xref: /openbmc/linux/drivers/net/ethernet/ibm/ibmvnic.c (revision 4d39c89f0b94bf4a6e1ccf42702e7d80d210a5fd)
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 		for (i = 0; i < adapter->req_rx_queues; i++)
1153 			napi_disable(&adapter->napi[i]);
1154 		release_resources(adapter);
1155 		return rc;
1156 	}
1157 
1158 	netif_tx_start_all_queues(netdev);
1159 
1160 	if (prev_state == VNIC_CLOSED) {
1161 		for (i = 0; i < adapter->req_rx_queues; i++)
1162 			napi_schedule(&adapter->napi[i]);
1163 	}
1164 
1165 	adapter->state = VNIC_OPEN;
1166 	return rc;
1167 }
1168 
1169 static int ibmvnic_open(struct net_device *netdev)
1170 {
1171 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1172 	int rc;
1173 
1174 	ASSERT_RTNL();
1175 
1176 	/* If device failover is pending or we are about to reset, just set
1177 	 * device state and return. Device operation will be handled by reset
1178 	 * routine.
1179 	 *
1180 	 * It should be safe to overwrite the adapter->state here. Since
1181 	 * we hold the rtnl, either the reset has not actually started or
1182 	 * the rtnl got dropped during the set_link_state() in do_reset().
1183 	 * In the former case, no one else is changing the state (again we
1184 	 * have the rtnl) and in the latter case, do_reset() will detect and
1185 	 * honor our setting below.
1186 	 */
1187 	if (adapter->failover_pending || (test_bit(0, &adapter->resetting))) {
1188 		netdev_dbg(netdev, "[S:%d FOP:%d] Resetting, deferring open\n",
1189 			   adapter->state, adapter->failover_pending);
1190 		adapter->state = VNIC_OPEN;
1191 		rc = 0;
1192 		goto out;
1193 	}
1194 
1195 	if (adapter->state != VNIC_CLOSED) {
1196 		rc = ibmvnic_login(netdev);
1197 		if (rc)
1198 			goto out;
1199 
1200 		rc = init_resources(adapter);
1201 		if (rc) {
1202 			netdev_err(netdev, "failed to initialize resources\n");
1203 			release_resources(adapter);
1204 			goto out;
1205 		}
1206 	}
1207 
1208 	rc = __ibmvnic_open(netdev);
1209 
1210 out:
1211 	/* If open failed and there is a pending failover or in-progress reset,
1212 	 * set device state and return. Device operation will be handled by
1213 	 * reset routine. See also comments above regarding rtnl.
1214 	 */
1215 	if (rc &&
1216 	    (adapter->failover_pending || (test_bit(0, &adapter->resetting)))) {
1217 		adapter->state = VNIC_OPEN;
1218 		rc = 0;
1219 	}
1220 	return rc;
1221 }
1222 
1223 static void clean_rx_pools(struct ibmvnic_adapter *adapter)
1224 {
1225 	struct ibmvnic_rx_pool *rx_pool;
1226 	struct ibmvnic_rx_buff *rx_buff;
1227 	u64 rx_entries;
1228 	int rx_scrqs;
1229 	int i, j;
1230 
1231 	if (!adapter->rx_pool)
1232 		return;
1233 
1234 	rx_scrqs = adapter->num_active_rx_pools;
1235 	rx_entries = adapter->req_rx_add_entries_per_subcrq;
1236 
1237 	/* Free any remaining skbs in the rx buffer pools */
1238 	for (i = 0; i < rx_scrqs; i++) {
1239 		rx_pool = &adapter->rx_pool[i];
1240 		if (!rx_pool || !rx_pool->rx_buff)
1241 			continue;
1242 
1243 		netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i);
1244 		for (j = 0; j < rx_entries; j++) {
1245 			rx_buff = &rx_pool->rx_buff[j];
1246 			if (rx_buff && rx_buff->skb) {
