_buffers - 1 :
1546 					  tx_pool->consumer_index - 1;
1547 		tx_buff = &tx_pool->tx_buff[index];
1548 		adapter->netdev->stats.tx_packets--;
1549 		adapter->netdev->stats.tx_bytes -= tx_buff->skb->len;
1550 		adapter->tx_stats_buffers[queue_num].packets--;
1551 		adapter->tx_stats_buffers[queue_num].bytes -=
1552 						tx_buff->skb->len;
1553 		dev_kfree_skb_any(tx_buff->skb);
1554 		tx_buff->skb = NULL;
1555 		adapter->netdev->stats.tx_dropped++;
1556 	}
1557 	ind_bufp->index = 0;
1558 	if (atomic_sub_return(entries, &tx_scrq->used) <=
1559 	    (adapter->req_tx_entries_per_subcrq / 2) &&
1560 	    __netif_subqueue_stopped(adapter->netdev, queue_num)) {
1561 		netif_wake_subqueue(adapter->netdev, queue_num);
1562 		netdev_dbg(adapter->netdev, "Started queue %d\n",
1563 			   queue_num);
1564 	}
1565 }
1566 
1567 static int ibmvnic_tx_scrq_flush(struct ibmvnic_adapter *adapter,
1568 				 struct ibmvnic_sub_crq_queue *tx_scrq)
1569 {
1570 	struct ibmvnic_ind_xmit_queue *ind_bufp;
1571 	u64 dma_addr;
1572 	u64 entries;
1573 	u64 handle;
1574 	int rc;
1575 
1576 	ind_bufp = &tx_scrq->ind_buf;
1577 	dma_addr = (u64)ind_bufp->indir_dma;
1578 	entries = (u64)ind_bufp->index;
1579 	handle = tx_scrq->handle;
1580 
1581 	if (!entries)
1582 		return 0;
1583 	rc = send_subcrq_indirect(adapter, handle, dma_addr, entries);
1584 	if (rc)
1585 		ibmvnic_tx_scrq_clean_buffer(adapter, tx_scrq);
1586 	else
1587 		ind_bufp->index = 0;
1588 	return 0;
1589 }
1590 
1591 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
1592 {
1593 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1594 	int queue_num = skb_get_queue_mapping(skb);
1595 	u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req;
1596 	struct device *dev = &adapter->vdev->dev;
1597 	struct ibmvnic_ind_xmit_queue *ind_bufp;
1598 	struct ibmvnic_tx_buff *tx_buff = NULL;
1599 	struct ibmvnic_sub_crq_queue *tx_scrq;
1600 	struct ibmvnic_tx_pool *tx_pool;
1601 	unsigned int tx_send_failed = 0;
1602 	netdev_tx_t ret = NETDEV_TX_OK;
1603 	unsigned int tx_map_failed = 0;
1604 	union sub_crq indir_arr[16];
1605 	unsigned int tx_dropped = 0;
1606 	unsigned int tx_packets = 0;
1607 	unsigned int tx_bytes = 0;
1608 	dma_addr_t data_dma_addr;
1609 	struct netdev_queue *txq;
1610 	unsigned long lpar_rc;
1611 	union sub_crq tx_crq;
1612 	unsigned int offset;
1613 	int num_entries = 1;
1614 	unsigned char *dst;
1615 	int index = 0;
1616 	u8 proto = 0;
1617 
1618 	tx_scrq = adapter->tx_scrq[queue_num];
1619 	txq = netdev_get_tx_queue(netdev, queue_num);
1620 	ind_bufp = &tx_scrq->ind_buf;
1621 
1622 	if (test_bit(0, &adapter->resetting)) {
1623 		if (!netif_subqueue_stopped(netdev, skb))
1624 			netif_stop_subqueue(netdev, queue_num);
1625 		dev_kfree_skb_any(skb);
1626 
1627 		tx_send_failed++;
1628 		tx_dropped++;
1629 		ret = NETDEV_TX_OK;
1630 		ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1631 		goto out;
1632 	}
1633 
1634 	if (ibmvnic_xmit_workarounds(skb, netdev)) {
1635 		tx_dropped++;
1636 		tx_send_failed++;
1637 		ret = NETDEV_TX_OK;
1638 		ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1639 		goto out;
1640 	}
1641 	if (skb_is_gso(skb))
1642 		tx_pool = &adapter->tso_pool[queue_num];
1643 	else
1644 		tx_pool = &adapter->tx_pool[queue_num];
1645 
1646 	index = tx_pool->free_map[tx_pool->consumer_index];
1647 
1648 	if (index == IBMVNIC_INVALID_MAP) {
1649 		dev_kfree_skb_any(skb);
1650 		tx_send_failed++;
1651 		tx_dropped++;
1652 		ret = NETDEV_TX_OK;
1653 		ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1654 		goto out;
1655 	}
1656 
1657 	tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP;
1658 
1659 	offset = index * tx_pool->buf_size;
1660 	dst = tx_pool->long_term_buff.buff + offset;
1661 	memset(dst, 0, tx_pool->buf_size);
1662 	data_dma_addr = tx_pool->long_term_buff.addr + offset;
1663 
1664 	if (skb_shinfo(skb)->nr_frags) {
1665 		int cur, i;
1666 
1667 		/* Copy the head */
1668 		skb_copy_from_linear_data(skb, dst, skb_headlen(skb));
1669 		cur = skb_headlen(skb);
1670 
1671 		/* Copy the frags */
1672 		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1673 			const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1674 
1675 			memcpy(dst + cur,
1676 			       page_address(skb_frag_page(frag)) +
1677 			       skb_frag_off(frag), skb_frag_size(frag));
1678 			cur += skb_frag_size(frag);
1679 		}
1680 	} else {
1681 		skb_copy_from_linear_data(skb, dst, skb->len);
1682 	}
1683 
1684 	/* post changes to long_term_buff *dst before VIOS accessing it */
1685 	dma_wmb();
1686 
1687 	tx_pool->consumer_index =
1688 	    (tx_pool->consumer_index + 1) % tx_pool->num_buffers;
1689 
1690 	tx_buff = &tx_pool->tx_buff[index];
1691 	tx_buff->skb = skb;
1692 	tx_buff->index = index;
1693 	tx_buff->pool_index = queue_num;
1694 
1695 	memset(&tx_crq, 0, sizeof(tx_crq));
1696 	tx_crq.v1.first = IBMVNIC_CRQ_CMD;
1697 	tx_crq.v1.type = IBMVNIC_TX_DESC;
1698 	tx_crq.v1.n_crq_elem = 1;
1699 	tx_crq.v1.n_sge = 1;
1700 	tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED;
1701 
1702 	if (skb_is_gso(skb))
1703 		tx_crq.v1.correlator =
1704 			cpu_to_be32(index | IBMVNIC_TSO_POOL_MASK);
1705 	else
1706 		tx_crq.v1.correlator = cpu_to_be32(index);
1707 	tx_crq.v1.dma_reg = cpu_to_be16(tx_pool->long_term_buff.map_id);
1708 	tx_crq.v1.sge_len = cpu_to_be32(skb->len);
1709 	tx_crq.v1.ioba = cpu_to_be64(data_dma_addr);
1710 
1711 	if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) {
1712 		tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT;
1713 		tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci);
1714 	}
1715 
1716 	if (skb->protocol == htons(ETH_P_IP)) {
1717 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4;
1718 		proto = ip_hdr(skb)->protocol;
1719 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
1720 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6;
1721 		proto = ipv6_hdr(skb)->nexthdr;
1722 	}
1723 
1724 	if (proto == IPPROTO_TCP)
1725 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP;
1726 	else if (proto == IPPROTO_UDP)
1727 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP;
1728 
1729 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
1730 		tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD;
1731 		hdrs += 2;
1732 	}
1733 	if (skb_is_gso(skb)) {
1734 		tx_crq.v1.flags1 |= IBMVNIC_TX_LSO;
1735 		tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size);
1736 		hdrs += 2;
1737 	}
1738 
1739 	if ((*hdrs >> 7) & 1)
1740 		build_hdr_descs_arr(skb, indir_arr, &num_entries, *hdrs);
1741 
1742 	tx_crq.v1.n_crq_elem = num_entries;
1743 	tx_buff->num_entries = num_entries;
1744 	/* flush buffer if current entry can not fit */
1745 	if (num_entries + ind_bufp->index > IBMVNIC_MAX_IND_DESCS) {
1746 		lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1747 		if (lpar_rc != H_SUCCESS)
1748 			goto tx_flush_err;
1749 	}
1750 
1751 	indir_arr[0] = tx_crq;
1752 	memcpy(&ind_bufp->indir_arr[ind_bufp->index], &indir_arr[0],
1753 	       num_entries * sizeof(struct ibmvnic_generic_scrq));
1754 	ind_bufp->index += num_entries;
1755 	if (__netdev_tx_sent_queue(txq, skb->len,
1756 				   netdev_xmit_more() &&
1757 				   ind_bufp->index < IBMVNIC_MAX_IND_DESCS)) {
1758 		lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1759 		if (lpar_rc != H_SUCCESS)
1760 			goto tx_err;
1761 	}
1762 
1763 	if (atomic_add_return(num_entries, &tx_scrq->used)
1764 					>= adapter->req_tx_entries_per_subcrq) {
1765 		netdev_dbg(netdev, "Stopping queue %d\n", queue_num);
1766 		netif_stop_subqueue(netdev, queue_num);
1767 	}
1768 
1769 	tx_packets++;
1770 	tx_bytes += skb->len;
1771 	txq->trans_start = jiffies;
1772 	ret = NETDEV_TX_OK;
1773 	goto out;
1774 
1775 tx_flush_err:
1776 	dev_kfree_skb_any(skb);
1777 	tx_buff->skb = NULL;
1778 	tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
1779 				  tx_pool->num_buffers - 1 :
1780 				  tx_pool->consumer_index - 1;
1781 	tx_dropped++;
1782 tx_err:
1783 	if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER)
1784 		dev_err_ratelimited(dev, "tx: send failed\n");
1785 
1786 	if (lpar_rc == H_CLOSED || adapter->failover_pending) {
1787 		/* Disable TX and report carrier off if queue is closed
1788 		 * or pending failover.
1789 		 * Firmware guarantees that a signal will be sent to the
1790 		 * driver, triggering a reset or some other action.
1791 		 */
1792 		netif_tx_stop_all_queues(netdev);
1793 		netif_carrier_off(netdev);
1794 	}
1795 out:
1796 	netdev->stats.tx_dropped += tx_dropped;
1797 	netdev->stats.tx_bytes += tx_bytes;
1798 	netdev->stats.tx_packets += tx_packets;
1799 	adapter->tx_send_failed += tx_send_failed;
1800 	adapter->tx_map_failed += tx_map_failed;
1801 	adapter->tx_stats_buffers[queue_num].packets += tx_packets;
1802 	adapter->tx_stats_buffers[queue_num].bytes += tx_bytes;
1803 	adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped;
1804 
1805 	return ret;
1806 }
1807 
1808 static void ibmvnic_set_multi(struct net_device *netdev)
1809 {
1810 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1811 	struct netdev_hw_addr *ha;
1812 	union ibmvnic_crq crq;
1813 
1814 	memset(&crq, 0, sizeof(crq));
1815 	crq.request_capability.first = IBMVNIC_CRQ_CMD;
1816 	crq.request_capability.cmd = REQUEST_CAPABILITY;
1817 
1818 	if (netdev->flags & IFF_PROMISC) {
1819 		if (!adapter->promisc_supported)
1820 			return;
1821 	} else {
1822 		if (netdev->flags & IFF_ALLMULTI) {
1823 			/* Accept all multicast */
1824 			memset(&crq, 0, sizeof(crq));
1825 			crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1826 			crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1827 			crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL;
1828 			ibmvnic_send_crq(adapter, &crq);
1829 		} else if (netdev_mc_empty(netdev)) {
1830 			/* Reject all multicast */
1831 			memset(&crq, 0, sizeof(crq));
1832 			crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1833 			crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1834 			crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL;
1835 			ibmvnic_send_crq(adapter, &crq);
1836 		} else {
1837 			/* Accept one or more multicast(s) */
1838 			netdev_for_each_mc_addr(ha, netdev) {
1839 				memset(&crq, 0, sizeof(crq));
1840 				crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1841 				crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1842 				crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC;
1843 				ether_addr_copy(&crq.multicast_ctrl.mac_addr[0],
1844 						ha->addr);
1845 				ibmvnic_send_crq(adapter, &crq);
1846 			}
1847 		}
1848 	}
1849 }
1850 
1851 static int __ibmvnic_set_mac(struct net_device *netdev, u8 *dev_addr)
1852 {
1853 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1854 	union ibmvnic_crq crq;
1855 	int rc;
1856 
1857 	if (!is_valid_ether_addr(dev_addr)) {
1858 		rc = -EADDRNOTAVAIL;
1859 		goto err;
1860 	}
1861 
1862 	memset(&crq, 0, sizeof(crq));
1863 	crq.change_mac_addr.first = IBMVNIC_CRQ_CMD;
1864 	crq.change_mac_addr.cmd = CHANGE_MAC_ADDR;
1865 	ether_addr_copy(&crq.change_mac_addr.mac_addr[0], dev_addr);
1866 
1867 	mutex_lock(&adapter->fw_lock);
1868 	adapter->fw_done_rc = 0;
1869 	reinit_completion(&adapter->fw_done);
1870 
1871 	rc = ibmvnic_send_crq(adapter, &crq);
1872 	if (rc) {
1873 		rc = -EIO;
1874 		mutex_unlock(&adapter->fw_lock);
1875 		goto err;
1876 	}
1877 
1878 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1879 	/* netdev->dev_addr is changed in handle_change_mac_rsp function */
1880 	if (rc || adapter->fw_done_rc) {
1881 		rc = -EIO;
1882 		mutex_unlock(&adapter->fw_lock);
1883 		goto err;
1884 	}
1885 	mutex_unlock(&adapter->fw_lock);
1886 	return 0;
1887 err:
1888 	ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
1889 	return rc;
1890 }
1891 
1892 static int ibmvnic_set_mac(struct net_device *netdev, void *p)
1893 {
1894 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1895 	struct sockaddr *addr = p;
1896 	int rc;
1897 
1898 	rc = 0;
1899 	if (!is_valid_ether_addr(addr->sa_data))
1900 		return -EADDRNOTAVAIL;
1901 
1902 	ether_addr_copy(adapter->mac_addr, addr->sa_data);
1903 	if (adapter->state != VNIC_PROBED)
1904 		rc = __ibmvnic_set_mac(netdev, addr->sa_data);
1905 
1906 	return rc;
1907 }
1908 
1909 /*
1910  * do_reset returns zero if we are able to keep processing reset events, or
1911  * non-zero if we hit a fatal error and must halt.
1912  */
1913 static int do_reset(struct ibmvnic_adapter *adapter,
1914 		    struct ibmvnic_rwi *rwi, u32 reset_state)
1915 {
1916 	u64 old_num_rx_queues, old_num_tx_queues;
1917 	u64 old_num_rx_slots, old_num_tx_slots;
1918 	struct net_device *netdev = adapter->netdev;
1919 	int rc;
1920 
1921 	netdev_dbg(adapter->netdev,
1922 		   "[S:%d FOP:%d] Reset reason %d, reset_state %d\n",
1923 		   adapter->state, adapter->failover_pending,
1924 		   rwi->reset_reason, reset_state);
1925 
1926 	adapter->reset_reason = rwi->reset_reason;
1927 	/* requestor of VNIC_RESET_CHANGE_PARAM already has the rtnl lock */
1928 	if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
1929 		rtnl_lock();
1930 
1931 	/* Now that we have the rtnl lock, clear any pending failover.
1932 	 * This will ensure ibmvnic_open() has either completed or will
1933 	 * block until failover is complete.
1934 	 */
1935 	if (rwi->reset_reason == VNIC_RESET_FAILOVER)
1936 		adapter->failover_pending = false;
1937 
1938 	/* read the state and check (again) after getting rtnl */
1939 	reset_state = adapter->state;
1940 
1941 	if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
1942 		rc = -EBUSY;
1943 		goto out;
1944 	}
1945 
1946 	netif_carrier_off(netdev);
1947 
1948 	old_num_rx_queues = adapter->req_rx_queues;
1949 	old_num_tx_queues = adapter->req_tx_queues;
1950 	old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq;
1951 	old_num_tx_slots = adapter->req_tx_entries_per_subcrq;
1952 
1953 	ibmvnic_cleanup(netdev);
1954 
1955 	if (reset_state == VNIC_OPEN &&
1956 	    adapter->reset_reason != VNIC_RESET_MOBILITY &&
1957 	    adapter->reset_reason != VNIC_RESET_FAILOVER) {
1958 		if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
1959 			rc = __ibmvnic_close(netdev);
1960 			if (rc)
1961 				goto out;
1962 		} else {
1963 			adapter->state = VNIC_CLOSING;
1964 
1965 			/* Release the RTNL lock before link state change and
1966 			 * re-acquire after the link state change to allow
1967 			 * linkwatch_event to grab the RTNL lock and run during
1968 			 * a reset.
1969 			 */
1970 			rtnl_unlock();
1971 			rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
1972 			rtnl_lock();
1973 			if (rc)
1974 				goto out;
1975 
1976 			if (adapter->state == VNIC_OPEN) {
1977 				/* When we dropped rtnl, ibmvnic_open() got
1978 				 * it and noticed that we are resetting and
1979 				 * set the adapter state to OPEN. Update our
1980 				 * new "target" state, and resume the reset
1981 				 * from VNIC_CLOSING state.
1982 				 */
1983 				netdev_dbg(netdev,
1984 					   "Open changed state from %d, updating.\n",
1985 					   reset_state);
1986 				reset_state = VNIC_OPEN;
1987 				adapter->state = VNIC_CLOSING;
1988 			}
1989 
1990 			if (adapter->state != VNIC_CLOSING) {
1991 				/* If someone else changed the adapter state
1992 				 * when we dropped the rtnl, fail the reset
1993 				 */
1994 				rc = -1;
1995 				goto out;
1996 			}
1997 			adapter->state = VNIC_CLOSED;
1998 		}
1999 	}
2000 
2001 	if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2002 		release_resources(adapter);
2003 		release_sub_crqs(adapter, 1);
2004 		release_crq_queue(adapter);
2005 	}
2006 
2007 	if (adapter->reset_reason != VNIC_RESET_NON_FATAL) {
2008 		/* remove the closed state so when we call open it appears
2009 		 * we are coming from the probed state.
2010 		 */
2011 		adapter->state = VNIC_PROBED;
2012 
2013 		if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2014 			rc = init_crq_queue(adapter);
2015 		} else if (adapter->reset_reason == VNIC_RESET_MOBILITY) {
2016 			rc = ibmvnic_reenable_crq_queue(adapter);
2017 			release_sub_crqs(adapter, 1);
2018 		} else {
2019 			rc = ibmvnic_reset_crq(adapter);
2020 			if (rc == H_CLOSED || rc == H_SUCCESS) {
2021 				rc = vio_enable_interrupts(adapter->vdev);
2022 				if (rc)
2023 					netdev_err(adapter->netdev,
2024 						   "Reset failed to enable interrupts. rc=%d\n",
2025 						   rc);
2026 			}
2027 		}
2028 
2029 		if (rc) {
2030 			netdev_err(adapter->netdev,
2031 				   "Reset couldn't initialize crq. rc=%d\n", rc);
2032 			goto out;
2033 		}
2034 
2035 		rc = ibmvnic_reset_init(adapter, true);
2036 		if (rc) {
2037 			rc = IBMVNIC_INIT_FAILED;
2038 			goto out;
2039 		}
2040 
2041 		/* If the adapter was in PROBE state prior to the reset,
2042 		 * exit here.