1247 				dev_kfree_skb_any(rx_buff->skb);
1248 				rx_buff->skb = NULL;
1249 			}
1250 		}
1251 	}
1252 }
1253 
1254 static void clean_one_tx_pool(struct ibmvnic_adapter *adapter,
1255 			      struct ibmvnic_tx_pool *tx_pool)
1256 {
1257 	struct ibmvnic_tx_buff *tx_buff;
1258 	u64 tx_entries;
1259 	int i;
1260 
1261 	if (!tx_pool || !tx_pool->tx_buff)
1262 		return;
1263 
1264 	tx_entries = tx_pool->num_buffers;
1265 
1266 	for (i = 0; i < tx_entries; i++) {
1267 		tx_buff = &tx_pool->tx_buff[i];
1268 		if (tx_buff && tx_buff->skb) {
1269 			dev_kfree_skb_any(tx_buff->skb);
1270 			tx_buff->skb = NULL;
1271 		}
1272 	}
1273 }
1274 
1275 static void clean_tx_pools(struct ibmvnic_adapter *adapter)
1276 {
1277 	int tx_scrqs;
1278 	int i;
1279 
1280 	if (!adapter->tx_pool || !adapter->tso_pool)
1281 		return;
1282 
1283 	tx_scrqs = adapter->num_active_tx_pools;
1284 
1285 	/* Free any remaining skbs in the tx buffer pools */
1286 	for (i = 0; i < tx_scrqs; i++) {
1287 		netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i);
1288 		clean_one_tx_pool(adapter, &adapter->tx_pool[i]);
1289 		clean_one_tx_pool(adapter, &adapter->tso_pool[i]);
1290 	}
1291 }
1292 
1293 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter)
1294 {
1295 	struct net_device *netdev = adapter->netdev;
1296 	int i;
1297 
1298 	if (adapter->tx_scrq) {
1299 		for (i = 0; i < adapter->req_tx_queues; i++)
1300 			if (adapter->tx_scrq[i]->irq) {
1301 				netdev_dbg(netdev,
1302 					   "Disabling tx_scrq[%d] irq\n", i);
1303 				disable_scrq_irq(adapter, adapter->tx_scrq[i]);
1304 				disable_irq(adapter->tx_scrq[i]->irq);
1305 			}
1306 	}
1307 
1308 	if (adapter->rx_scrq) {
1309 		for (i = 0; i < adapter->req_rx_queues; i++) {
1310 			if (adapter->rx_scrq[i]->irq) {
1311 				netdev_dbg(netdev,
1312 					   "Disabling rx_scrq[%d] irq\n", i);
1313 				disable_scrq_irq(adapter, adapter->rx_scrq[i]);
1314 				disable_irq(adapter->rx_scrq[i]->irq);
1315 			}
1316 		}
1317 	}
1318 }
1319 
1320 static void ibmvnic_cleanup(struct net_device *netdev)
1321 {
1322 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1323 
1324 	/* ensure that transmissions are stopped if called by do_reset */
1325 	if (test_bit(0, &adapter->resetting))
1326 		netif_tx_disable(netdev);
1327 	else
1328 		netif_tx_stop_all_queues(netdev);
1329 
1330 	ibmvnic_napi_disable(adapter);
1331 	ibmvnic_disable_irqs(adapter);
1332 
1333 	clean_rx_pools(adapter);
1334 	clean_tx_pools(adapter);
1335 }
1336 
1337 static int __ibmvnic_close(struct net_device *netdev)
1338 {
1339 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1340 	int rc = 0;
1341 
1342 	adapter->state = VNIC_CLOSING;
1343 	rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
1344 	adapter->state = VNIC_CLOSED;
1345 	return rc;
1346 }
1347 
1348 static int ibmvnic_close(struct net_device *netdev)
1349 {
1350 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1351 	int rc;
1352 
1353 	netdev_dbg(netdev, "[S:%d FOP:%d FRR:%d] Closing\n",
1354 		   adapter->state, adapter->failover_pending,
1355 		   adapter->force_reset_recovery);
1356 
1357 	/* If device failover is pending, just set device state and return.
1358 	 * Device operation will be handled by reset routine.
1359 	 */
1360 	if (adapter->failover_pending) {
1361 		adapter->state = VNIC_CLOSED;
1362 		return 0;
1363 	}
1364 
1365 	rc = __ibmvnic_close(netdev);
1366 	ibmvnic_cleanup(netdev);
1367 
1368 	return rc;
1369 }
1370 
1371 /**
1372  * build_hdr_data - creates L2/L3/L4 header data buffer
1373  * @hdr_field: bitfield determining needed headers
1374  * @skb: socket buffer
1375  * @hdr_len: array of header lengths
1376  * @hdr_data: buffer to write the header to
1377  *
1378  * Reads hdr_field to determine which headers are needed by firmware.
1379  * Builds a buffer containing these headers.  Saves individual header
1380  * lengths and total buffer length to be used to build descriptors.
1381  */
1382 static int build_hdr_data(u8 hdr_field, struct sk_buff *skb,
1383 			  int *hdr_len, u8 *hdr_data)
1384 {
1385 	int len = 0;
1386 	u8 *hdr;
1387 
1388 	if (skb_vlan_tagged(skb) && !skb_vlan_tag_present(skb))
1389 		hdr_len[0] = sizeof(struct vlan_ethhdr);
1390 	else
1391 		hdr_len[0] = sizeof(struct ethhdr);
1392 
1393 	if (skb->protocol == htons(ETH_P_IP)) {
1394 		hdr_len[1] = ip_hdr(skb)->ihl * 4;
1395 		if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1396 			hdr_len[2] = tcp_hdrlen(skb);
1397 		else if (ip_hdr(skb)->protocol == IPPROTO_UDP)
1398 			hdr_len[2] = sizeof(struct udphdr);
1399 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
1400 		hdr_len[1] = sizeof(struct ipv6hdr);
1401 		if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1402 			hdr_len[2] = tcp_hdrlen(skb);
1403 		else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
1404 			hdr_len[2] = sizeof(struct udphdr);
1405 	} else if (skb->protocol == htons(ETH_P_ARP)) {
1406 		hdr_len[1] = arp_hdr_len(skb->dev);
1407 		hdr_len[2] = 0;
1408 	}
1409 
1410 	memset(hdr_data, 0, 120);
1411 	if ((hdr_field >> 6) & 1) {
1412 		hdr = skb_mac_header(skb);
1413 		memcpy(hdr_data, hdr, hdr_len[0]);
1414 		len += hdr_len[0];
1415 	}
1416 
1417 	if ((hdr_field >> 5) & 1) {
1418 		hdr = skb_network_header(skb);
1419 		memcpy(hdr_data + len, hdr, hdr_len[1]);
1420 		len += hdr_len[1];
1421 	}
1422 
1423 	if ((hdr_field >> 4) & 1) {
1424 		hdr = skb_transport_header(skb);
1425 		memcpy(hdr_data + len, hdr, hdr_len[2]);
1426 		len += hdr_len[2];
1427 	}
1428 	return len;
1429 }
1430 
1431 /**
1432  * create_hdr_descs - create header and header extension descriptors
1433  * @hdr_field: bitfield determining needed headers
1434  * @hdr_data: buffer containing header data
1435  * @len: length of data buffer
1436  * @hdr_len: array of individual header lengths
1437  * @scrq_arr: descriptor array
1438  *
1439  * Creates header and, if needed, header extension descriptors and
1440  * places them in a descriptor array, scrq_arr
1441  */
1442 
1443 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len,
1444 			    union sub_crq *scrq_arr)