2043 		 */
2044 		if (reset_state == VNIC_PROBED) {
2045 			rc = 0;
2046 			goto out;
2047 		}
2048 
2049 		rc = ibmvnic_login(netdev);
2050 		if (rc)
2051 			goto out;
2052 
2053 		if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2054 			rc = init_resources(adapter);
2055 			if (rc)
2056 				goto out;
2057 		} else if (adapter->req_rx_queues != old_num_rx_queues ||
2058 		    adapter->req_tx_queues != old_num_tx_queues ||
2059 		    adapter->req_rx_add_entries_per_subcrq !=
2060 		    old_num_rx_slots ||
2061 		    adapter->req_tx_entries_per_subcrq !=
2062 		    old_num_tx_slots ||
2063 		    !adapter->rx_pool ||
2064 		    !adapter->tso_pool ||
2065 		    !adapter->tx_pool) {
2066 			release_rx_pools(adapter);
2067 			release_tx_pools(adapter);
2068 			release_napi(adapter);
2069 			release_vpd_data(adapter);
2070 
2071 			rc = init_resources(adapter);
2072 			if (rc)
2073 				goto out;
2074 
2075 		} else {
2076 			rc = reset_tx_pools(adapter);
2077 			if (rc) {
2078 				netdev_dbg(adapter->netdev, "reset tx pools failed (%d)\n",
2079 					   rc);
2080 				goto out;
2081 			}
2082 
2083 			rc = reset_rx_pools(adapter);
2084 			if (rc) {
2085 				netdev_dbg(adapter->netdev, "reset rx pools failed (%d)\n",
2086 					   rc);
2087 				goto out;
2088 			}
2089 		}
2090 		ibmvnic_disable_irqs(adapter);
2091 	}
2092 	adapter->state = VNIC_CLOSED;
2093 
2094 	if (reset_state == VNIC_CLOSED) {
2095 		rc = 0;
2096 		goto out;
2097 	}
2098 
2099 	rc = __ibmvnic_open(netdev);
2100 	if (rc) {
2101 		rc = IBMVNIC_OPEN_FAILED;
2102 		goto out;
2103 	}
2104 
2105 	/* refresh device's multicast list */
2106 	ibmvnic_set_multi(netdev);
2107 
2108 	if (adapter->reset_reason == VNIC_RESET_FAILOVER ||
2109 	    adapter->reset_reason == VNIC_RESET_MOBILITY)
2110 		__netdev_notify_peers(netdev);
2111 
2112 	rc = 0;
2113 
2114 out:
2115 	/* restore the adapter state if reset failed */
2116 	if (rc)
2117 		adapter->state = reset_state;
2118 	/* requestor of VNIC_RESET_CHANGE_PARAM should still hold the rtnl lock */
2119 	if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
2120 		rtnl_unlock();
2121 
2122 	netdev_dbg(adapter->netdev, "[S:%d FOP:%d] Reset done, rc %d\n",
2123 		   adapter->state, adapter->failover_pending, rc);
2124 	return rc;
2125 }
2126 
2127 static int do_hard_reset(struct ibmvnic_adapter *adapter,
2128 			 struct ibmvnic_rwi *rwi, u32 reset_state)
2129 {
2130 	struct net_device *netdev = adapter->netdev;
2131 	int rc;
2132 
2133 	netdev_dbg(adapter->netdev, "Hard resetting driver (%d)\n",
2134 		   rwi->reset_reason);
2135 
2136 	/* read the state and check (again) after getting rtnl */
2137 	reset_state = adapter->state;
2138 
2139 	if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
2140 		rc = -EBUSY;
2141 		goto out;
2142 	}
2143 
2144 	netif_carrier_off(netdev);
2145 	adapter->reset_reason = rwi->reset_reason;
2146 
2147 	ibmvnic_cleanup(netdev);
2148 	release_resources(adapter);
2149 	release_sub_crqs(adapter, 0);
2150 	release_crq_queue(adapter);
2151 
2152 	/* remove the closed state so when we call open it appears
2153 	 * we are coming from the probed state.
2154 	 */
2155 	adapter->state = VNIC_PROBED;
2156 
2157 	reinit_completion(&adapter->init_done);
2158 	rc = init_crq_queue(adapter);
2159 	if (rc) {
2160 		netdev_err(adapter->netdev,
2161 			   "Couldn't initialize crq. rc=%d\n", rc);
2162 		goto out;
2163 	}
2164 
2165 	rc = ibmvnic_reset_init(adapter, false);
2166 	if (rc)
2167 		goto out;
2168 
2169 	/* If the adapter was in PROBE state prior to the reset,
2170 	 * exit here.
2171 	 */
2172 	if (reset_state == VNIC_PROBED)
2173 		goto out;
2174 
2175 	rc = ibmvnic_login(netdev);
2176 	if (rc)
2177 		goto out;
2178 
2179 	rc = init_resources(adapter);
2180 	if (rc)
2181 		goto out;
2182 
2183 	ibmvnic_disable_irqs(adapter);
2184 	adapter->state = VNIC_CLOSED;
2185 
2186 	if (reset_state == VNIC_CLOSED)
2187 		goto out;
2188 
2189 	rc = __ibmvnic_open(netdev);
2190 	if (rc) {
2191 		rc = IBMVNIC_OPEN_FAILED;
2192 		goto out;
2193 	}
2194 
2195 	__netdev_notify_peers(netdev);
2196 out:
2197 	/* restore adapter state if reset failed */
2198 	if (rc)
2199 		adapter->state = reset_state;
2200 	netdev_dbg(adapter->netdev, "[S:%d FOP:%d] Hard reset done, rc %d\n",
2201 		   adapter->state, adapter->failover_pending, rc);
2202 	return rc;
2203 }
2204 
2205 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter)
2206 {
2207 	struct ibmvnic_rwi *rwi;
2208 	unsigned long flags;
2209 
2210 	spin_lock_irqsave(&adapter->rwi_lock, flags);
2211 
2212 	if (!list_empty(&adapter->rwi_list)) {
2213 		rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi,
2214 				       list);
2215 		list_del(&rwi->list);
2216 	} else {
2217 		rwi = NULL;
2218 	}
2219 
2220 	spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2221 	return rwi;
2222 }
2223 
2224 static void __ibmvnic_reset(struct work_struct *work)
2225 {
2226 	struct ibmvnic_rwi *rwi;
2227 	struct ibmvnic_adapter *adapter;
2228 	bool saved_state = false;
2229 	unsigned long flags;
2230 	u32 reset_state;
2231 	int rc = 0;
2232 
2233 	adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset);
2234 
2235 	if (test_and_set_bit_lock(0, &adapter->resetting)) {
2236 		schedule_delayed_work(&adapter->ibmvnic_delayed_reset,
2237 				      IBMVNIC_RESET_DELAY);
2238 		return;
2239 	}
2240 
2241 	rwi = get_next_rwi(adapter);
2242 	while (rwi) {
2243 		spin_lock_irqsave(&adapter->state_lock, flags);
2244 
2245 		if (adapter->state == VNIC_REMOVING ||
2246 		    adapter->state == VNIC_REMOVED) {
2247 			spin_unlock_irqrestore(&adapter->state_lock, flags);
2248 			kfree(rwi);
2249 			rc = EBUSY;
2250 			break;
2251 		}
2252 
2253 		if (!saved_state) {
2254 			reset_state = adapter->state;
2255 			saved_state = true;
2256 		}
2257 		spin_unlock_irqrestore(&adapter->state_lock, flags);
2258 
2259 		if (adapter->force_reset_recovery) {
2260 			/* Since we are doing a hard reset now, clear the
2261 			 * failover_pending flag so we don't ignore any
2262 			 * future MOBILITY or other resets.
2263 			 */
2264 			adapter->failover_pending = false;
2265 
2266 			/* Transport event occurred during previous reset */
2267 			if (adapter->wait_for_reset) {
2268 				/* Previous was CHANGE_PARAM; caller locked */
2269 				adapter->force_reset_recovery = false;
2270 				rc = do_hard_reset(adapter, rwi, reset_state);
2271 			} else {
2272 				rtnl_lock();
2273 				adapter->force_reset_recovery = false;
2274 				rc = do_hard_reset(adapter, rwi, reset_state);
2275 				rtnl_unlock();
2276 			}
2277 			if (rc) {
2278 				/* give backing device time to settle down */
2279 				netdev_dbg(adapter->netdev,
2280 					   "[S:%d] Hard reset failed, waiting 60 secs\n",
2281 					   adapter->state);
2282 				set_current_state(TASK_UNINTERRUPTIBLE);
2283 				schedule_timeout(60 * HZ);
2284 			}
2285 		} else {
2286 			rc = do_reset(adapter, rwi, reset_state);
2287 		}
2288 		kfree(rwi);
2289 		adapter->last_reset_time = jiffies;
2290 
2291 		if (rc)
2292 			netdev_dbg(adapter->netdev, "Reset failed, rc=%d\n", rc);
2293 
2294 		rwi = get_next_rwi(adapter);
2295 
2296 		if (rwi && (rwi->reset_reason == VNIC_RESET_FAILOVER ||
2297 			    rwi->reset_reason == VNIC_RESET_MOBILITY))
2298 			adapter->force_reset_recovery = true;
2299 	}
2300 
2301 	if (adapter->wait_for_reset) {
2302 		adapter->reset_done_rc = rc;
2303 		complete(&adapter->reset_done);
2304 	}
2305 
2306 	clear_bit_unlock(0, &adapter->resetting);
2307 
2308 	netdev_dbg(adapter->netdev,
2309 		   "[S:%d FRR:%d WFR:%d] Done processing resets\n",
2310 		   adapter->state, adapter->force_reset_recovery,
2311 		   adapter->wait_for_reset);
2312 }
2313 
2314 static void __ibmvnic_delayed_reset(struct work_struct *work)
2315 {
2316 	struct ibmvnic_adapter *adapter;
2317 
2318 	adapter = container_of(work, struct ibmvnic_adapter,
2319 			       ibmvnic_delayed_reset.work);
2320 	__ibmvnic_reset(&adapter->ibmvnic_reset);
2321 }
2322 
2323 static int ibmvnic_reset(struct ibmvnic_adapter *adapter,
2324 			 enum ibmvnic_reset_reason reason)
2325 {
2326 	struct list_head *entry, *tmp_entry;
2327 	struct ibmvnic_rwi *rwi, *tmp;
2328 	struct net_device *netdev = adapter->netdev;
2329 	unsigned long flags;
2330 	int ret;
2331 
2332 	spin_lock_irqsave(&adapter->rwi_lock, flags);
2333 
2334 	/* If failover is pending don't schedule any other reset.
2335 	 * Instead let the failover complete. If there is already a
2336 	 * a failover reset scheduled, we will detect and drop the
2337 	 * duplicate reset when walking the ->rwi_list below.
2338 	 */
2339 	if (adapter->state == VNIC_REMOVING ||
2340 	    adapter->state == VNIC_REMOVED ||
2341 	    (adapter->failover_pending && reason != VNIC_RESET_FAILOVER)) {
2342 		ret = EBUSY;
2343 		netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n");
2344 		goto err;
2345 	}
2346 
2347 	if (adapter->state == VNIC_PROBING) {
2348 		netdev_warn(netdev, "Adapter reset during probe\n");
2349 		adapter->init_done_rc = EAGAIN;
2350 		ret = EAGAIN;
2351 		goto err;
2352 	}
2353 
2354 	list_for_each(entry, &adapter->rwi_list) {
2355 		tmp = list_entry(entry, struct ibmvnic_rwi, list);
2356 		if (tmp->reset_reason == reason) {
2357 			netdev_dbg(netdev, "Skipping matching reset, reason=%d\n",
2358 				   reason);
2359 			ret = EBUSY;
2360 			goto err;
2361 		}
2362 	}
2363 
2364 	rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC);
2365 	if (!rwi) {
2366 		ret = ENOMEM;
2367 		goto err;
2368 	}
2369 	/* if we just received a transport event,
2370 	 * flush reset queue and process this reset
2371 	 */
2372 	if (adapter->force_reset_recovery && !list_empty(&adapter->rwi_list)) {
2373 		list_for_each_safe(entry, tmp_entry, &adapter->rwi_list)
2374 			list_del(entry);
2375 	}
2376 	rwi->reset_reason = reason;
2377 	list_add_tail(&rwi->list, &adapter->rwi_list);
2378 	netdev_dbg(adapter->netdev, "Scheduling reset (reason %d)\n", reason);
2379 	schedule_work(&adapter->ibmvnic_reset);
2380 
2381 	ret = 0;
2382 err:
2383 	/* ibmvnic_close() below can block, so drop the lock first */
2384 	spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2385 
2386 	if (ret == ENOMEM)
2387 		ibmvnic_close(netdev);
2388 
2389 	return -ret;
2390 }
2391 
2392 static void ibmvnic_tx_timeout(struct net_device *dev, unsigned int txqueue)
2393 {
2394 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
2395 
2396 	if (test_bit(0, &adapter->resetting)) {
2397 		netdev_err(adapter->netdev,
2398 			   "Adapter is resetting, skip timeout reset\n");
2399 		return;
2400 	}
2401 	/* No queuing up reset until at least 5 seconds (default watchdog val)
2402 	 * after last reset
2403 	 */
2404 	if (time_before(jiffies, (adapter->last_reset_time + dev->watchdog_timeo))) {
2405 		netdev_dbg(dev, "Not yet time to tx timeout.\n");
2406 		return;
2407 	}
2408 	ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT);
2409 }
2410 
2411 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter,
2412 				  struct ibmvnic_rx_buff *rx_buff)
2413 {
2414 	struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index];
2415 
2416 	rx_buff->skb = NULL;
2417 
2418 	pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff);
2419 	pool->next_alloc = (pool->next_alloc + 1) % pool->size;
2420 
2421 	atomic_dec(&pool->available);
2422 }
2423 
2424 static int ibmvnic_poll(struct napi_struct *napi, int budget)
2425 {
2426 	struct ibmvnic_sub_crq_queue *rx_scrq;
2427 	struct ibmvnic_adapter *adapter;
2428 	struct net_device *netdev;
2429 	int frames_processed;
2430 	int scrq_num;
2431 
2432 	netdev = napi->dev;
2433 	adapter = netdev_priv(netdev);
2434 	scrq_num = (int)(napi - adapter->napi);
2435 	frames_processed = 0;
2436 	rx_scrq = adapter->rx_scrq[scrq_num];
2437 
2438 restart_poll:
2439 	while (frames_processed < budget) {
2440 		struct sk_buff *skb;
2441 		struct ibmvnic_rx_buff *rx_buff;
2442 		union sub_crq *next;
2443 		u32 length;
2444 		u16 offset;
2445 		u8 flags = 0;
2446 
2447 		if (unlikely(test_bit(0, &adapter->resetting) &&
2448 			     adapter->reset_reason != VNIC_RESET_NON_FATAL)) {
2449 			enable_scrq_irq(adapter, rx_scrq);
2450 			napi_complete_done(napi, frames_processed);
2451 			return frames_processed;
2452 		}
2453 
2454 		if (!pending_scrq(adapter, rx_scrq))
2455 			break;
2456 		next = ibmvnic_next_scrq(adapter, rx_scrq);
2457 		rx_buff = (struct ibmvnic_rx_buff *)
2458 			  be64_to_cpu(next->rx_comp.correlator);
2459 		/* do error checking */
2460 		if (next->rx_comp.rc) {
2461 			netdev_dbg(netdev, "rx buffer returned with rc %x\n",
2462 				   be16_to_cpu(next->rx_comp.rc));
2463 			/* free the entry */
2464 			next->rx_comp.first = 0;
2465 			dev_kfree_skb_any(rx_buff->skb);
2466 			remove_buff_from_pool(adapter, rx_buff);
2467 			continue;
2468 		} else if (!rx_buff->skb) {
2469 			/* free the entry */
2470 			next->rx_comp.first = 0;
2471 			remove_buff_from_pool(adapter, rx_buff);
2472 			continue;
2473 		}
2474 
2475 		length = be32_to_cpu(next->rx_comp.len);
2476 		offset = be16_to_cpu(next->rx_comp.off_frame_data);
2477 		flags = next->rx_comp.flags;
2478 		skb = rx_buff->skb;
2479 		/* load long_term_buff before copying to skb */
2480 		dma_rmb();
2481 		skb_copy_to_linear_data(skb, rx_buff->data + offset,
2482 					length);
2483 
2484 		/* VLAN Header has been stripped by the system firmware and
2485 		 * needs to be inserted by the driver
2486 		 */
2487 		if (adapter->rx_vlan_header_insertion &&
2488 		    (flags & IBMVNIC_VLAN_STRIPPED))
2489 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
2490 					       ntohs(next->rx_comp.vlan_tci));
2491 
2492 		/* free the entry */
2493 		next->rx_comp.first = 0;
2494 		remove_buff_from_pool(adapter, rx_buff);
2495 
2496 		skb_put(skb, length);
2497 		skb->protocol = eth_type_trans(skb, netdev);
2498 		skb_record_rx_queue(skb, scrq_num);
2499 
2500 		if (flags & IBMVNIC_IP_CHKSUM_GOOD &&
2501 		    flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) {
2502 			skb->ip_summed = CHECKSUM_UNNECESSARY;
2503 		}
2504 
2505 		length = skb->len;
2506 		napi_gro_receive(napi, skb); /* send it up */
2507 		netdev->stats.rx_packets++;
2508 		netdev->stats.rx_bytes += length;
2509 		adapter->rx_stats_buffers[scrq_num].packets++;
2510 		adapter->rx_stats_buffers[scrq_num].bytes += length;
2511 		frames_processed++;
2512 	}
2513 
2514 	if (adapter->state != VNIC_CLOSING &&
2515 	    ((atomic_read(&adapter->rx_pool[scrq_num].available) <
2516 	      adapter->req_rx_add_entries_per_subcrq / 2) ||
2517 	      frames_processed < budget))
2518 		replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]);
2519 	if (frames_processed < budget) {
2520 		if (napi_complete_done(napi, frames_processed)) {
2521 			enable_scrq_irq(adapter, rx_scrq);
2522 			if (pending_scrq(adapter, rx_scrq)) {
2523 				if (napi_reschedule(napi)) {
2524 					disable_scrq_irq(adapter, rx_scrq);
2525 					goto restart_poll;
2526 				}
2527 			}
2528 		}
2529 	}
2530 	return frames_processed;
2531 }
2532 
2533 static int wait_for_reset(struct ibmvnic_adapter *adapter)
2534 {
2535 	int rc, ret;
2536 
2537 	adapter->fallback.mtu = adapter->req_mtu;
2538 	adapter->fallback.rx_queues = adapter->req_rx_queues;
2539 	adapter->fallback.tx_queues = adapter->req_tx_queues;
2540 	adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq;
2541 	adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq;
2542 
2543 	reinit_completion(&adapter->reset_done);
2544 	adapter->wait_for_reset = true;
2545 	rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2546 
2547 	if (rc) {
2548 		ret = rc;
2549 		goto out;
2550 	}
2551 	rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 60000);
2552 	if (rc) {
2553 		ret = -ENODEV;
2554 		goto out;
2555 	}
2556 
2557 	ret = 0;
2558 	if (adapter->reset_done_rc) {
2559 		ret = -EIO;
2560 		adapter->desired.mtu = adapter->fallback.mtu;
2561 		adapter->desired.rx_queues = adapter->fallback.rx_queues;
2562 		adapter->desired.tx_queues = adapter->fallback.tx_queues;
2563 		adapter->desired.rx_entries = adapter->fallback.rx_entries;
2564 		adapter->desired.tx_entries = adapter->fallback.tx_entries;
2565 
2566 		reinit_completion(&adapter->reset_done);
2567 		adapter->wait_for_reset = true;
2568 		rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2569 		if (rc) {
2570 			ret = rc;
2571 			goto out;
2572 		}
2573 		rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done,
2574 						 60000);
2575 		if (rc) {
2576 			ret = -ENODEV;
2577 			goto out;
2578 		}
2579 	}
2580 out:
2581 	adapter->wait_for_reset = false;
2582 
2583 	return ret;
2584 }
2585 
2586 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu)
2587 {
2588 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2589 
2590 	adapter->desired.mtu = new_mtu + ETH_HLEN;
2591 
2592 	return wait_for_reset(adapter);
2593 }
2594 
2595 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb,
2596 						struct net_device *dev,
2597 						netdev_features_t features)
2598 {
2599 	/* Some backing hardware adapters can not
2600 	 * handle packets with a MSS less than 224
2601 	 * or with only one segment.