1445 {
1446 	union sub_crq hdr_desc;
1447 	int tmp_len = len;
1448 	int num_descs = 0;
1449 	u8 *data, *cur;
1450 	int tmp;
1451 
1452 	while (tmp_len > 0) {
1453 		cur = hdr_data + len - tmp_len;
1454 
1455 		memset(&hdr_desc, 0, sizeof(hdr_desc));
1456 		if (cur != hdr_data) {
1457 			data = hdr_desc.hdr_ext.data;
1458 			tmp = tmp_len > 29 ? 29 : tmp_len;
1459 			hdr_desc.hdr_ext.first = IBMVNIC_CRQ_CMD;
1460 			hdr_desc.hdr_ext.type = IBMVNIC_HDR_EXT_DESC;
1461 			hdr_desc.hdr_ext.len = tmp;
1462 		} else {
1463 			data = hdr_desc.hdr.data;
1464 			tmp = tmp_len > 24 ? 24 : tmp_len;
1465 			hdr_desc.hdr.first = IBMVNIC_CRQ_CMD;
1466 			hdr_desc.hdr.type = IBMVNIC_HDR_DESC;
1467 			hdr_desc.hdr.len = tmp;
1468 			hdr_desc.hdr.l2_len = (u8)hdr_len[0];
1469 			hdr_desc.hdr.l3_len = cpu_to_be16((u16)hdr_len[1]);
1470 			hdr_desc.hdr.l4_len = (u8)hdr_len[2];
1471 			hdr_desc.hdr.flag = hdr_field << 1;
1472 		}
1473 		memcpy(data, cur, tmp);
1474 		tmp_len -= tmp;
1475 		*scrq_arr = hdr_desc;
1476 		scrq_arr++;
1477 		num_descs++;
1478 	}
1479 
1480 	return num_descs;
1481 }
1482 
1483 /**
1484  * build_hdr_descs_arr - build a header descriptor array
1485  * @txbuff: tx buffer
1486  * @num_entries: number of descriptors to be sent
1487  * @hdr_field: bit field determining which headers will be sent
1488  *
1489  * This function will build a TX descriptor array with applicable
1490  * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect.
1491  */
1492 
1493 static void build_hdr_descs_arr(struct sk_buff *skb,
1494 				union sub_crq *indir_arr,
1495 				int *num_entries, u8 hdr_field)
1496 {
1497 	int hdr_len[3] = {0, 0, 0};
1498 	u8 hdr_data[140] = {0};
1499 	int tot_len;
1500 
1501 	tot_len = build_hdr_data(hdr_field, skb, hdr_len,
1502 				 hdr_data);
1503 	*num_entries += create_hdr_descs(hdr_field, hdr_data, tot_len, hdr_len,
1504 					 indir_arr + 1);
1505 }
1506 
1507 static int ibmvnic_xmit_workarounds(struct sk_buff *skb,
1508 				    struct net_device *netdev)
1509 {
1510 	/* For some backing devices, mishandling of small packets
1511 	 * can result in a loss of connection or TX stall. Device
1512 	 * architects recommend that no packet should be smaller
1513 	 * than the minimum MTU value provided to the driver, so
1514 	 * pad any packets to that length
1515 	 */
1516 	if (skb->len < netdev->min_mtu)
1517 		return skb_put_padto(skb, netdev->min_mtu);
1518 
1519 	return 0;
1520 }
1521 
1522 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
1523 					 struct ibmvnic_sub_crq_queue *tx_scrq)
1524 {
1525 	struct ibmvnic_ind_xmit_queue *ind_bufp;
1526 	struct ibmvnic_tx_buff *tx_buff;
1527 	struct ibmvnic_tx_pool *tx_pool;
1528 	union sub_crq tx_scrq_entry;
1529 	int queue_num;
1530 	int entries;
1531 	int index;
1532 	int i;
1533 
1534 	ind_bufp = &tx_scrq->ind_buf;
1535 	entries = (u64)ind_bufp->index;
1536 	queue_num = tx_scrq->pool_index;