2602 	 */
2603 	if (skb_is_gso(skb)) {
2604 		if (skb_shinfo(skb)->gso_size < 224 ||
2605 		    skb_shinfo(skb)->gso_segs == 1)
2606 			features &= ~NETIF_F_GSO_MASK;
2607 	}
2608 
2609 	return features;
2610 }
2611 
2612 static const struct net_device_ops ibmvnic_netdev_ops = {
2613 	.ndo_open		= ibmvnic_open,
2614 	.ndo_stop		= ibmvnic_close,
2615 	.ndo_start_xmit		= ibmvnic_xmit,
2616 	.ndo_set_rx_mode	= ibmvnic_set_multi,
2617 	.ndo_set_mac_address	= ibmvnic_set_mac,
2618 	.ndo_validate_addr	= eth_validate_addr,
2619 	.ndo_tx_timeout		= ibmvnic_tx_timeout,
2620 	.ndo_change_mtu		= ibmvnic_change_mtu,
2621 	.ndo_features_check     = ibmvnic_features_check,
2622 };
2623 
2624 /* ethtool functions */
2625 
2626 static int ibmvnic_get_link_ksettings(struct net_device *netdev,
2627 				      struct ethtool_link_ksettings *cmd)
2628 {
2629 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2630 	int rc;
2631 
2632 	rc = send_query_phys_parms(adapter);
2633 	if (rc) {
2634 		adapter->speed = SPEED_UNKNOWN;
2635 		adapter->duplex = DUPLEX_UNKNOWN;
2636 	}
2637 	cmd->base.speed = adapter->speed;
2638 	cmd->base.duplex = adapter->duplex;
2639 	cmd->base.port = PORT_FIBRE;
2640 	cmd->base.phy_address = 0;
2641 	cmd->base.autoneg = AUTONEG_ENABLE;
2642 
2643 	return 0;
2644 }
2645 
2646 static void ibmvnic_get_drvinfo(struct net_device *netdev,
2647 				struct ethtool_drvinfo *info)
2648 {
2649 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2650 
2651 	strscpy(info->driver, ibmvnic_driver_name, sizeof(info->driver));
2652 	strscpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version));
2653 	strscpy(info->fw_version, adapter->fw_version,
2654 		sizeof(info->fw_version));
2655 }
2656 
2657 static u32 ibmvnic_get_msglevel(struct net_device *netdev)
2658 {
2659 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2660 
2661 	return adapter->msg_enable;
2662 }
2663 
2664 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data)
2665 {
2666 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2667 
2668 	adapter->msg_enable = data;
2669 }
2670 
2671 static u32 ibmvnic_get_link(struct net_device *netdev)
2672 {
2673 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2674 
2675 	/* Don't need to send a query because we request a logical link up at
2676 	 * init and then we wait for link state indications
2677 	 */
2678 	return adapter->logical_link_state;
2679 }
2680 
2681 static void ibmvnic_get_ringparam(struct net_device *netdev,
2682 				  struct ethtool_ringparam *ring)
2683 {
2684 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2685 
2686 	if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
2687 		ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq;
2688 		ring->tx_max_pending = adapter->max_tx_entries_per_subcrq;
2689 	} else {
2690 		ring->rx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
2691 		ring->tx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
2692 	}
2693 	ring->rx_mini_max_pending = 0;
2694 	ring->rx_jumbo_max_pending = 0;
2695 	ring->rx_pending = adapter->req_rx_add_entries_per_subcrq;
2696 	ring->tx_pending = adapter->req_tx_entries_per_subcrq;
2697 	ring->rx_mini_pending = 0;
2698 	ring->rx_jumbo_pending = 0;
2699 }
2700 
2701 static int ibmvnic_set_ringparam(struct net_device *netdev,
2702 				 struct ethtool_ringparam *ring)
2703 {
2704 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2705 	int ret;
2706 
2707 	ret = 0;
2708 	adapter->desired.rx_entries = ring->rx_pending;
2709 	adapter->desired.tx_entries = ring->tx_pending;
2710 
2711 	ret = wait_for_reset(adapter);
2712 
2713 	if (!ret &&
2714 	    (adapter->req_rx_add_entries_per_subcrq != ring->rx_pending ||
2715 	     adapter->req_tx_entries_per_subcrq != ring->tx_pending))
2716 		netdev_info(netdev,
2717 			    "Could not match full ringsize request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
2718 			    ring->rx_pending, ring->tx_pending,
2719 			    adapter->req_rx_add_entries_per_subcrq,
2720 			    adapter->req_tx_entries_per_subcrq);
2721 	return ret;
2722 }
2723 
2724 static void ibmvnic_get_channels(struct net_device *netdev,
2725 				 struct ethtool_channels *channels)
2726 {
2727 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2728 
2729 	if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
2730 		channels->max_rx = adapter->max_rx_queues;
2731 		channels->max_tx = adapter->max_tx_queues;
2732 	} else {
2733 		channels->max_rx = IBMVNIC_MAX_QUEUES;
2734 		channels->max_tx = IBMVNIC_MAX_QUEUES;
2735 	}
2736 
2737 	channels->max_other = 0;
2738 	channels->max_combined = 0;
2739 	channels->rx_count = adapter->req_rx_queues;
2740 	channels->tx_count = adapter->req_tx_queues;
2741 	channels->other_count = 0;
2742 	channels->combined_count = 0;
2743 }
2744 
2745 static int ibmvnic_set_channels(struct net_device *netdev,
2746 				struct ethtool_channels *channels)
2747 {
2748 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2749 	int ret;
2750 
2751 	ret = 0;
2752 	adapter->desired.rx_queues = channels->rx_count;
2753 	adapter->desired.tx_queues = channels->tx_count;
2754 
2755 	ret = wait_for_reset(adapter);
2756 
2757 	if (!ret &&
2758 	    (adapter->req_rx_queues != channels->rx_count ||
2759 	     adapter->req_tx_queues != channels->tx_count))
2760 		netdev_info(netdev,
2761 			    "Could not match full channels request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
2762 			    channels->rx_count, channels->tx_count,
2763 			    adapter->req_rx_queues, adapter->req_tx_queues);
2764 	return ret;
2765 }
2766 
2767 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data)
2768 {
2769 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
2770 	int i;
2771 
2772 	switch (stringset) {
2773 	case ETH_SS_STATS:
2774 		for (i = 0; i < ARRAY_SIZE(ibmvnic_stats);
2775 				i++, data += ETH_GSTRING_LEN)
2776 			memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN);
2777 
2778 		for (i = 0; i < adapter->req_tx_queues; i++) {
2779 			snprintf(data, ETH_GSTRING_LEN, "tx%d_packets", i);
2780 			data += ETH_GSTRING_LEN;
2781 
2782 			snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i);
2783 			data += ETH_GSTRING_LEN;
2784 
2785 			snprintf(data, ETH_GSTRING_LEN,
2786 				 "tx%d_dropped_packets", i);
2787 			data += ETH_GSTRING_LEN;
2788 		}
2789 
2790 		for (i = 0; i < adapter->req_rx_queues; i++) {
2791 			snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i);
2792 			data += ETH_GSTRING_LEN;
2793 
2794 			snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i);
2795 			data += ETH_GSTRING_LEN;
2796 
2797 			snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i);
2798 			data += ETH_GSTRING_LEN;
2799 		}
2800 		break;
2801 
2802 	case ETH_SS_PRIV_FLAGS:
2803 		for (i = 0; i < ARRAY_SIZE(ibmvnic_priv_flags); i++)
2804 			strcpy(data + i * ETH_GSTRING_LEN,
2805 			       ibmvnic_priv_flags[i]);
2806 		break;
2807 	default:
2808 		return;
2809 	}
2810 }
2811 
2812 static int ibmvnic_get_sset_count(struct net_device *dev, int sset)
2813 {
2814 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
2815 
2816 	switch (sset) {
2817 	case ETH_SS_STATS:
2818 		return ARRAY_SIZE(ibmvnic_stats) +
2819 		       adapter->req_tx_queues * NUM_TX_STATS +
2820 		       adapter->req_rx_queues * NUM_RX_STATS;
2821 	case ETH_SS_PRIV_FLAGS:
2822 		return ARRAY_SIZE(ibmvnic_priv_flags);
2823 	default:
2824 		return -EOPNOTSUPP;
2825 	}
2826 }
2827 
2828 static void ibmvnic_get_ethtool_stats(struct net_device *dev,
2829 				      struct ethtool_stats *stats, u64 *data)
2830 {
2831 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
2832 	union ibmvnic_crq crq;
2833 	int i, j;
2834 	int rc;
2835 
2836 	memset(&crq, 0, sizeof(crq));
2837 	crq.request_statistics.first = IBMVNIC_CRQ_CMD;
2838 	crq.request_statistics.cmd = REQUEST_STATISTICS;
2839 	crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token);
2840 	crq.request_statistics.len =
2841 	    cpu_to_be32(sizeof(struct ibmvnic_statistics));
2842 
2843 	/* Wait for data to be written */
2844 	reinit_completion(&adapter->stats_done);
2845 	rc = ibmvnic_send_crq(adapter, &crq);
2846 	if (rc)
2847 		return;
2848 	rc = ibmvnic_wait_for_completion(adapter, &adapter->stats_done, 10000);
2849 	if (rc)
2850 		return;
2851 
2852 	for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++)
2853 		data[i] = be64_to_cpu(IBMVNIC_GET_STAT
2854 				      (adapter, ibmvnic_stats[i].offset));
2855 
2856 	for (j = 0; j < adapter->req_tx_queues; j++) {
2857 		data[i] = adapter->tx_stats_buffers[j].packets;
2858 		i++;
2859 		data[i] = adapter->tx_stats_buffers[j].bytes;
2860 		i++;
2861 		data[i] = adapter->tx_stats_buffers[j].dropped_packets;
2862 		i++;
2863 	}
2864 
2865 	for (j = 0; j < adapter->req_rx_queues; j++) {
2866 		data[i] = adapter->rx_stats_buffers[j].packets;
2867 		i++;
2868 		data[i] = adapter->rx_stats_buffers[j].bytes;
2869 		i++;
2870 		data[i] = adapter->rx_stats_buffers[j].interrupts;
2871 		i++;
2872 	}
2873 }
2874 
2875 static u32 ibmvnic_get_priv_flags(struct net_device *netdev)
2876 {
2877 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2878 
2879 	return adapter->priv_flags;
2880 }
2881 
2882 static int ibmvnic_set_priv_flags(struct net_device *netdev, u32 flags)
2883 {
2884 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2885 	bool which_maxes = !!(flags & IBMVNIC_USE_SERVER_MAXES);
2886 
2887 	if (which_maxes)
2888 		adapter->priv_flags |= IBMVNIC_USE_SERVER_MAXES;
2889 	else
2890 		adapter->priv_flags &= ~IBMVNIC_USE_SERVER_MAXES;
2891 
2892 	return 0;
2893 }
2894 
2895 static const struct ethtool_ops ibmvnic_ethtool_ops = {
2896 	.get_drvinfo		= ibmvnic_get_drvinfo,
2897 	.get_msglevel		= ibmvnic_get_msglevel,
2898 	.set_msglevel		= ibmvnic_set_msglevel,
2899 	.get_link		= ibmvnic_get_link,
2900 	.get_ringparam		= ibmvnic_get_ringparam,
2901 	.set_ringparam		= ibmvnic_set_ringparam,
2902 	.get_channels		= ibmvnic_get_channels,
2903 	.set_channels		= ibmvnic_set_channels,
2904 	.get_strings            = ibmvnic_get_strings,
2905 	.get_sset_count         = ibmvnic_get_sset_count,
2906 	.get_ethtool_stats	= ibmvnic_get_ethtool_stats,
2907 	.get_link_ksettings	= ibmvnic_get_link_ksettings,
2908 	.get_priv_flags		= ibmvnic_get_priv_flags,
2909 	.set_priv_flags		= ibmvnic_set_priv_flags,
2910 };
2911 
2912 /* Routines for managing CRQs/sCRQs  */
2913 
2914 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter,
2915 				   struct ibmvnic_sub_crq_queue *scrq)
2916 {
2917 	int rc;
2918 
2919 	if (!scrq) {
2920 		netdev_dbg(adapter->netdev, "Invalid scrq reset.\n");
2921 		return -EINVAL;
2922 	}
2923 
2924 	if (scrq->irq) {
2925 		free_irq(scrq->irq, scrq);
2926 		irq_dispose_mapping(scrq->irq);
2927 		scrq->irq = 0;
2928 	}
2929 
2930 	if (scrq->msgs) {
2931 		memset(scrq->msgs, 0, 4 * PAGE_SIZE);
2932 		atomic_set(&scrq->used, 0);
2933 		scrq->cur = 0;
2934 		scrq->ind_buf.index = 0;
2935 	} else {
2936 		netdev_dbg(adapter->netdev, "Invalid scrq reset\n");
2937 		return -EINVAL;
2938 	}
2939 
2940 	rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
2941 			   4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
2942 	return rc;
2943 }
2944 
2945 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter)
2946 {
2947 	int i, rc;
2948 
2949 	if (!adapter->tx_scrq || !adapter->rx_scrq)
2950 		return -EINVAL;
2951 
2952 	for (i = 0; i < adapter->req_tx_queues; i++) {
2953 		netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i);
2954 		rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]);
2955 		if (rc)
2956 			return rc;
2957 	}
2958 
2959 	for (i = 0; i < adapter->req_rx_queues; i++) {
2960 		netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i);
2961 		rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]);
2962 		if (rc)
2963 			return rc;
2964 	}
2965 
2966 	return rc;
2967 }
2968 
2969 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter,
2970 				  struct ibmvnic_sub_crq_queue *scrq,
2971 				  bool do_h_free)
2972 {
2973 	struct device *dev = &adapter->vdev->dev;
2974 	long rc;
2975 
2976 	netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n");
2977 
2978 	if (do_h_free) {
2979 		/* Close the sub-crqs */
2980 		do {
2981 			rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
2982 						adapter->vdev->unit_address,
2983 						scrq->crq_num);
2984 		} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
2985 
2986 		if (rc) {
2987 			netdev_err(adapter->netdev,
2988 				   "Failed to release sub-CRQ %16lx, rc = %ld\n",
2989 				   scrq->crq_num, rc);
2990 		}
2991 	}
2992 
2993 	dma_free_coherent(dev,
2994 			  IBMVNIC_IND_ARR_SZ,
2995 			  scrq->ind_buf.indir_arr,
2996 			  scrq->ind_buf.indir_dma);
2997 
2998 	dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
2999 			 DMA_BIDIRECTIONAL);
3000 	free_pages((unsigned long)scrq->msgs, 2);
3001 	kfree(scrq);
3002 }
3003 
3004 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter
3005 							*adapter)
3006 {
3007 	struct device *dev = &adapter->vdev->dev;
3008 	struct ibmvnic_sub_crq_queue *scrq;
3009 	int rc;
3010 
3011 	scrq = kzalloc(sizeof(*scrq), GFP_KERNEL);
3012 	if (!scrq)
3013 		return NULL;
3014 
3015 	scrq->msgs =
3016 		(union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2);
3017 	if (!scrq->msgs) {
3018 		dev_warn(dev, "Couldn't allocate crq queue messages page\n");
3019 		goto zero_page_failed;
3020 	}
3021 
3022 	scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE,
3023 					 DMA_BIDIRECTIONAL);
3024 	if (dma_mapping_error(dev, scrq->msg_token)) {
3025 		dev_warn(dev, "Couldn't map crq queue messages page\n");
3026 		goto map_failed;
3027 	}
3028 
3029 	rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
3030 			   4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
3031 
3032 	if (rc == H_RESOURCE)
3033 		rc = ibmvnic_reset_crq(adapter);
3034 
3035 	if (rc == H_CLOSED) {
3036 		dev_warn(dev, "Partner adapter not ready, waiting.\n");
3037 	} else if (rc) {
3038 		dev_warn(dev, "Error %d registering sub-crq\n", rc);
3039 		goto reg_failed;
3040 	}
3041 
3042 	scrq->adapter = adapter;
3043 	scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs);
3044 	scrq->ind_buf.index = 0;
3045 
3046 	scrq->ind_buf.indir_arr =
3047 		dma_alloc_coherent(dev,
3048 				   IBMVNIC_IND_ARR_SZ,
3049 				   &scrq->ind_buf.indir_dma,
3050 				   GFP_KERNEL);
3051 
3052 	if (!scrq->ind_buf.indir_arr)
3053 		goto indir_failed;
3054 
3055 	spin_lock_init(&scrq->lock);
3056 
3057 	netdev_dbg(adapter->netdev,
3058 		   "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n",
3059 		   scrq->crq_num, scrq->hw_irq, scrq->irq);
3060 
3061 	return scrq;
3062 
3063 indir_failed:
3064 	do {
3065 		rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3066 					adapter->vdev->unit_address,
3067 					scrq->crq_num);
3068 	} while (rc == H_BUSY || rc == H_IS_LONG_BUSY(rc));
3069 reg_failed:
3070 	dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3071 			 DMA_BIDIRECTIONAL);
3072 map_failed:
3073 	free_pages((unsigned long)scrq->msgs, 2);
3074 zero_page_failed:
3075 	kfree(scrq);
3076 
3077 	return NULL;
3078 }
3079 
3080 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free)
3081 {
3082 	int i;
3083 
3084 	if (adapter->tx_scrq) {
3085 		for (i = 0; i < adapter->num_active_tx_scrqs; i++) {
3086 			if (!adapter->tx_scrq[i])
3087 				continue;
3088 
3089 			netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n",
3090 				   i);
3091 			if (adapter->tx_scrq[i]->irq) {
3092 				free_irq(adapter->tx_scrq[i]->irq,
3093 					 adapter->tx_scrq[i]);
3094 				irq_dispose_mapping(adapter->tx_scrq[i]->irq);
3095 				adapter->tx_scrq[i]->irq = 0;
3096 			}
3097 
3098 			release_sub_crq_queue(adapter, adapter->tx_scrq[i],
3099 					      do_h_free);
3100 		}
3101 
3102 		kfree(adapter->tx_scrq);
3103 		adapter->tx_scrq = NULL;
3104 		adapter->num_active_tx_scrqs = 0;
3105 	}
3106 
3107 	if (adapter->rx_scrq) {
3108 		for (i = 0; i < adapter->num_active_rx_scrqs; i++) {
3109 			if (!adapter->rx_scrq[i])
3110 				continue;
3111 
3112 			netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n",
3113 				   i);
3114 			if (adapter->rx_scrq[i]->irq) {
3115 				free_irq(adapter->rx_scrq[i]->irq,
3116 					 adapter->rx_scrq[i]);
3117 				irq_dispose_mapping(adapter->rx_scrq[i]->irq);
3118 				adapter->rx_scrq[i]->irq = 0;
3119 			}
3120 
3121 			release_sub_crq_queue(adapter, adapter->rx_scrq[i],
3122 					      do_h_free);
3123 		}
3124 
3125 		kfree(adapter->rx_scrq);
3126 		adapter->rx_scrq = NULL;
3127 		adapter->num_active_rx_scrqs = 0;
3128 	}
3129 }
3130 
3131 static int disable_scrq_irq(struct ibmvnic_adapter *adapter,
3132 			    struct ibmvnic_sub_crq_queue *scrq)
3133 {
3134 	struct device *dev = &adapter->vdev->dev;
3135 	unsigned long rc;
3136 
3137 	rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3138 				H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3139 	if (rc)
3140 		dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n",
3141 			scrq->hw_irq, rc);
3142 	return rc;
3143 }
3144 
3145 static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
3146 			   struct ibmvnic_sub_crq_queue *scrq)
3147 {
3148 	struct device *dev = &adapter->vdev->dev;
3149 	unsigned long rc;
3150 
3151 	if (scrq->hw_irq > 0x100000000ULL) {
3152 		dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq);
3153 		return 1;
3154 	}
3155 
3156 	if (test_bit(0, &adapter->resetting) &&
3157 	    adapter->reset_reason == VNIC_RESET_MOBILITY) {
3158 		u64 val = (0xff000000) | scrq->hw_irq;
3159 
3160 		rc = plpar_hcall_norets(H_EOI, val);
3161 		/* H_EOI would fail with rc = H_FUNCTION when running
3162 		 * in XIVE mode which is expected, but not an error.
3163 		 */
3164 		if (rc && rc != H_FUNCTION)
3165 			dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n",
3166 				val, rc);
3167 	}
3168 
3169 	rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3170 				H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3171 	if (rc)
3172 		dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n",
3173 			scrq->hw_irq, rc);
3174 	return rc;
3175 }
3176 
3177 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter,
3178 			       struct ibmvnic_sub_crq_queue *scrq)
3179 {
3180 	struct device *dev = &adapter->vdev->dev;
3181 	struct ibmvnic_tx_pool *tx_pool;
3182 	struct ibmvnic_tx_buff *txbuff;
3183 	struct netdev_queue *txq;
3184 	union sub_crq *next;
3185 	int index;
3186 	int i;
3187 
3188 restart_loop:
3189 	while (pending_scrq(adapter, scrq)) {
3190 		unsigned int pool = scrq->pool_index;
3191 		int num_entries = 0;
3192 		int total_bytes = 0;
3193 		int num_packets = 0;
3194 
3195 		next = ibmvnic_next_scrq(adapter, scrq);
3196 		for (i = 0; i < next->tx_comp.num_comps; i++) {
3197 			index = be32_to_cpu(next->tx_comp.correlators[i]);
3198 			if (index & IBMVNIC_TSO_POOL_MASK) {
3199 				tx_pool = &adapter->tso_pool[pool];
3200 				index &= ~IBMVNIC_TSO_POOL_MASK;
3201 			} else {
3202 				tx_pool = &adapter->tx_pool[pool];
3203 			}
3204 
3205 			txbuff = &tx_pool->tx_buff[index];
3206 			num_packets++;
3207 			num_entries += txbuff->num_entries;
3208 			if (txbuff->skb) {
3209 				total_bytes += txbuff->skb->len;
3210 				if (next->tx_comp.rcs[i]) {
3211 					dev_err(dev, "tx error %x\n",
3212 						next->tx_comp.rcs[i]);
3213 					dev_kfree_skb_irq(txbuff->skb);
3214 				} else {
3215 					dev_consume_skb_irq(txbuff->skb);
3216 				}
3217 				txbuff->skb = NULL;
3218 			} else {
3219 				netdev_warn(adapter->netdev,
3220 					    "TX completion received with NULL socket buffer\n");
3221 			}
3222 			tx_pool->free_map[tx_pool->producer_index] = index;
3223 			tx_pool->producer_index =
3224 				(tx_pool->producer_index + 1) %
3225 					tx_pool->num_buffers;
3226 		}
3227 		/* remove tx_comp scrq*/
3228 		next->tx_comp.first = 0;
3229 
3230 		txq = netdev_get_tx_queue(adapter->netdev, scrq->pool_index);
3231 		netdev_tx_completed_queue(txq, num_packets, total_bytes);
3232 
3233 		if (atomic_sub_return(num_entries, &scrq->used) <=
3234 		    (adapter->req_tx_entries_per_subcrq / 2) &&
3235 		    __netif_subqueue_stopped(adapter->netdev,
3236 					     scrq->pool_index)) {
3237 			netif_wake_subqueue(adapter->netdev, scrq->pool_index);
3238 			netdev_dbg(adapter->netdev, "Started queue %d\n",
3239 				   scrq->pool_index);
3240 		}
3241 	}
3242 
3243 	enable_scrq_irq(adapter, scrq);
3244 
3245 	if (pending_scrq(adapter, scrq)) {
3246 		disable_scrq_irq(adapter, scrq);
3247 		goto restart_loop;
3248 	}
3249 
3250 	return 0;
3251 }
3252 
3253 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance)
3254 {
3255 	struct ibmvnic_sub_crq_queue *scrq = instance;
3256 	struct ibmvnic_adapter *adapter = scrq->adapter;
3257 
3258 	disable_scrq_irq(adapter, scrq);
3259 	ibmvnic_complete_tx(adapter, scrq);
3260 
3261 	return IRQ_HANDLED;
3262 }
3263 
3264 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance)
3265 {
3266 	struct ibmvnic_sub_crq_queue *scrq = instance;
3267 	struct ibmvnic_adapter *adapter = scrq->adapter;
3268 
3269 	/* When booting a kdump kernel we can hit pending interrupts
3270 	 * prior to completing driver initialization.
3271 	 */
3272 	if (unlikely(adapter->state != VNIC_OPEN))
3273 		return IRQ_NONE;
3274 
3275 	adapter->rx_stats_buffers[scrq->scrq_num].interrupts++;
3276 
3277 	if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) {
3278 		disable_scrq_irq(adapter, scrq);
3279 		__napi_schedule(&adapter->napi[scrq->scrq_num]);
3280 	}
3281 
3282 	return IRQ_HANDLED;
3283 }
3284 
3285 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter)
3286 {
3287 	struct device *dev = &adapter->vdev->dev;
3288 	struct ibmvnic_sub_crq_queue *scrq;
3289 	int i = 0, j = 0;
3290 	int rc = 0;
3291 
3292 	for (i = 0; i < adapter->req_tx_queues; i++) {
3293 		netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n",
3294 			   i);
3295 		scrq = adapter->tx_scrq[i];
3296 		scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3297 
3298 		if (!scrq->irq) {
3299 			rc = -EINVAL;
3300 			dev_err(dev, "Error mapping irq\n");
3301 			goto req_tx_irq_failed;
3302 		}
3303 
3304 		snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-tx%d",
3305 			 adapter->vdev->unit_address, i);
3306 		rc = request_irq(scrq->irq, ibmvnic_interrupt_tx,
3307 				 0, scrq->name, scrq);
3308 
3309 		if (rc) {
3310 			dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n",
3311 				scrq->irq, rc);
3312 			irq_dispose_mapping(scrq->irq);
3313 			goto req_tx_irq_failed;
3314 		}
3315 	}
3316 
3317 	for (i = 0; i < adapter->req_rx_queues; i++) {
3318 		netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n",
3319 			   i);
3320 		scrq = adapter->rx_scrq[i];
3321 		scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3322 		if (!scrq->irq) {
3323 			rc = -EINVAL;
3324 			dev_err(dev, "Error mapping irq\n");
3325 			goto req_rx_irq_failed;
3326 		}
3327 		snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-rx%d",
3328 			 adapter->vdev->unit_address, i);
3329 		rc = request_irq(scrq->irq, ibmvnic_interrupt_rx,
3330 				 0, scrq->name, scrq);
3331 		if (rc) {
3332 			dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n",
3333 				scrq->irq, rc);
3334 			irq_dispose_mapping(scrq->irq);
3335 			goto req_rx_irq_failed;
3336 		}
3337 	}
3338 	return rc;
3339 
3340 req_rx_irq_failed:
3341 	for (j = 0; j < i; j++) {
3342 		free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]);
3343 		irq_dispose_mapping(adapter->rx_scrq[j]->irq);
3344 	}
3345 	i = adapter->req_tx_queues;
3346 req_tx_irq_failed:
3347 	for (j = 0; j < i; j++) {
3348 		free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]);
3349 		irq_dispose_mapping(adapter->tx_scrq[j]->irq);
3350 	}
3351 	release_sub_crqs(adapter, 1);
3352 	return rc;
3353 }
3354 
3355 static int init_sub_crqs(struct ibmvnic_adapter *adapter)
3356 {
3357 	struct device *dev = &adapter->vdev->dev;
3358 	struct ibmvnic_sub_crq_queue **allqueues;
3359 	int registered_queues = 0;
3360 	int total_queues;
3361 	int more = 0;
3362 	int i;
3363 
3364 	total_queues = adapter->req_tx_queues + adapter->req_rx_queues;
3365 
3366 	allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL);
3367 	if (!allqueues)
3368 		return -1;
3369 
3370 	for (i = 0; i < total_queues; i++) {
3371 		allqueues[i] = init_sub_crq_queue(adapter);
3372 		if (!allqueues[i]) {
3373 			dev_warn(dev, "Couldn't allocate all sub-crqs\n");
3374 			break;
3375 		}
3376 		registered_queues++;
3377 	}
3378 
3379 	/* Make sure we were able to register the minimum number of queues */
3380 	if (registered_queues <
3381 	    adapter->min_tx_queues + adapter->min_rx_queues) {
3382 		dev_err(dev, "Fatal: Couldn't init  min number of sub-crqs\n");
3383 		goto tx_failed;
3384 	}
3385 
3386 	/* Distribute the failed allocated queues*/
3387 	for (i = 0; i < total_queues - registered_queues + more ; i++) {
3388 		netdev_dbg(adapter->netdev, "Reducing number of queues\n");
3389 		switch (i % 3) {
3390 		case 0:
3391 			if (adapter->req_rx_queues > adapter->min_rx_queues)
3392 				adapter->req_rx_queues--;
3393 			else
3394 				more++;
3395 			break;
3396 		case 1:
3397 			if (adapter->req_tx_queues > adapter->min_tx_queues)
3398 				adapter->req_tx_queues--;
3399 			else
3400 				more++;
3401 			break;
3402 		}
3403 	}
3404 
3405 	adapter->tx_scrq = kcalloc(adapter->req_tx_queues,
3406 				   sizeof(*adapter->tx_scrq), GFP_KERNEL);
3407 	if (!adapter->tx_scrq)
3408 		goto tx_failed;
3409 
3410 	for (i = 0; i < adapter->req_tx_queues; i++) {
3411 		adapter->tx_scrq[i] = allqueues[i];
3412 		adapter->tx_scrq[i]->pool_index = i;
3413 		adapter->num_active_tx_scrqs++;
3414 	}
3415 
3416 	adapter->rx_scrq = kcalloc(adapter->req_rx_queues,
3417 				   sizeof(*adapter->rx_scrq), GFP_KERNEL);
3418 	if (!adapter->rx_scrq)
3419 		goto rx_failed;
3420 
3421 	for (i = 0; i < adapter->req_rx_queues; i++) {
3422 		adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues];
3423 		adapter->rx_scrq[i]->scrq_num = i;
3424 		adapter->num_active_rx_scrqs++;
3425 	}
3426 
3427 	kfree(allqueues);
3428 	return 0;
3429 
3430 rx_failed:
3431 	kfree(adapter->tx_scrq);
3432 	adapter->tx_scrq = NULL;
3433 tx_failed:
3434 	for (i = 0; i < registered_queues; i++)
3435 		release_sub_crq_queue(adapter, allqueues[i], 1);
3436 	kfree(allqueues);
3437 	return -1;
3438 }
3439 
3440 static void send_request_cap(struct ibmvnic_adapter *adapter, int retry)
3441 {
3442 	struct device *dev = &adapter->vdev->dev;
3443 	union ibmvnic_crq crq;
3444 	int max_entries;
3445 
3446 	if (!retry) {
3447 		/* Sub-CRQ entries are 32 byte long */
3448 		int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4);
3449 
3450 		if (adapter->min_tx_entries_per_subcrq > entries_page ||
3451 		    adapter->min_rx_add_entries_per_subcrq > entries_page) {
3452 			dev_err(dev, "Fatal, invalid entries per sub-crq\n");
3453 			return;
3454 		}
3455 
3456 		if (adapter->desired.mtu)
3457 			adapter->req_mtu = adapter->desired.mtu;
3458 		else
3459 			adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN;
3460 
3461 		if (!adapter->desired.tx_entries)
3462 			adapter->desired.tx_entries =
3463 					adapter->max_tx_entries_per_subcrq;
3464 		if (!adapter->desired.rx_entries)
3465 			adapter->desired.rx_entries =
3466 					adapter->max_rx_add_entries_per_subcrq;
3467 
3468 		max_entries = IBMVNIC_MAX_LTB_SIZE /
3469 			      (adapter->req_mtu + IBMVNIC_BUFFER_HLEN);
3470 
3471 		if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3472 			adapter->desired.tx_entries > IBMVNIC_MAX_LTB_SIZE) {
3473 			adapter->desired.tx_entries = max_entries;
3474 		}
3475 
3476 		if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3477 			adapter->desired.rx_entries > IBMVNIC_MAX_LTB_SIZE) {
3478 			adapter->desired.rx_entries = max_entries;
3479 		}
3480 
3481 		if (adapter->desired.tx_entries)
3482 			adapter->req_tx_entries_per_subcrq =
3483 					adapter->desired.tx_entries;
3484 		else
3485 			adapter->req_tx_entries_per_subcrq =
3486 					adapter->max_tx_entries_per_subcrq;
3487 
3488 		if (adapter->desired.rx_entries)
3489 			adapter->req_rx_add_entries_per_subcrq =
3490 					adapter->desired.rx_entries;
3491 		else
3492 			adapter->req_rx_add_entries_per_subcrq =
3493 					adapter->max_rx_add_entries_per_subcrq;
3494 
3495 		if (adapter->desired.tx_queues)
3496 			adapter->req_tx_queues =
3497 					adapter->desired.tx_queues;
3498 		else
3499 			adapter->req_tx_queues =
3500 					adapter->opt_tx_comp_sub_queues;
3501 
3502 		if (adapter->desired.rx_queues)
3503 			adapter->req_rx_queues =
3504 					adapter->desired.rx_queues;
3505 		else
3506 			adapter->req_rx_queues =
3507 					adapter->opt_rx_comp_queues;
3508 
3509 		adapter->req_rx_add_queues = adapter->max_rx_add_queues;
3510 	}
3511 
3512 	memset(&crq, 0, sizeof(crq));
3513 	crq.request_capability.first = IBMVNIC_CRQ_CMD;
3514 	crq.request_capability.cmd = REQUEST_CAPABILITY;
3515 
3516 	crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES);
3517 	crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues);
3518 	atomic_inc(&adapter->running_cap_crqs);
3519 	ibmvnic_send_crq(adapter, &crq);
3520 
3521 	crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES);
3522 	crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues);
3523 	atomic_inc(&adapter->running_cap_crqs);
3524 	ibmvnic_send_crq(adapter, &crq);
3525 
3526 	crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES);
3527 	crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues);
3528 	atomic_inc(&adapter->running_cap_crqs);
3529 	ibmvnic_send_crq(adapter, &crq);
3530 
3531 	crq.request_capability.capability =
3532 	    cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ);
3533 	crq.request_capability.number =
3534 	    cpu_to_be64(adapter->req_tx_entries_per_subcrq);
3535 	atomic_inc(&adapter->running_cap_crqs);
3536 	ibmvnic_send_crq(adapter, &crq);
3537 
3538 	crq.request_capability.capability =
3539 	    cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ);
3540 	crq.request_capability.number =
3541 	    cpu_to_be64(adapter->req_rx_add_entries_per_subcrq);
3542 	atomic_inc(&adapter->running_cap_crqs);
3543 	ibmvnic_send_crq(adapter, &crq);
3544 
3545 	crq.request_capability.capability = cpu_to_be16(REQ_MTU);
3546 	crq.request_capability.number = cpu_to_be64(adapter->req_mtu);
3547 	atomic_inc(&adapter->running_cap_crqs);
3548 	ibmvnic_send_crq(adapter, &crq);
3549 
3550 	if (adapter->netdev->flags & IFF_PROMISC) {
3551 		if (adapter->promisc_supported) {
3552 			crq.request_capability.capability =
3553 			    cpu_to_be16(PROMISC_REQUESTED);
3554 			crq.request_capability.number = cpu_to_be64(1);
3555 			atomic_inc(&adapter->running_cap_crqs);
3556 			ibmvnic_send_crq(adapter, &crq);
3557 		}
3558 	} else {
3559 		crq.request_capability.capability =
3560 		    cpu_to_be16(PROMISC_REQUESTED);
3561 		crq.request_capability.number = cpu_to_be64(0);
3562 		atomic_inc(&adapter->running_cap_crqs);
3563 		ibmvnic_send_crq(adapter, &crq);
3564 	}
3565 }
3566 
3567 static int pending_scrq(struct ibmvnic_adapter *adapter,
3568 			struct ibmvnic_sub_crq_queue *scrq)
3569 {
3570 	union sub_crq *entry = &scrq->msgs[scrq->cur];
3571 	int rc;
3572 
3573 	rc = !!(entry->generic.first & IBMVNIC_CRQ_CMD_RSP);
3574 
3575 	/* Ensure that the SCRQ valid flag is loaded prior to loading the
3576 	 * contents of the SCRQ descriptor
3577 	 */
3578 	dma_rmb();
3579 
3580 	return rc;
3581 }
3582 
3583 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter,
3584 					struct ibmvnic_sub_crq_queue *scrq)
3585 {
3586 	union sub_crq *entry;
3587 	unsigned long flags;
3588 
3589 	spin_lock_irqsave(&scrq->lock, flags);
3590 	entry = &scrq->msgs[scrq->cur];
3591 	if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) {
3592 		if (++scrq->cur == scrq->size)
3593 			scrq->cur = 0;
3594 	} else {
3595 		entry = NULL;
3596 	}
3597 	spin_unlock_irqrestore(&scrq->lock, flags);
3598 
3599 	/* Ensure that the SCRQ valid flag is loaded prior to loading the
3600 	 * contents of the SCRQ descriptor
3601 	 */
3602 	dma_rmb();
3603 
3604 	return entry;
3605 }
3606 
3607 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter)
3608 {
3609 	struct ibmvnic_crq_queue *queue = &adapter->crq;
3610 	union ibmvnic_crq *crq;
3611 
3612 	crq = &queue->msgs[queue->cur];
3613 	if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) {
3614 		if (++queue->cur == queue->size)
3615 			queue->cur = 0;
3616 	} else {
3617 		crq = NULL;
3618 	}
3619 
3620 	return crq;
3621 }
3622 
3623 static void print_subcrq_error(struct device *dev, int rc, const char *func)
3624 {
3625 	switch (rc) {
3626 	case H_PARAMETER:
3627 		dev_warn_ratelimited(dev,
3628 				     "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n",
3629 				     func, rc);
3630 		break;
3631 	case H_CLOSED:
3632 		dev_warn_ratelimited(dev,
3633 				     "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n",
3634 				     func, rc);
3635 		break;
3636 	default:
3637 		dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc);
3638 		break;
3639 	}
3640 }
3641 
3642 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter,
3643 				u64 remote_handle, u64 ioba, u64 num_entries)
3644 {
3645 	unsigned int ua = adapter->vdev->unit_address;
3646 	struct device *dev = &adapter->vdev->dev;
3647 	int rc;
3648 
3649 	/* Make sure the hypervisor sees the complete request */
3650 	dma_wmb();
3651 	rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua,
3652 				cpu_to_be64(remote_handle),
3653 				ioba, num_entries);
3654 
3655 	if (rc)
3656 		print_subcrq_error(dev, rc, __func__);
3657 
3658 	return rc;
3659 }
3660 
3661 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter,
3662 			    union ibmvnic_crq *crq)
3663 {
3664 	unsigned int ua = adapter->vdev->unit_address;
3665 	struct device *dev = &adapter->vdev->dev;
3666 	u64 *u64_crq = (u64 *)crq;
3667 	int rc;
3668 
3669 	netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n",
3670 		   (unsigned long)cpu_to_be64(u64_crq[0]),
3671 		   (unsigned long)cpu_to_be64(u64_crq[1]));
3672 
3673 	if (!adapter->crq.active &&
3674 	    crq->generic.first != IBMVNIC_CRQ_INIT_CMD) {
3675 		dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n");
3676 		return -EINVAL;
3677 	}
3678 
3679 	/* Make sure the hypervisor sees the complete request */
3680 	dma_wmb();
3681 
3682 	rc = plpar_hcall_norets(H_SEND_CRQ, ua,
3683 				cpu_to_be64(u64_crq[0]),
3684 				cpu_to_be64(u64_crq[1]));
3685 
3686 	if (rc) {
3687 		if (rc == H_CLOSED) {
3688 			dev_warn(dev, "CRQ Queue closed\n");
3689 			/* do not reset, report the fail, wait for passive init from server */
3690 		}
3691 
3692 		dev_warn(dev, "Send error (rc=%d)\n", rc);
3693 	}
3694 
3695 	return rc;
3696 }
3697 
3698 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter)
3699 {
3700 	struct device *dev = &adapter->vdev->dev;
3701 	union ibmvnic_crq crq;
3702 	int retries = 100;
3703 	int rc;
3704 
3705 	memset(&crq, 0, sizeof(crq));
3706 	crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
3707 	crq.generic.cmd = IBMVNIC_CRQ_INIT;
3708 	netdev_dbg(adapter->netdev, "Sending CRQ init\n");
3709 
3710 	do {
3711 		rc = ibmvnic_send_crq(adapter, &crq);
3712 		if (rc != H_CLOSED)
3713 			break;
3714 		retries--;
3715 		msleep(50);
3716 
3717 	} while (retries > 0);
3718 
3719 	if (rc) {
3720 		dev_err(dev, "Failed to send init request, rc = %d\n", rc);
3721 		return rc;
3722 	}
3723 
3724 	return 0;
3725 }
3726 
3727 static int send_version_xchg(struct ibmvnic_adapter *adapter)
3728 {
3729 	union ibmvnic_crq crq;
3730 
3731 	memset(&crq, 0, sizeof(crq));
3732 	crq.version_exchange.first = IBMVNIC_CRQ_CMD;
3733 	crq.version_exchange.cmd = VERSION_EXCHANGE;
3734 	crq.version_exchange.version = cpu_to_be16(ibmvnic_version);
3735 
3736 	return ibmvnic_send_crq(adapter, &crq);
3737 }
3738 
3739 struct vnic_login_client_data {
3740 	u8	type;
3741 	__be16	len;
3742 	char	name[];
3743 } __packed;
3744 
3745 static int vnic_client_data_len(struct ibmvnic_adapter *adapter)
3746 {
3747 	int len;
3748 
3749 	/* Calculate the amount of buffer space needed for the
3750 	 * vnic client data in the login buffer. There are four entries,
3751 	 * OS name, LPAR name, device name, and a null last entry.
3752 	 */
3753 	len = 4 * sizeof(struct vnic_login_client_data);
3754 	len += 6; /* "Linux" plus NULL */
3755 	len += strlen(utsname()->nodename) + 1;
3756 	len += strlen(adapter->netdev->name) + 1;
3757 
3758 	return len;
3759 }
3760 
3761 static void vnic_add_client_data(struct ibmvnic_adapter *adapter,
3762 				 struct vnic_login_client_data *vlcd)
3763 {
3764 	const char *os_name = "Linux";
3765 	int len;
3766 
3767 	/* Type 1 - LPAR OS */
3768 	vlcd->type = 1;
3769 	len = strlen(os_name) + 1;
3770 	vlcd->len = cpu_to_be16(len);
3771 	strncpy(vlcd->name, os_name, len);
3772 	vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
3773 
3774 	/* Type 2 - LPAR name */
3775 	vlcd->type = 2;
3776 	len = strlen(utsname()->nodename) + 1;
3777 	vlcd->len = cpu_to_be16(len);
3778 	strncpy(vlcd->name, utsname()->nodename, len);
3779 	vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
3780 
3781 	/* Type 3 - device name */
3782 	vlcd->type = 3;
3783 	len = strlen(adapter->netdev->name) + 1;
3784 	vlcd->len = cpu_to_be16(len);
3785 	strncpy(vlcd->name, adapter->netdev->name, len);
3786 }
3787 
3788 static int send_login(struct ibmvnic_adapter *adapter)
3789 {
3790 	struct ibmvnic_login_rsp_buffer *login_rsp_buffer;
3791 	struct ibmvnic_login_buffer *login_buffer;
3792 	struct device *dev = &adapter->vdev->dev;
3793 	struct vnic_login_client_data *vlcd;
3794 	dma_addr_t rsp_buffer_token;
3795 	dma_addr_t buffer_token;
3796 	size_t rsp_buffer_size;
3797 	union ibmvnic_crq crq;
3798 	int client_data_len;
3799 	size_t buffer_size;
3800 	__be64 *tx_list_p;
3801 	__be64 *rx_list_p;
3802 	int rc;
3803 	int i;
3804 
3805 	if (!adapter->tx_scrq || !adapter->rx_scrq) {
3806 		netdev_err(adapter->netdev,
3807 			   "RX or TX queues are not allocated, device login failed\n");
3808 		return -1;
3809 	}
3810 
3811 	release_login_buffer(adapter);
3812 	release_login_rsp_buffer(adapter);
3813 
3814 	client_data_len = vnic_client_data_len(adapter);
3815 
3816 	buffer_size =
3817 	    sizeof(struct ibmvnic_login_buffer) +
3818 	    sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) +
3819 	    client_data_len;
3820 
3821 	login_buffer = kzalloc(buffer_size, GFP_ATOMIC);
3822 	if (!login_buffer)
3823 		goto buf_alloc_failed;
3824 
3825 	buffer_token = dma_map_single(dev, login_buffer, buffer_size,
3826 				      DMA_TO_DEVICE);
3827 	if (dma_mapping_error(dev, buffer_token)) {
3828 		dev_err(dev, "Couldn't map login buffer\n");
3829 		goto buf_map_failed;
3830 	}
3831 
3832 	rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) +
3833 			  sizeof(u64) * adapter->req_tx_queues +
3834 			  sizeof(u64) * adapter->req_rx_queues +
3835 			  sizeof(u64) * adapter->req_rx_queues +
3836 			  sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS;
3837 
3838 	login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC);
3839 	if (!login_rsp_buffer)
3840 		goto buf_rsp_alloc_failed;
3841 
3842 	rsp_buffer_token = dma_map_single(dev, login_rsp_buffer,
3843 					  rsp_buffer_size, DMA_FROM_DEVICE);
3844 	if (dma_mapping_error(dev, rsp_buffer_token)) {
3845 		dev_err(dev, "Couldn't map login rsp buffer\n");
3846 		goto buf_rsp_map_failed;
3847 	}
3848 
3849 	adapter->login_buf = login_buffer;
3850 	adapter->login_buf_token = buffer_token;
3851 	adapter->login_buf_sz = buffer_size;
3852 	adapter->login_rsp_buf = login_rsp_buffer;
3853 	adapter->login_rsp_buf_token = rsp_buffer_token;
3854 	adapter->login_rsp_buf_sz = rsp_buffer_size;
3855 
3856 	login_buffer->len = cpu_to_be32(buffer_size);
3857 	login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB);
3858 	login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues);
3859 	login_buffer->off_txcomp_subcrqs =
3860 	    cpu_to_be32(sizeof(struct ibmvnic_login_buffer));
3861 	login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues);
3862 	login_buffer->off_rxcomp_subcrqs =
3863 	    cpu_to_be32(sizeof(struct ibmvnic_login_buffer) +
3864 			sizeof(u64) * adapter->req_tx_queues);
3865 	login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token);
3866 	login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size);
3867 
3868 	tx_list_p = (__be64 *)((char *)login_buffer +
3869 				      sizeof(struct ibmvnic_login_buffer));
3870 	rx_list_p = (__be64 *)((char *)login_buffer +
3871 				      sizeof(struct ibmvnic_login_buffer) +
3872 				      sizeof(u64) * adapter->req_tx_queues);
3873 
3874 	for (i = 0; i < adapter->req_tx_queues; i++) {
3875 		if (adapter->tx_scrq[i]) {
3876 			tx_list_p[i] =
3877 				cpu_to_be64(adapter->tx_scrq[i]->crq_num);
3878 		}
3879 	}
3880 
3881 	for (i = 0; i < adapter->req_rx_queues; i++) {
3882 		if (adapter->rx_scrq[i]) {
3883 			rx_list_p[i] =
3884 				cpu_to_be64(adapter->rx_scrq[i]->crq_num);
3885 		}
3886 	}
3887 
3888 	/* Insert vNIC login client data */
3889 	vlcd = (struct vnic_login_client_data *)
3890 		((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues));
3891 	login_buffer->client_data_offset =
3892 			cpu_to_be32((char *)vlcd - (char *)login_buffer);
3893 	login_buffer->client_data_len = cpu_to_be32(client_data_len);
3894 
3895 	vnic_add_client_data(adapter, vlcd);
3896 
3897 	netdev_dbg(adapter->netdev, "Login Buffer:\n");
3898 	for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) {
3899 		netdev_dbg(adapter->netdev, "%016lx\n",
3900 			   ((unsigned long *)(adapter->login_buf))[i]);
3901 	}
3902 
3903 	memset(&crq, 0, sizeof(crq));
3904 	crq.login.first = IBMVNIC_CRQ_CMD;
3905 	crq.login.cmd = LOGIN;
3906 	crq.login.ioba = cpu_to_be32(buffer_token);
3907 	crq.login.len = cpu_to_be32(buffer_size);
3908 
3909 	adapter->login_pending = true;
3910 	rc = ibmvnic_send_crq(adapter, &crq);
3911 	if (rc) {
3912 		adapter->login_pending = false;
3913 		netdev_err(adapter->netdev, "Failed to send login, rc=%d\n", rc);
3914 		goto buf_rsp_map_failed;
3915 	}
3916 
3917 	return 0;
3918 
3919 buf_rsp_map_failed:
3920 	kfree(login_rsp_buffer);
3921 	adapter->login_rsp_buf = NULL;
3922 buf_rsp_alloc_failed:
3923 	dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE);
3924 buf_map_failed:
3925 	kfree(login_buffer);
3926 	adapter->login_buf = NULL;
3927 buf_alloc_failed:
3928 	return -1;
3929 }
3930 
3931 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr,
3932 			    u32 len, u8 map_id)
3933 {
3934 	union ibmvnic_crq crq;
3935 
3936 	memset(&crq, 0, sizeof(crq));
3937 	crq.request_map.first = IBMVNIC_CRQ_CMD;
3938 	crq.request_map.cmd = REQUEST_MAP;
3939 	crq.request_map.map_id = map_id;
3940 	crq.request_map.ioba = cpu_to_be32(addr);
3941 	crq.request_map.len = cpu_to_be32(len);
3942 	return ibmvnic_send_crq(adapter, &crq);
3943 }
3944 
3945 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id)
3946 {
3947 	union ibmvnic_crq crq;
3948 
3949 	memset(&crq, 0, sizeof(crq));
3950 	crq.request_unmap.first = IBMVNIC_CRQ_CMD;
3951 	crq.request_unmap.cmd = REQUEST_UNMAP;
3952 	crq.request_unmap.map_id = map_id;
3953 	return ibmvnic_send_crq(adapter, &crq);
3954 }
3955 
3956 static void send_query_map(struct ibmvnic_adapter *adapter)
3957 {
3958 	union ibmvnic_crq crq;
3959 
3960 	memset(&crq, 0, sizeof(crq));
3961 	crq.query_map.first = IBMVNIC_CRQ_CMD;
3962 	crq.query_map.cmd = QUERY_MAP;
3963 	ibmvnic_send_crq(adapter, &crq);
3964 }
3965 
3966 /* Send a series of CRQs requesting various capabilities of the VNIC server */
3967 static void send_query_cap(struct ibmvnic_adapter *adapter)
3968 {
3969 	union ibmvnic_crq crq;
3970 
3971 	atomic_set(&adapter->running_cap_crqs, 0);
3972 	memset(&crq, 0, sizeof(crq));
3973 	crq.query_capability.first = IBMVNIC_CRQ_CMD;
3974 	crq.query_capability.cmd = QUERY_CAPABILITY;
3975 
3976 	crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES);
3977 	atomic_inc(&adapter->running_cap_crqs);
3978 	ibmvnic_send_crq(adapter, &crq);
3979 
3980 	crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES);
3981 	atomic_inc(&adapter->running_cap_crqs);
3982 	ibmvnic_send_crq(adapter, &crq);
3983 
3984 	crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES);
3985 	atomic_inc(&adapter->running_cap_crqs);
3986 	ibmvnic_send_crq(adapter, &crq);
3987 
3988 	crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES);
3989 	atomic_inc(&adapter->running_cap_crqs);
3990 	ibmvnic_send_crq(adapter, &crq);
3991 
3992 	crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES);
3993 	atomic_inc(&adapter->running_cap_crqs);
3994 	ibmvnic_send_crq(adapter, &crq);
3995 
3996 	crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES);
3997 	atomic_inc(&adapter->running_cap_crqs);
3998 	ibmvnic_send_crq(adapter, &crq);
3999 
4000 	crq.query_capability.capability =
4001 	    cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ);
4002 	atomic_inc(&adapter->running_cap_crqs);
4003 	ibmvnic_send_crq(adapter, &crq);
4004 
4005 	crq.query_capability.capability =
4006 	    cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ);
4007 	atomic_inc(&adapter->running_cap_crqs);
4008 	ibmvnic_send_crq(adapter, &crq);
4009 
4010 	crq.query_capability.capability =
4011 	    cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ);
4012 	atomic_inc(&adapter->running_cap_crqs);
4013 	ibmvnic_send_crq(adapter, &crq);
4014 
4015 	crq.query_capability.capability =
4016 	    cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ);
4017 	atomic_inc(&adapter->running_cap_crqs);
4018 	ibmvnic_send_crq(adapter, &crq);
4019 
4020 	crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD);
4021 	atomic_inc(&adapter->running_cap_crqs);
4022 	ibmvnic_send_crq(adapter, &crq);
4023 
4024 	crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED);
4025 	atomic_inc(&adapter->running_cap_crqs);
4026 	ibmvnic_send_crq(adapter, &crq);
4027 
4028 	crq.query_capability.capability = cpu_to_be16(MIN_MTU);
4029 	atomic_inc(&adapter->running_cap_crqs);
4030 	ibmvnic_send_crq(adapter, &crq);
4031 
4032 	crq.query_capability.capability = cpu_to_be16(MAX_MTU);
4033 	atomic_inc(&adapter->running_cap_crqs);
4034 	ibmvnic_send_crq(adapter, &crq);
4035 
4036 	crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS);
4037 	atomic_inc(&adapter->running_cap_crqs);
4038 	ibmvnic_send_crq(adapter, &crq);
4039 
4040 	crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION);
4041 	atomic_inc(&adapter->running_cap_crqs);
4042 	ibmvnic_send_crq(adapter, &crq);
4043 
4044 	crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION);
4045 	atomic_inc(&adapter->running_cap_crqs);
4046 	ibmvnic_send_crq(adapter, &crq);
4047 
4048 	crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES);
4049 	atomic_inc(&adapter->running_cap_crqs);
4050 	ibmvnic_send_crq(adapter, &crq);
4051 
4052 	crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED);
4053 	atomic_inc(&adapter->running_cap_crqs);
4054 	ibmvnic_send_crq(adapter, &crq);
4055 
4056 	crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES);
4057 	atomic_inc(&adapter->running_cap_crqs);
4058 	ibmvnic_send_crq(adapter, &crq);
4059 
4060 	crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES);
4061 	atomic_inc(&adapter->running_cap_crqs);
4062 	ibmvnic_send_crq(adapter, &crq);
4063 
4064 	crq.query_capability.capability =
4065 			cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q);
4066 	atomic_inc(&adapter->running_cap_crqs);
4067 	ibmvnic_send_crq(adapter, &crq);
4068 
4069 	crq.query_capability.capability =
4070 			cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ);
4071 	atomic_inc(&adapter->running_cap_crqs);
4072 	ibmvnic_send_crq(adapter, &crq);
4073 
4074 	crq.query_capability.capability =
4075 			cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ);
4076 	atomic_inc(&adapter->running_cap_crqs);
4077 	ibmvnic_send_crq(adapter, &crq);
4078 
4079 	crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ);
4080 	atomic_inc(&adapter->running_cap_crqs);
4081 	ibmvnic_send_crq(adapter, &crq);
4082 }
4083 
4084 static void send_query_ip_offload(struct ibmvnic_adapter *adapter)
4085 {
4086 	int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer);
4087 	struct device *dev = &adapter->vdev->dev;
4088 	union ibmvnic_crq crq;
4089 
4090 	adapter->ip_offload_tok =
4091 		dma_map_single(dev,
4092 			       &adapter->ip_offload_buf,
4093 			       buf_sz,
4094 			       DMA_FROM_DEVICE);
4095 
4096 	if (dma_mapping_error(dev, adapter->ip_offload_tok)) {
4097 		if (!firmware_has_feature(FW_FEATURE_CMO))
4098 			dev_err(dev, "Couldn't map offload buffer\n");
4099 		return;
4100 	}
4101 
4102 	memset(&crq, 0, sizeof(crq));
4103 	crq.query_ip_offload.first = IBMVNIC_CRQ_CMD;
4104 	crq.query_ip_offload.cmd = QUERY_IP_OFFLOAD;
4105 	crq.query_ip_offload.len = cpu_to_be32(buf_sz);
4106 	crq.query_ip_offload.ioba =
4107 	    cpu_to_be32(adapter->ip_offload_tok);
4108 
4109 	ibmvnic_send_crq(adapter, &crq);
4110 }
4111 
4112 static void send_control_ip_offload(struct ibmvnic_adapter *adapter)
4113 {
4114 	struct ibmvnic_control_ip_offload_buffer *ctrl_buf = &adapter->ip_offload_ctrl;
4115 	struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4116 	struct device *dev = &adapter->vdev->dev;
4117 	netdev_features_t old_hw_features = 0;
4118 	union ibmvnic_crq crq;
4119 
4120 	adapter->ip_offload_ctrl_tok =
4121 		dma_map_single(dev,
4122 			       ctrl_buf,
4123 			       sizeof(adapter->ip_offload_ctrl),
4124 			       DMA_TO_DEVICE);
4125 
4126 	if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) {
4127 		dev_err(dev, "Couldn't map ip offload control buffer\n");
4128 		return;
4129 	}
4130 
4131 	ctrl_buf->len = cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4132 	ctrl_buf->version = cpu_to_be32(INITIAL_VERSION_IOB);
4133 	ctrl_buf->ipv4_chksum = buf->ipv4_chksum;
4134 	ctrl_buf->ipv6_chksum = buf->ipv6_chksum;
4135 	ctrl_buf->tcp_ipv4_chksum = buf->tcp_ipv4_chksum;
4136 	ctrl_buf->udp_ipv4_chksum = buf->udp_ipv4_chksum;
4137 	ctrl_buf->tcp_ipv6_chksum = buf->tcp_ipv6_chksum;
4138 	ctrl_buf->udp_ipv6_chksum = buf->udp_ipv6_chksum;
4139 	ctrl_buf->large_tx_ipv4 = buf->large_tx_ipv4;
4140 	ctrl_buf->large_tx_ipv6 = buf->large_tx_ipv6;
4141 
4142 	/* large_rx disabled for now, additional features needed */
4143 	ctrl_buf->large_rx_ipv4 = 0;
4144 	ctrl_buf->large_rx_ipv6 = 0;
4145 
4146 	if (adapter->state != VNIC_PROBING) {
4147 		old_hw_features = adapter->netdev->hw_features;
4148 		adapter->netdev->hw_features = 0;
4149 	}
4150 
4151 	adapter->netdev->hw_features = NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
4152 
4153 	if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum)
4154 		adapter->netdev->hw_features |= NETIF_F_IP_CSUM;
4155 
4156 	if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum)
4157 		adapter->netdev->hw_features |= NETIF_F_IPV6_CSUM;
4158 
4159 	if ((adapter->netdev->features &
4160 	    (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
4161 		adapter->netdev->hw_features |= NETIF_F_RXCSUM;
4162 
4163 	if (buf->large_tx_ipv4)
4164 		adapter->netdev->hw_features |= NETIF_F_TSO;
4165 	if (buf->large_tx_ipv6)
4166 		adapter->netdev->hw_features |= NETIF_F_TSO6;
4167 
4168 	if (adapter->state == VNIC_PROBING) {
4169 		adapter->netdev->features |= adapter->netdev->hw_features;
4170 	} else if (old_hw_features != adapter->netdev->hw_features) {
4171 		netdev_features_t tmp = 0;
4172 
4173 		/* disable features no longer supported */
4174 		adapter->netdev->features &= adapter->netdev->hw_features;
4175 		/* turn on features now supported if previously enabled */
4176 		tmp = (old_hw_features ^ adapter->netdev->hw_features) &
4177 			adapter->netdev->hw_features;
4178 		adapter->netdev->features |=
4179 				tmp & adapter->netdev->wanted_features;
4180 	}
4181 
4182 	memset(&crq, 0, sizeof(crq));
4183 	crq.control_ip_offload.first = IBMVNIC_CRQ_CMD;
4184 	crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD;
4185 	crq.control_ip_offload.len =
4186 	    cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4187 	crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok);
4188 	ibmvnic_send_crq(adapter, &crq);
4189 }
4190 
4191 static void handle_vpd_size_rsp(union ibmvnic_crq *crq,
4192 				struct ibmvnic_adapter *adapter)
4193 {
4194 	struct device *dev = &adapter->vdev->dev;
4195 
4196 	if (crq->get_vpd_size_rsp.rc.code) {
4197 		dev_err(dev, "Error retrieving VPD size, rc=%x\n",
4198 			crq->get_vpd_size_rsp.rc.code);
4199 		complete(&adapter->fw_done);
4200 		return;
4201 	}
4202 
4203 	adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len);
4204 	complete(&adapter->fw_done);
4205 }
4206 
4207 static void handle_vpd_rsp(union ibmvnic_crq *crq,
4208 			   struct ibmvnic_adapter *adapter)
4209 {
4210 	struct device *dev = &adapter->vdev->dev;
4211 	unsigned char *substr = NULL;
4212 	u8 fw_level_len = 0;
4213 
4214 	memset(adapter->fw_version, 0, 32);
4215 
4216 	dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len,
4217 			 DMA_FROM_DEVICE);
4218 
4219 	if (crq->get_vpd_rsp.rc.code) {
4220 		dev_err(dev, "Error retrieving VPD from device, rc=%x\n",
4221 			crq->get_vpd_rsp.rc.code);
4222 		goto complete;
4223 	}
4224 
4225 	/* get the position of the firmware version info
4226 	 * located after the ASCII 'RM' substring in the buffer
4227 	 */
4228 	substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len);
4229 	if (!substr) {
4230 		dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n");
4231 		goto complete;
4232 	}
4233 
4234 	/* get length of firmware level ASCII substring */
4235 	if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) {
4236 		fw_level_len = *(substr + 2);
4237 	} else {
4238 		dev_info(dev, "Length of FW substr extrapolated VDP buff\n");
4239 		goto complete;
4240 	}
4241 
4242 	/* copy firmware version string from vpd into adapter */
4243 	if ((substr + 3 + fw_level_len) <
4244 	    (adapter->vpd->buff + adapter->vpd->len)) {
4245 		strncpy((char *)adapter->fw_version, substr + 3, fw_level_len);
4246 	} else {
4247 		dev_info(dev, "FW substr extrapolated VPD buff\n");
4248 	}
4249 
4250 complete:
4251 	if (adapter->fw_version[0] == '\0')
4252 		strncpy((char *)adapter->fw_version, "N/A", 3 * sizeof(char));
4253 	complete(&adapter->fw_done);
4254 }
4255 
4256 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter)
4257 {
4258 	struct device *dev = &adapter->vdev->dev;
4259 	struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4260 	int i;
4261 
4262 	dma_unmap_single(dev, adapter->ip_offload_tok,
4263 			 sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE);
4264 
4265 	netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n");
4266 	for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++)
4267 		netdev_dbg(adapter->netdev, "%016lx\n",
4268 			   ((unsigned long *)(buf))[i]);
4269 
4270 	netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum);
4271 	netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum);
4272 	netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n",
4273 		   buf->tcp_ipv4_chksum);
4274 	netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n",
4275 		   buf->tcp_ipv6_chksum);
4276 	netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n",
4277 		   buf->udp_ipv4_chksum);
4278 	netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n",
4279 		   buf->udp_ipv6_chksum);
4280 	netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n",
4281 		   buf->large_tx_ipv4);
4282 	netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n",
4283 		   buf->large_tx_ipv6);
4284 	netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n",
4285 		   buf->large_rx_ipv4);
4286 	netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n",
4287 		   buf->large_rx_ipv6);
4288 	netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n",
4289 		   buf->max_ipv4_header_size);
4290 	netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n",
4291 		   buf->max_ipv6_header_size);
4292 	netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n",
4293 		   buf->max_tcp_header_size);
4294 	netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n",
4295 		   buf->max_udp_header_size);
4296 	netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n",
4297 		   buf->max_large_tx_size);
4298 	netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n",
4299 		   buf->max_large_rx_size);
4300 	netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n",
4301 		   buf->ipv6_extension_header);
4302 	netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n",
4303 		   buf->tcp_pseudosum_req);
4304 	netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n",
4305 		   buf->num_ipv6_ext_headers);
4306 	netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n",
4307 		   buf->off_ipv6_ext_headers);
4308 
4309 	send_control_ip_offload(adapter);
4310 }
4311 
4312 static const char *ibmvnic_fw_err_cause(u16 cause)
4313 {
4314 	switch (cause) {
4315 	case ADAPTER_PROBLEM:
4316 		return "adapter problem";
4317 	case BUS_PROBLEM:
4318 		return "bus problem";
4319 	case FW_PROBLEM:
4320 		return "firmware problem";
4321 	case DD_PROBLEM:
4322 		return "device driver problem";
4323 	case EEH_RECOVERY:
4324 		return "EEH recovery";
4325 	case FW_UPDATED:
4326 		return "firmware updated";
4327 	case LOW_MEMORY:
4328 		return "low Memory";
4329 	default:
4330 		return "unknown";
4331 	}
4332 }
4333 
4334 static void handle_error_indication(union ibmvnic_crq *crq,
4335 				    struct ibmvnic_adapter *adapter)
4336 {
4337 	struct device *dev = &adapter->vdev->dev;
4338 	u16 cause;
4339 
4340 	cause = be16_to_cpu(crq->error_indication.error_cause);
4341 
4342 	dev_warn_ratelimited(dev,
4343 			     "Firmware reports %serror, cause: %s. Starting recovery...\n",
4344 			     crq->error_indication.flags
4345 				& IBMVNIC_FATAL_ERROR ? "FATAL " : "",
4346 			     ibmvnic_fw_err_cause(cause));
4347 
4348 	if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR)
4349 		ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4350 	else
4351 		ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL);
4352 }
4353 
4354 static int handle_change_mac_rsp(union ibmvnic_crq *crq,
4355 				 struct ibmvnic_adapter *adapter)
4356 {
4357 	struct net_device *netdev = adapter->netdev;
4358 	struct device *dev = &adapter->vdev->dev;
4359 	long rc;
4360 
4361 	rc = crq->change_mac_addr_rsp.rc.code;
4362 	if (rc) {
4363 		dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc);
4364 		goto out;
4365 	}
4366 	/* crq->change_mac_addr.mac_addr is the requested one
4367 	 * crq->change_mac_addr_rsp.mac_addr is the returned valid one.
4368 	 */
4369 	ether_addr_copy(netdev->dev_addr,
4370 			&crq->change_mac_addr_rsp.mac_addr[0]);
4371 	ether_addr_copy(adapter->mac_addr,
4372 			&crq->change_mac_addr_rsp.mac_addr[0]);
4373 out:
4374 	complete(&adapter->fw_done);
4375 	return rc;
4376 }
4377 
4378 static void handle_request_cap_rsp(union ibmvnic_crq *crq,
4379 				   struct ibmvnic_adapter *adapter)
4380 {
4381 	struct device *dev = &adapter->vdev->dev;
4382 	u64 *req_value;
4383 	char *name;
4384 
4385 	atomic_dec(&adapter->running_cap_crqs);
4386 	switch (be16_to_cpu(crq->request_capability_rsp.capability)) {
4387 	case REQ_TX_QUEUES:
4388 		req_value = &adapter->req_tx_queues;
4389 		name = "tx";
4390 		break;
4391 	case REQ_RX_QUEUES:
4392 		req_value = &adapter->req_rx_queues;
4393 		name = "rx";
4394 		break;
4395 	case REQ_RX_ADD_QUEUES:
4396 		req_value = &adapter->req_rx_add_queues;
4397 		name = "rx_add";
4398 		break;
4399 	case REQ_TX_ENTRIES_PER_SUBCRQ:
4400 		req_value = &adapter->req_tx_entries_per_subcrq;
4401 		name = "tx_entries_per_subcrq";
4402 		break;
4403 	case REQ_RX_ADD_ENTRIES_PER_SUBCRQ:
4404 		req_value = &adapter->req_rx_add_entries_per_subcrq;
4405 		name = "rx_add_entries_per_subcrq";
4406 		break;
4407 	case REQ_MTU:
4408 		req_value = &adapter->req_mtu;
4409 		name = "mtu";
4410 		break;
4411 	case PROMISC_REQUESTED:
4412 		req_value = &adapter->promisc;
4413 		name = "promisc";
4414 		break;
4415 	default:
4416 		dev_err(dev, "Got invalid cap request rsp %d\n",
4417 			crq->request_capability.capability);
4418 		return;
4419 	}
4420 
4421 	switch (crq->request_capability_rsp.rc.code) {
4422 	case SUCCESS:
4423 		break;
4424 	case PARTIALSUCCESS:
4425 		dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n",
4426 			 *req_value,
4427 			 (long)be64_to_cpu(crq->request_capability_rsp.number),
4428 			 name);
4429 
4430 		if (be16_to_cpu(crq->request_capability_rsp.capability) ==
4431 		    REQ_MTU) {
4432 			pr_err("mtu of %llu is not supported. Reverting.\n",
4433 			       *req_value);
4434 			*req_value = adapter->fallback.mtu;
4435 		} else {
4436 			*req_value =
4437 				be64_to_cpu(crq->request_capability_rsp.number);
4438 		}
4439 
4440 		send_request_cap(adapter, 1);
4441 		return;
4442 	default:
4443 		dev_err(dev, "Error %d in request cap rsp\n",
4444 			crq->request_capability_rsp.rc.code);
4445 		return;
4446 	}
4447 
4448 	/* Done receiving requested capabilities, query IP offload support */
4449 	if (atomic_read(&adapter->running_cap_crqs) == 0) {
4450 		adapter->wait_capability = false;
4451 		send_query_ip_offload(adapter);
4452 	}
4453 }
4454 
4455 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq,
4456 			    struct ibmvnic_adapter *adapter)
4457 {
4458 	struct device *dev = &adapter->vdev->dev;
4459 	struct net_device *netdev = adapter->netdev;
4460 	struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf;
4461 	struct ibmvnic_login_buffer *login = adapter->login_buf;
4462 	u64 *tx_handle_array;
4463 	u64 *rx_handle_array;
4464 	int num_tx_pools;
4465 	int num_rx_pools;
4466 	u64 *size_array;
4467 	int i;
4468 
4469 	/* CHECK: Test/set of login_pending does not need to be atomic
4470 	 * because only ibmvnic_tasklet tests/clears this.
4471 	 */
4472 	if (!adapter->login_pending) {
4473 		netdev_warn(netdev, "Ignoring unexpected login response\n");
4474 		return 0;
4475 	}
4476 	adapter->login_pending = false;
4477 
4478 	dma_unmap_single(dev, adapter->login_buf_token, adapter->login_buf_sz,
4479 			 DMA_TO_DEVICE);
4480 	dma_unmap_single(dev, adapter->login_rsp_buf_token,
4481 			 adapter->login_rsp_buf_sz, DMA_FROM_DEVICE);
4482 
4483 	/* If the number of queues requested can't be allocated by the
4484 	 * server, the login response will return with code 1. We will need
4485 	 * to resend the login buffer with fewer queues requested.
4486 	 */
4487 	if (login_rsp_crq->generic.rc.code) {
4488 		adapter->init_done_rc = login_rsp_crq->generic.rc.code;
4489 		complete(&adapter->init_done);
4490 		return 0;
4491 	}
4492 
4493 	netdev->mtu = adapter->req_mtu - ETH_HLEN;
4494 
4495 	netdev_dbg(adapter->netdev, "Login Response Buffer:\n");
4496 	for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) {
4497 		netdev_dbg(adapter->netdev, "%016lx\n",
4498 			   ((unsigned long *)(adapter->login_rsp_buf))[i]);
4499 	}
4500 
4501 	/* Sanity checks */
4502 	if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs ||
4503 	    (be32_to_cpu(login->num_rxcomp_subcrqs) *
4504 	     adapter->req_rx_add_queues !=
4505 	     be32_to_cpu(login_rsp->num_rxadd_subcrqs))) {
4506 		dev_err(dev, "FATAL: Inconsistent login and login rsp\n");
4507 		ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4508 		return -EIO;
4509 	}
4510 	size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4511 		be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size));
4512 	/* variable buffer sizes are not supported, so just read the
4513 	 * first entry.
4514 	 */
4515 	adapter->cur_rx_buf_sz = be64_to_cpu(size_array[0]);
4516 
4517 	num_tx_pools = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs);
4518 	num_rx_pools = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
4519 
4520 	tx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4521 				  be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs));
4522 	rx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4523 				  be32_to_cpu(adapter->login_rsp_buf->off_rxadd_subcrqs));
4524 
4525 	for (i = 0; i < num_tx_pools; i++)
4526 		adapter->tx_scrq[i]->handle = tx_handle_array[i];
4527 
4528 	for (i = 0; i < num_rx_pools; i++)
4529 		adapter->rx_scrq[i]->handle = rx_handle_array[i];
4530 
4531 	adapter->num_active_tx_scrqs = num_tx_pools;
4532 	adapter->num_active_rx_scrqs = num_rx_pools;
4533 	release_login_rsp_buffer(adapter);
4534 	release_login_buffer(adapter);
4535 	complete(&adapter->init_done);
4536 
4537 	return 0;
4538 }
4539 
4540 static void handle_request_unmap_rsp(union ibmvnic_crq *crq,
4541 				     struct ibmvnic_adapter *adapter)
4542 {
4543 	struct device *dev = &adapter->vdev->dev;
4544 	long rc;
4545 
4546 	rc = crq->request_unmap_rsp.rc.code;
4547 	if (rc)
4548 		dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc);
4549 }
4550 
4551 static void handle_query_map_rsp(union ibmvnic_crq *crq,
4552 				 struct ibmvnic_adapter *adapter)
4553 {
4554 	struct net_device *netdev = adapter->netdev;
4555 	struct device *dev = &adapter->vdev->dev;
4556 	long rc;
4557 
4558 	rc = crq->query_map_rsp.rc.code;
4559 	if (rc) {
4560 		dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc);
4561 		return;
4562 	}
4563 	netdev_dbg(netdev, "page_size = %d\ntot_pages = %d\nfree_pages = %d\n",
4564 		   crq->query_map_rsp.page_size, crq->query_map_rsp.tot_pages,
4565 		   crq->query_map_rsp.free_pages);
4566 }
4567 
4568 static void handle_query_cap_rsp(union ibmvnic_crq *crq,
4569 				 struct ibmvnic_adapter *adapter)
4570 {
4571 	struct net_device *netdev = adapter->netdev;
4572 	struct device *dev = &adapter->vdev->dev;
4573 	long rc;
4574 
4575 	atomic_dec(&adapter->running_cap_crqs);
4576 	netdev_dbg(netdev, "Outstanding queries: %d\n",
4577 		   atomic_read(&adapter->running_cap_crqs));
4578 	rc = crq->query_capability.rc.code;
4579 	if (rc) {
4580 		dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc);
4581 		goto out;
4582 	}
4583 
4584 	switch (be16_to_cpu(crq->query_capability.capability)) {
4585 	case MIN_TX_QUEUES:
4586 		adapter->min_tx_queues =
4587 		    be64_to_cpu(crq->query_capability.number);
4588 		netdev_dbg(netdev, "min_tx_queues = %lld\n",
4589 			   adapter->min_tx_queues);
4590 		break;
4591 	case MIN_RX_QUEUES:
4592 		adapter->min_rx_queues =
4593 		    be64_to_cpu(crq->query_capability.number);
4594 		netdev_dbg(netdev, "min_rx_queues = %lld\n",
4595 			   adapter->min_rx_queues);
4596 		break;
4597 	case MIN_RX_ADD_QUEUES:
4598 		adapter->min_rx_add_queues =
4599 		    be64_to_cpu(crq->query_capability.number);
4600 		netdev_dbg(netdev, "min_rx_add_queues = %lld\n",
4601 			   adapter->min_rx_add_queues);
4602 		break;
4603 	case MAX_TX_QUEUES:
4604 		adapter->max_tx_queues =
4605 		    be64_to_cpu(crq->query_capability.number);
4606 		netdev_dbg(netdev, "max_tx_queues = %lld\n",
4607 			   adapter->max_tx_queues);
4608 		break;
4609 	case MAX_RX_QUEUES:
4610 		adapter->max_rx_queues =
4611 		    be64_to_cpu(crq->query_capability.number);
4612 		netdev_dbg(netdev, "max_rx_queues = %lld\n",
4613 			   adapter->max_rx_queues);
4614 		break;
4615 	case MAX_RX_ADD_QUEUES:
4616 		adapter->max_rx_add_queues =
4617 		    be64_to_cpu(crq->query_capability.number);
4618 		netdev_dbg(netdev, "max_rx_add_queues = %lld\n",
4619 			   adapter->max_rx_add_queues);
4620 		break;
4621 	case MIN_TX_ENTRIES_PER_SUBCRQ:
4622 		adapter->min_tx_entries_per_subcrq =
4623 		    be64_to_cpu(crq->query_capability.number);
4624 		netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n",
4625 			   adapter->min_tx_entries_per_subcrq);
4626 		break;
4627 	case MIN_RX_ADD_ENTRIES_PER_SUBCRQ:
4628 		adapter->min_rx_add_entries_per_subcrq =
4629 		    be64_to_cpu(crq->query_capability.number);
4630 		netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n",
4631 			   adapter->min_rx_add_entries_per_subcrq);
4632 		break;
4633 	case MAX_TX_ENTRIES_PER_SUBCRQ:
4634 		adapter->max_tx_entries_per_subcrq =
4635 		    be64_to_cpu(crq->query_capability.number);
4636 		netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n",
4637 			   adapter->max_tx_entries_per_subcrq);
4638 		break;
4639 	case MAX_RX_ADD_ENTRIES_PER_SUBCRQ:
4640 		adapter->max_rx_add_entries_per_subcrq =
4641 		    be64_to_cpu(crq->query_capability.number);
4642 		netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n",
4643 			   adapter->max_rx_add_entries_per_subcrq);
4644 		break;
4645 	case TCP_IP_OFFLOAD:
4646 		adapter->tcp_ip_offload =
4647 		    be64_to_cpu(crq->query_capability.number);
4648 		netdev_dbg(netdev, "tcp_ip_offload = %lld\n",
4649 			   adapter->tcp_ip_offload);
4650 		break;
4651 	case PROMISC_SUPPORTED:
4652 		adapter->promisc_supported =
4653 		    be64_to_cpu(crq->query_capability.number);
4654 		netdev_dbg(netdev, "promisc_supported = %lld\n",
4655 			   adapter->promisc_supported);
4656 		break;
4657 	case MIN_MTU:
4658 		adapter->min_mtu = be64_to_cpu(crq->query_capability.number);
4659 		netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
4660 		netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu);
4661 		break;
4662 	case MAX_MTU:
4663 		adapter->max_mtu = be64_to_cpu(crq->query_capability.number);
4664 		netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
4665 		netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu);
4666 		break;
4667 	case MAX_MULTICAST_FILTERS:
4668 		adapter->max_multicast_filters =
4669 		    be64_to_cpu(crq->query_capability.number);
4670 		netdev_dbg(netdev, "max_multicast_filters = %lld\n",
4671 			   adapter->max_multicast_filters);
4672 		break;
4673 	case VLAN_HEADER_INSERTION:
4674 		adapter->vlan_header_insertion =
4675 		    be64_to_cpu(crq->query_capability.number);
4676 		if (adapter->vlan_header_insertion)
4677 			netdev->features |= NETIF_F_HW_VLAN_STAG_TX;
4678 		netdev_dbg(netdev, "vlan_header_insertion = %lld\n",
4679 			   adapter->vlan_header_insertion);
4680 		break;
4681 	case RX_VLAN_HEADER_INSERTION:
4682 		adapter->rx_vlan_header_insertion =
4683 		    be64_to_cpu(crq->query_capability.number);
4684 		netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n",
4685 			   adapter->rx_vlan_header_insertion);
4686 		break;
4687 	case MAX_TX_SG_ENTRIES:
4688 		adapter->max_tx_sg_entries =
4689 		    be64_to_cpu(crq->query_capability.number);
4690 		netdev_dbg(netdev, "max_tx_sg_entries = %lld\n",
4691 			   adapter->max_tx_sg_entries);
4692 		break;
4693 	case RX_SG_SUPPORTED:
4694 		adapter->rx_sg_supported =
4695 		    be64_to_cpu(crq->query_capability.number);
4696 		netdev_dbg(netdev, "rx_sg_supported = %lld\n",
4697 			   adapter->rx_sg_supported);
4698 		break;
4699 	case OPT_TX_COMP_SUB_QUEUES:
4700 		adapter->opt_tx_comp_sub_queues =
4701 		    be64_to_cpu(crq->query_capability.number);
4702 		netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n",
4703 			   adapter->opt_tx_comp_sub_queues);
4704 		break;
4705 	case OPT_RX_COMP_QUEUES:
4706 		adapter->opt_rx_comp_queues =
4707 		    be64_to_cpu(crq->query_capability.number);
4708 		netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n",
4709 			   adapter->opt_rx_comp_queues);
4710 		break;
4711 	case OPT_RX_BUFADD_Q_PER_RX_COMP_Q:
4712 		adapter->opt_rx_bufadd_q_per_rx_comp_q =
4713 		    be64_to_cpu(crq->query_capability.number);
4714 		netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n",
4715 			   adapter->opt_rx_bufadd_q_per_rx_comp_q);
4716 		break;
4717 	case OPT_TX_ENTRIES_PER_SUBCRQ:
4718 		adapter->opt_tx_entries_per_subcrq =
4719 		    be64_to_cpu(crq->query_capability.number);
4720 		netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n",
4721 			   adapter->opt_tx_entries_per_subcrq);
4722 		break;
4723 	case OPT_RXBA_ENTRIES_PER_SUBCRQ:
4724 		adapter->opt_rxba_entries_per_subcrq =
4725 		    be64_to_cpu(crq->query_capability.number);
4726 		netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n",
4727 			   adapter->opt_rxba_entries_per_subcrq);
4728 		break;
4729 	case TX_RX_DESC_REQ:
4730 		adapter->tx_rx_desc_req = crq->query_capability.number;
4731 		netdev_dbg(netdev, "tx_rx_desc_req = %llx\n",
4732 			   adapter->tx_rx_desc_req);
4733 		break;
4734 
4735 	default:
4736 		netdev_err(netdev, "Got invalid cap rsp %d\n",
4737 			   crq->query_capability.capability);
4738 	}
4739 
4740 out:
4741 	if (atomic_read(&adapter->running_cap_crqs) == 0) {
4742 		adapter->wait_capability = false;
4743 		send_request_cap(adapter, 0);
4744 	}
4745 }
4746 
4747 static int send_query_phys_parms(struct ibmvnic_adapter *adapter)
4748 {
4749 	union ibmvnic_crq crq;
4750 	int rc;
4751 
4752 	memset(&crq, 0, sizeof(crq));
4753 	crq.query_phys_parms.first = IBMVNIC_CRQ_CMD;
4754 	crq.query_phys_parms.cmd = QUERY_PHYS_PARMS;
4755 
4756 	mutex_lock(&adapter->fw_lock);
4757 	adapter->fw_done_rc = 0;
4758 	reinit_completion(&adapter->fw_done);
4759 
4760 	rc = ibmvnic_send_crq(adapter, &crq);
4761 	if (rc) {
4762 		mutex_unlock(&adapter->fw_lock);
4763 		return rc;
4764 	}
4765 
4766 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
4767 	if (rc) {
4768 		mutex_unlock(&adapter->fw_lock);
4769 		return rc;
4770 	}
4771 
4772 	mutex_unlock(&adapter->fw_lock);
4773 	return adapter->fw_done_rc ? -EIO : 0;
4774 }
4775 
4776 static int handle_query_phys_parms_rsp(union ibmvnic_crq *crq,
4777 				       struct ibmvnic_adapter *adapter)
4778 {
4779 	struct net_device *netdev = adapter->netdev;
4780 	int rc;
4781 	__be32 rspeed = cpu_to_be32(crq->query_phys_parms_rsp.speed);
4782 
4783 	rc = crq->query_phys_parms_rsp.rc.code;
4784 	if (rc) {
4785 		netdev_err(netdev, "Error %d in QUERY_PHYS_PARMS\n", rc);
4786 		return rc;
4787 	}
4788 	switch (rspeed) {
4789 	case IBMVNIC_10MBPS:
4790 		adapter->speed = SPEED_10;
4791 		break;
4792 	case IBMVNIC_100MBPS:
4793 		adapter->speed = SPEED_100;
4794 		break;
4795 	case IBMVNIC_1GBPS:
4796 		adapter->speed = SPEED_1000;
4797 		break;
4798 	case IBMVNIC_10GBPS:
4799 		adapter->speed = SPEED_10000;
4800 		break;
4801 	case IBMVNIC_25GBPS:
4802 		adapter->speed = SPEED_25000;
4803 		break;
4804 	case IBMVNIC_40GBPS:
4805 		adapter->speed = SPEED_40000;
4806 		break;
4807 	case IBMVNIC_50GBPS:
4808 		adapter->speed = SPEED_50000;
4809 		break;
4810 	case IBMVNIC_100GBPS:
4811 		adapter->speed = SPEED_100000;
4812 		break;
4813 	case IBMVNIC_200GBPS:
4814 		adapter->speed = SPEED_200000;
4815 		break;
4816 	default:
4817 		if (netif_carrier_ok(netdev))
4818 			netdev_warn(netdev, "Unknown speed 0x%08x\n", rspeed);
4819 		adapter->speed = SPEED_UNKNOWN;
4820 	}
4821 	if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_FULL_DUPLEX)
4822 		adapter->duplex = DUPLEX_FULL;
4823 	else if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_HALF_DUPLEX)
4824 		adapter->duplex = DUPLEX_HALF;
4825 	else
4826 		adapter->duplex = DUPLEX_UNKNOWN;
4827 
4828 	return rc;
4829 }
4830 
4831 static void ibmvnic_handle_crq(union ibmvnic_crq *crq,
4832 			       struct ibmvnic_adapter *adapter)
4833 {
4834 	struct ibmvnic_generic_crq *gen_crq = &crq->generic;
4835 	struct net_device *netdev = adapter->netdev;
4836 	struct device *dev = &adapter->vdev->dev;
4837 	u64 *u64_crq = (u64 *)crq;
4838 	long rc;
4839 
4840 	netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n",
4841 		   (unsigned long)cpu_to_be64(u64_crq[0]),
4842 		   (unsigned long)cpu_to_be64(u64_crq[1]));
4843 	switch (gen_crq->first) {
4844 	case IBMVNIC_CRQ_INIT_RSP:
4845 		switch (gen_crq->cmd) {
4846 		case IBMVNIC_CRQ_INIT:
4847 			dev_info(dev, "Partner initialized\n");
4848 			adapter->from_passive_init = true;
4849 			/* Discard any stale login responses from prev reset.
4850 			 * CHECK: should we clear even on INIT_COMPLETE?
4851 			 */
4852 			adapter->login_pending = false;
4853 
4854 			if (!completion_done(&adapter->init_done)) {
4855 				complete(&adapter->init_done);
4856 				adapter->init_done_rc = -EIO;
4857 			}
4858 			rc = ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
4859 			if (rc && rc != -EBUSY) {
4860 				/* We were unable to schedule the failover
4861 				 * reset either because the adapter was still
4862 				 * probing (eg: during kexec) or we could not
4863 				 * allocate memory. Clear the failover_pending
4864 				 * flag since no one else will. We ignore
4865 				 * EBUSY because it means either FAILOVER reset
4866 				 * is already scheduled or the adapter is
4867 				 * being removed.
4868 				 */
4869 				netdev_err(netdev,
4870 					   "Error %ld scheduling failover reset\n",
4871 					   rc);
4872 				adapter->failover_pending = false;
4873 			}
4874 			break;
4875 		case IBMVNIC_CRQ_INIT_COMPLETE:
4876 			dev_info(dev, "Partner initialization complete\n");
4877 			adapter->crq.active = true;
4878 			send_version_xchg(adapter);
4879 			break;
4880 		default:
4881 			dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd);
4882 		}
4883 		return;
4884 	case IBMVNIC_CRQ_XPORT_EVENT:
4885 		netif_carrier_off(netdev);
4886 		adapter->crq.active = false;
4887 		/* terminate any thread waiting for a response
4888 		 * from the device
4889 		 */
4890 		if (!completion_done(&adapter->fw_done)) {
4891 			adapter->fw_done_rc = -EIO;
4892 			complete(&adapter->fw_done);
4893 		}
4894 		if (!completion_done(&adapter->stats_done))
4895 			complete(&adapter->stats_done);
4896 		if (test_bit(0, &adapter->resetting))
4897 			adapter->force_reset_recovery = true;
4898 		if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) {
4899 			dev_info(dev, "Migrated, re-enabling adapter\n");
4900 			ibmvnic_reset(adapter, VNIC_RESET_MOBILITY);
4901 		} else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) {
4902 			dev_info(dev, "Backing device failover detected\n");
4903 			adapter->failover_pending = true;
4904 		} else {
4905 			/* The adapter lost the connection */
4906 			dev_err(dev, "Virtual Adapter failed (rc=%d)\n",
4907 				gen_crq->cmd);
4908 			ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4909 		}
4910 		return;
4911 	case IBMVNIC_CRQ_CMD_RSP:
4912 		break;
4913 	default:
4914 		dev_err(dev, "Got an invalid msg type 0x%02x\n",
4915 			gen_crq->first);
4916 		return;
4917 	}
4918 
4919 	switch (gen_crq->cmd) {
4920 	case VERSION_EXCHANGE_RSP:
4921 		rc = crq->version_exchange_rsp.rc.code;
4922 		if (rc) {
4923 			dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc);
4924 			break;
4925 		}
4926 		ibmvnic_version =
4927 			    be16_to_cpu(crq->version_exchange_rsp.version);
4928 		dev_info(dev, "Partner protocol version is %d\n",
4929 			 ibmvnic_version);
4930 		send_query_cap(adapter);
4931 		break;
4932 	case QUERY_CAPABILITY_RSP:
4933 		handle_query_cap_rsp(crq, adapter);
4934 		break;
4935 	case QUERY_MAP_RSP:
4936 		handle_query_map_rsp(crq, adapter);
4937 		break;
4938 	case REQUEST_MAP_RSP:
4939 		adapter->fw_done_rc = crq->request_map_rsp.rc.code;
4940 		complete(&adapter->fw_done);
4941 		break;
4942 	case REQUEST_UNMAP_RSP:
4943 		handle_request_unmap_rsp(crq, adapter);
4944 		break;
4945 	case REQUEST_CAPABILITY_RSP:
4946 		handle_request_cap_rsp(crq, adapter);
4947 		break;
4948 	case LOGIN_RSP:
4949 		netdev_dbg(netdev, "Got Login Response\n");
4950 		handle_login_rsp(crq, adapter);
4951 		break;
4952 	case LOGICAL_LINK_STATE_RSP:
4953 		netdev_dbg(netdev,
4954 			   "Got Logical Link State Response, state: %d rc: %d\n",
4955 			   crq->logical_link_state_rsp.link_state,
4956 			   crq->logical_link_state_rsp.rc.code);
4957 		adapter->logical_link_state =
4958 		    crq->logical_link_state_rsp.link_state;
4959 		adapter->init_done_rc = crq->logical_link_state_rsp.rc.code;
4960 		complete(&adapter->init_done);
4961 		break;
4962 	case LINK_STATE_INDICATION:
4963 		netdev_dbg(netdev, "Got Logical Link State Indication\n");
4964 		adapter->phys_link_state =
4965 		    crq->link_state_indication.phys_link_state;
4966 		adapter->logical_link_state =
4967 		    crq->link_state_indication.logical_link_state;
4968 		if (adapter->phys_link_state && adapter->logical_link_state)
4969 			netif_carrier_on(netdev);
4970 		else
4971 			netif_carrier_off(netdev);
4972 		break;
4973 	case CHANGE_MAC_ADDR_RSP:
4974 		netdev_dbg(netdev, "Got MAC address change Response\n");
4975 		adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter);
4976 		break;
4977 	case ERROR_INDICATION:
4978 		netdev_dbg(netdev, "Got Error Indication\n");
4979 		handle_error_indication(crq, adapter);
4980 		break;
4981 	case REQUEST_STATISTICS_RSP:
4982 		netdev_dbg(netdev, "Got Statistics Response\n");
4983 		complete(&adapter->stats_done);
4984 		break;
4985 	case QUERY_IP_OFFLOAD_RSP:
4986 		netdev_dbg(netdev, "Got Query IP offload Response\n");
4987 		handle_query_ip_offload_rsp(adapter);
4988 		break;
4989 	case MULTICAST_CTRL_RSP:
4990 		netdev_dbg(netdev, "Got multicast control Response\n");
4991 		break;
4992 	case CONTROL_IP_OFFLOAD_RSP:
4993 		netdev_dbg(netdev, "Got Control IP offload Response\n");
4994 		dma_unmap_single(dev, adapter->ip_offload_ctrl_tok,
4995 				 sizeof(adapter->ip_offload_ctrl),
4996 				 DMA_TO_DEVICE);
4997 		complete(&adapter->init_done);
4998 		break;
4999 	case COLLECT_FW_TRACE_RSP:
5000 		netdev_dbg(netdev, "Got Collect firmware trace Response\n");
5001 		complete(&adapter->fw_done);
5002 		break;
5003 	case GET_VPD_SIZE_RSP:
5004 		handle_vpd_size_rsp(crq, adapter);
5005 		break;
5006 	case GET_VPD_RSP:
5007 		handle_vpd_rsp(crq, adapter);
5008 		break;
5009 	case QUERY_PHYS_PARMS_RSP:
5010 		adapter->fw_done_rc = handle_query_phys_parms_rsp(crq, adapter);
5011 		complete(&adapter->fw_done);
5012 		break;
5013 	default:
5014 		netdev_err(netdev, "Got an invalid cmd type 0x%02x\n",
5015 			   gen_crq->cmd);
5016 	}
5017 }
5018 
5019 static irqreturn_t ibmvnic_interrupt(int irq, void *instance)
5020 {
5021 	struct ibmvnic_adapter *adapter = instance;
5022 
5023 	tasklet_schedule(&adapter->tasklet);
5024 	return IRQ_HANDLED;
5025 }
5026 
5027 static void ibmvnic_tasklet(struct tasklet_struct *t)
5028 {
5029 	struct ibmvnic_adapter *adapter = from_tasklet(adapter, t, tasklet);
5030 	struct ibmvnic_crq_queue *queue = &adapter->crq;
5031 	union ibmvnic_crq *crq;
5032 	unsigned long flags;
5033 	bool done = false;
5034 
5035 	spin_lock_irqsave(&queue->lock, flags);
5036 	while (!done) {
5037 		/* Pull all the valid messages off the CRQ */
5038 		while ((crq = ibmvnic_next_crq(adapter)) != NULL) {
5039 			/* This barrier makes sure ibmvnic_next_crq()'s
5040 			 * crq->generic.first & IBMVNIC_CRQ_CMD_RSP is loaded
5041 			 * before ibmvnic_handle_crq()'s
5042 			 * switch(gen_crq->first) and switch(gen_crq->cmd).
5043 			 */
5044 			dma_rmb();
5045 			ibmvnic_handle_crq(crq, adapter);
5046 			crq->generic.first = 0;
5047 		}
5048 
5049 		/* remain in tasklet until all
5050 		 * capabilities responses are received
5051 		 */
5052 		if (!adapter->wait_capability)
5053 			done = true;
5054 	}
5055 	/* if capabilities CRQ's were sent in this tasklet, the following
5056 	 * tasklet must wait until all responses are received
5057 	 */
5058 	if (atomic_read(&adapter->running_cap_crqs) != 0)
5059 		adapter->wait_capability = true;
5060 	spin_unlock_irqrestore(&queue->lock, flags);
5061 }
5062 
5063 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter)
5064 {
5065 	struct vio_dev *vdev = adapter->vdev;
5066 	int rc;
5067 
5068 	do {
5069 		rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
5070 	} while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc));
5071 
5072 	if (rc)
5073 		dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc);
5074 
5075 	return rc;
5076 }
5077 
5078 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter)
5079 {
5080 	struct ibmvnic_crq_queue *crq = &adapter->crq;
5081 	struct device *dev = &adapter->vdev->dev;
5082 	struct vio_dev *vdev = adapter->vdev;
5083 	int rc;
5084 
5085 	/* Close the CRQ */
5086 	do {
5087 		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5088 	} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5089 
5090 	/* Clean out the queue */
5091 	if (!crq->msgs)
5092 		return -EINVAL;
5093 
5094 	memset(crq->msgs, 0, PAGE_SIZE);
5095 	crq->cur = 0;
5096 	crq->active = false;
5097 
5098 	/* And re-open it again */
5099 	rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5100 				crq->msg_token, PAGE_SIZE);
5101 
5102 	if (rc == H_CLOSED)
5103 		/* Adapter is good, but other end is not ready */
5104 		dev_warn(dev, "Partner adapter not ready\n");
5105 	else if (rc != 0)
5106 		dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc);
5107 
5108 	return rc;
5109 }
5110 
5111 static void release_crq_queue(struct ibmvnic_adapter *adapter)
5112 {
5113 	struct ibmvnic_crq_queue *crq = &adapter->crq;
5114 	struct vio_dev *vdev = adapter->vdev;
5115 	long rc;
5116 
5117 	if (!crq->msgs)
5118 		return;
5119 
5120 	netdev_dbg(adapter->netdev, "Releasing CRQ\n");
5121 	free_irq(vdev->irq, adapter);
5122 	tasklet_kill(&adapter->tasklet);
5123 	do {
5124 		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5125 	} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5126 
5127 	dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE,
5128 			 DMA_BIDIRECTIONAL);
5129 	free_page((unsigned long)crq->msgs);
5130 	crq->msgs = NULL;
5131 	crq->active = false;
5132 }
5133 
5134 static int init_crq_queue(struct ibmvnic_adapter *adapter)
5135 {
5136 	struct ibmvnic_crq_queue *crq = &adapter->crq;
5137 	struct device *dev = &adapter->vdev->dev;
5138 	struct vio_dev *vdev = adapter->vdev;
5139 	int rc, retrc = -ENOMEM;
5140 
5141 	if (crq->msgs)
5142 		return 0;
5143 
5144 	crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL);
5145 	/* Should we allocate more than one page? */
5146 
5147 	if (!crq->msgs)
5148 		return -ENOMEM;
5149 
5150 	crq->size = PAGE_SIZE / sizeof(*crq->msgs);
5151 	crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE,
5152 					DMA_BIDIRECTIONAL);
5153 	if (dma_mapping_error(dev, crq->msg_token))
5154 		goto map_failed;
5155 
5156 	rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5157 				crq->msg_token, PAGE_SIZE);
5158 
5159 	if (rc == H_RESOURCE)
5160 		/* maybe kexecing and resource is busy. try a reset */
5161 		rc = ibmvnic_reset_crq(adapter);
5162 	retrc = rc;
5163 
5164 	if (rc == H_CLOSED) {
5165 		dev_warn(dev, "Partner adapter not ready\n");
5166 	} else if (rc) {
5167 		dev_warn(dev, "Error %d opening adapter\n", rc);
5168 		goto reg_crq_failed;
5169 	}
5170 
5171 	retrc = 0;
5172 
5173 	tasklet_setup(&adapter->tasklet, (void *)ibmvnic_tasklet);
5174 
5175 	netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq);
5176 	snprintf(crq->name, sizeof(crq->name), "ibmvnic-%x",
5177 		 adapter->vdev->unit_address);
5178 	rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, crq->name, adapter);
5179 	if (rc) {
5180 		dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n",
5181 			vdev->irq, rc);
5182 		goto req_irq_failed;
5183 	}
5184 
5185 	rc = vio_enable_interrupts(vdev);
5186 	if (rc) {
5187 		dev_err(dev, "Error %d enabling interrupts\n", rc);
5188 		goto req_irq_failed;
5189 	}
5190 
5191 	crq->cur = 0;
5192 	spin_lock_init(&crq->lock);
5193 
5194 	return retrc;
5195 
5196 req_irq_failed:
5197 	tasklet_kill(&adapter->tasklet);
5198 	do {
5199 		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5200 	} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5201 reg_crq_failed:
5202 	dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL);
5203 map_failed:
5204 	free_page((unsigned long)crq->msgs);
5205 	crq->msgs = NULL;
5206 	return retrc;
5207 }
5208 
5209 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset)
5210 {
5211 	struct device *dev = &adapter->vdev->dev;
5212 	unsigned long timeout = msecs_to_jiffies(20000);
5213 	u64 old_num_rx_queues = adapter->req_rx_queues;
5214 	u64 old_num_tx_queues = adapter->req_tx_queues;
5215 	int rc;
5216 
5217 	adapter->from_passive_init = false;
5218 
5219 	if (reset)
5220 		reinit_completion(&adapter->init_done);
5221 
5222 	adapter->init_done_rc = 0;
5223 	rc = ibmvnic_send_crq_init(adapter);
5224 	if (rc) {
5225 		dev_err(dev, "Send crq init failed with error %d\n", rc);
5226 		return rc;
5227 	}
5228 
5229 	if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
5230 		dev_err(dev, "Initialization sequence timed out\n");
5231 		return -1;
5232 	}
5233 
5234 	if (adapter->init_done_rc) {
5235 		release_crq_queue(adapter);
5236 		return adapter->init_done_rc;
5237 	}
5238 
5239 	if (adapter->from_passive_init) {
5240 		adapter->state = VNIC_OPEN;
5241 		adapter->from_passive_init = false;
5242 		return -1;
5243 	}
5244 
5245 	if (reset &&
5246 	    test_bit(0, &adapter->resetting) && !adapter->wait_for_reset &&
5247 	    adapter->reset_reason != VNIC_RESET_MOBILITY) {
5248 		if (adapter->req_rx_queues != old_num_rx_queues ||
5249 		    adapter->req_tx_queues != old_num_tx_queues) {
5250 			release_sub_crqs(adapter, 0);
5251 			rc = init_sub_crqs(adapter);
5252 		} else {
5253 			rc = reset_sub_crq_queues(adapter);
5254 		}
5255 	} else {
5256 		rc = init_sub_crqs(adapter);
5257 	}
5258 
5259 	if (rc) {
5260 		dev_err(dev, "Initialization of sub crqs failed\n");
5261 		release_crq_queue(adapter);
5262 		return rc;
5263 	}
5264 
5265 	rc = init_sub_crq_irqs(adapter);
5266 	if (rc) {
5267 		dev_err(dev, "Failed to initialize sub crq irqs\n");
5268 		release_crq_queue(adapter);
5269 	}
5270 
5271 	return rc;
5272 }
5273 
5274 static struct device_attribute dev_attr_failover;
5275 
5276 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
5277 {
5278 	struct ibmvnic_adapter *adapter;
5279 	struct net_device *netdev;
5280 	unsigned char *mac_addr_p;
5281 	int rc;
5282 
5283 	dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n",
5284 		dev->unit_address);
5285 
5286 	mac_addr_p = (unsigned char *)vio_get_attribute(dev,
5287 							VETH_MAC_ADDR, NULL);
5288 	if (!mac_addr_p) {
5289 		dev_err(&dev->dev,
5290 			"(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n",
5291 			__FILE__, __LINE__);
5292 		return 0;
5293 	}
5294 
5295 	netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter),
5296 				   IBMVNIC_MAX_QUEUES);
5297 	if (!netdev)
5298 		return -ENOMEM;
5299 
5300 	adapter = netdev_priv(netdev);
5301 	adapter->state = VNIC_PROBING;
5302 	dev_set_drvdata(&dev->dev, netdev);
5303 	adapter->vdev = dev;
5304 	adapter->netdev = netdev;
5305 	adapter->login_pending = false;
5306 
5307 	ether_addr_copy(adapter->mac_addr, mac_addr_p);
5308 	ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
5309 	netdev->irq = dev->irq;
5310 	netdev->netdev_ops = &ibmvnic_netdev_ops;
5311 	netdev->ethtool_ops = &ibmvnic_ethtool_ops;
5312 	SET_NETDEV_DEV(netdev, &dev->dev);
5313 
5314 	INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset);
5315 	INIT_DELAYED_WORK(&adapter->ibmvnic_delayed_reset,
5316 			  __ibmvnic_delayed_reset);
5317 	INIT_LIST_HEAD(&adapter->rwi_list);
5318 	spin_lock_init(&adapter->rwi_lock);
5319 	spin_lock_init(&adapter->state_lock);
5320 	mutex_init(&adapter->fw_lock);
5321 	init_completion(&adapter->init_done);
5322 	init_completion(&adapter->fw_done);
5323 	init_completion(&adapter->reset_done);
5324 	init_completion(&adapter->stats_done);
5325 	clear_bit(0, &adapter->resetting);
5326 
5327 	do {
5328 		rc = init_crq_queue(adapter);
5329 		if (rc) {
5330 			dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n",
5331 				rc);
5332 			goto ibmvnic_init_fail;
5333 		}
5334 
5335 		rc = ibmvnic_reset_init(adapter, false);
5336 		if (rc && rc != EAGAIN)
5337 			goto ibmvnic_init_fail;
5338 	} while (rc == EAGAIN);
5339 
5340 	rc = init_stats_buffers(adapter);
5341 	if (rc)
5342 		goto ibmvnic_init_fail;
5343 
5344 	rc = init_stats_token(adapter);
5345 	if (rc)
5346 		goto ibmvnic_stats_fail;
5347 
5348 	netdev->mtu = adapter->req_mtu - ETH_HLEN;
5349 	netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
5350 	netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
5351 
5352 	rc = device_create_file(&dev->dev, &dev_attr_failover);
5353 	if (rc)
5354 		goto ibmvnic_dev_file_err;
5355 
5356 	netif_carrier_off(netdev);
5357 	rc = register_netdev(netdev);
5358 	if (rc) {
5359 		dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc);
5360 		goto ibmvnic_register_fail;
5361 	}
5362 	dev_info(&dev->dev, "ibmvnic registered\n");
5363 
5364 	adapter->state = VNIC_PROBED;
5365 
5366 	adapter->wait_for_reset = false;
5367 	adapter->last_reset_time = jiffies;
5368 	return 0;
5369 
5370 ibmvnic_register_fail:
5371 	device_remove_file(&dev->dev, &dev_attr_failover);
5372 
5373 ibmvnic_dev_file_err:
5374 	release_stats_token(adapter);
5375 
5376 ibmvnic_stats_fail:
5377 	release_stats_buffers(adapter);
5378 
5379 ibmvnic_init_fail:
5380 	release_sub_crqs(adapter, 1);
5381 	release_crq_queue(adapter);
5382 	mutex_destroy(&adapter->fw_lock);
5383 	free_netdev(netdev);
5384 
5385 	return rc;
5386 }
5387 
5388 static void ibmvnic_remove(struct vio_dev *dev)
5389 {
5390 	struct net_device *netdev = dev_get_drvdata(&dev->dev);
5391 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5392 	unsigned long flags;
5393 
5394 	spin_lock_irqsave(&adapter->state_lock, flags);
5395 
5396 	/* If ibmvnic_reset() is scheduling a reset, wait for it to
5397 	 * finish. Then, set the state to REMOVING to prevent it from
5398 	 * scheduling any more work and to have reset functions ignore
5399 	 * any resets that have already been scheduled. Drop the lock
5400 	 * after setting state, so __ibmvnic_reset() which is called
5401 	 * from the flush_work() below, can make progress.
5402 	 */
5403 	spin_lock(&adapter->rwi_lock);
5404 	adapter->state = VNIC_REMOVING;
5405 	spin_unlock(&adapter->rwi_lock);
5406 
5407 	spin_unlock_irqrestore(&adapter->state_lock, flags);
5408 
5409 	flush_work(&adapter->ibmvnic_reset);
5410 	flush_delayed_work(&adapter->ibmvnic_delayed_reset);
5411 
5412 	rtnl_lock();
5413 	unregister_netdevice(netdev);
5414 
5415 	release_resources(adapter);
5416 	release_sub_crqs(adapter, 1);
5417 	release_crq_queue(adapter);
5418 
5419 	release_stats_token(adapter);
5420 	release_stats_buffers(adapter);
5421 
5422 	adapter->state = VNIC_REMOVED;
5423 
5424 	rtnl_unlock();
5425 	mutex_destroy(&adapter->fw_lock);
5426 	device_remove_file(&dev->dev, &dev_attr_failover);
5427 	free_netdev(netdev);
5428 	dev_set_drvdata(&dev->dev, NULL);
5429 }
5430 
5431 static ssize_t failover_store(struct device *dev, struct device_attribute *attr,
5432 			      const char *buf, size_t count)
5433 {
5434 	struct net_device *netdev = dev_get_drvdata(dev);
5435 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5436 	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
5437 	__be64 session_token;
5438 	long rc;
5439 
5440 	if (!sysfs_streq(buf, "1"))
5441 		return -EINVAL;
5442 
5443 	rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address,
5444 			 H_GET_SESSION_TOKEN, 0, 0, 0);
5445 	if (rc) {
5446 		netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n",
5447 			   rc);
5448 		return -EINVAL;
5449 	}
5450 
5451 	session_token = (__be64)retbuf[0];
5452 	netdev_dbg(netdev, "Initiating client failover, session id %llx\n",
5453 		   be64_to_cpu(session_token));
5454 	rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
5455 				H_SESSION_ERR_DETECTED, session_token, 0, 0);
5456 	if (rc) {
5457 		netdev_err(netdev, "Client initiated failover failed, rc %ld\n",
5458 			   rc);
5459 		return -EINVAL;
5460 	}
5461 
5462 	return count;
5463 }
5464 
5465 static DEVICE_ATTR_WO(failover);
5466 
5467 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev)
5468 {
5469 	struct net_device *netdev = dev_get_drvdata(&vdev->dev);
5470 	struct ibmvnic_adapter *adapter;
5471 	struct iommu_table *tbl;
5472 	unsigned long ret = 0;
5473 	int i;
5474 
5475 	tbl = get_iommu_table_base(&vdev->dev);
5476 
5477 	/* netdev inits at probe time along with the structures we need below*/
5478 	if (!netdev)
5479 		return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl);
5480 
5481 	adapter = netdev_priv(netdev);
5482 
5483 	ret += PAGE_SIZE; /* the crq message queue */
5484 	ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl);
5485 
5486 	for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++)
5487 		ret += 4 * PAGE_SIZE; /* the scrq message queue */
5488 
5489 	for (i = 0; i < adapter->num_active_rx_pools; i++)
5490 		ret += adapter->rx_pool[i].size *
5491 		    IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl);
5492 
5493 	return ret;
5494 }
5495 
5496 static int ibmvnic_resume(struct device *dev)
5497 {
5498 	struct net_device *netdev = dev_get_drvdata(dev);
5499 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5500 
5501 	if (adapter->state != VNIC_OPEN)
5502 		return 0;
5503 
5504 	tasklet_schedule(&adapter->tasklet);
5505 
5506 	return 0;
5507 }
5508 
5509 static const struct vio_device_id ibmvnic_device_table[] = {
5510 	{"network", "IBM,vnic"},
5511 	{"", "" }
5512 };
5513 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table);
5514 
5515 static const struct dev_pm_ops ibmvnic_pm_ops = {
5516 	.resume = ibmvnic_resume
5517 };
5518 
5519 static struct vio_driver ibmvnic_driver = {
5520 	.id_table       = ibmvnic_device_table,
5521 	.probe          = ibmvnic_probe,
5522 	.remove         = ibmvnic_remove,
5523 	.get_desired_dma = ibmvnic_get_desired_dma,
5524 	.name		= ibmvnic_driver_name,
5525 	.pm		= &ibmvnic_pm_ops,
5526 };
5527 
5528 /* module functions */
5529 static int __init ibmvnic_module_init(void)
5530 {
5531 	pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string,
5532 		IBMVNIC_DRIVER_VERSION);
5533 
5534 	return vio_register_driver(&ibmvnic_driver);
5535 }
5536 
5537 static void __exit ibmvnic_module_exit(void)
5538 {
5539 	vio_unregister_driver(&ibmvnic_driver);
5540 }
5541 
5542 module_init(ibmvnic_module_init);
5543 module_exit(ibmvnic_module_exit);
5544