1537 
1538 	for (i = entries - 1; i >= 0; --i) {
1539 		tx_scrq_entry = ind_bufp->indir_arr[i];
1540 		if (tx_scrq_entry.v1.type != IBMVNIC_TX_DESC)
1541 			continue;
1542 		index = be32_to_cpu(tx_scrq_entry.v1.correlator);
1543 		if (index & IBMVNIC_TSO_POOL_MASK) {
1544 			tx_pool = &adapter->tso_pool[queue_num];
1545 			index &= ~IBMVNIC_TSO_POOL_MASK;
1546 		} else {
1547 			tx_pool = &adapter->tx_pool[queue_num];
1548 		}
1549 		tx_pool->free_map[tx_pool->consumer_index] = index;
1550 		tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
1551 					  tx_pool->num_buffers - 1 :
1552 					  tx_pool->consumer_index - 1;
1553 		tx_buff = &tx_pool->tx_buff[index];
1554 		adapter->netdev->stats.tx_packets--;
1555 		adapter->netdev->stats.tx_bytes -= tx_buff->skb->len;
1556 		adapter->tx_stats_buffers[queue_num].packets--;
1557 		adapter->tx_stats_buffers[queue_num].bytes -=
1558 						tx_buff->skb->len;
1559 		dev_kfree_skb_any(tx_buff->skb);
1560 		tx_buff->skb = NULL;
1561 		adapter->netdev->stats.tx_dropped++;
1562 	}
1563 	ind_bufp->index = 0;
1564 	if (atomic_sub_return(entries, &tx_scrq->used) <=
1565 	    (adapter->req_tx_entries_per_subcrq / 2) &&
1566 	    __netif_subqueue_stopped(adapter->netdev, queue_num)) {
1567 		netif_wake_subqueue(adapter->netdev, queue_num);
1568 		netdev_dbg(adapter->netdev, "Started queue %d\n",
1569 			   queue_num);
1570 	}
1571 }
1572 
1573 static int ibmvnic_tx_scrq_flush(struct ibmvnic_adapter *adapter,
1574 				 struct ibmvnic_sub_crq_queue *tx_scrq)
1575 {
1576 	struct ibmvnic_ind_xmit_queue *ind_bufp;
1577 	u64 dma_addr;
1578 	u64 entries;
1579 	u64 handle;
1580 	int rc;
1581 
1582 	ind_bufp = &tx_scrq->ind_buf;
1583 	dma_addr = (u64)ind_bufp->indir_dma;
1584 	entries = (u64)ind_bufp->index;
1585 	handle = tx_scrq->handle;
1586 
1587 	if (!entries)
1588 		return 0;
1589 	rc = send_subcrq_indirect(adapter, handle, dma_addr, entries);
1590 	if (rc)
1591 		ibmvnic_tx_scrq_clean_buffer(adapter, tx_scrq);
1592 	else
1593 		ind_bufp->index = 0;
1594 	return 0;
1595 }
1596 
1597 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
1598 {
1599 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1600 	int queue_num = skb_get_queue_mapping(skb);
1601 	u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req;
1602 	struct device *dev = &adapter->vdev->dev;
1603 	struct ibmvnic_ind_xmit_queue *ind_bufp;
1604 	struct ibmvnic_tx_buff *tx_buff = NULL;
1605 	struct ibmvnic_sub_crq_queue *tx_scrq;
1606 	struct ibmvnic_tx_pool *tx_pool;
1607 	unsigned int tx_send_failed = 0;
1608 	netdev_tx_t ret = NETDEV_TX_OK;
1609 	unsigned int tx_map_failed = 0;
1610 	union sub_crq indir_arr[16];
1611 	unsigned int tx_dropped = 0;
1612 	unsigned int tx_packets = 0;
1613 	unsigned int tx_bytes = 0;
1614 	dma_addr_t data_dma_addr;
1615 	struct netdev_queue *txq;
1616 	unsigned long lpar_rc;
1617 	union sub_crq tx_crq;
1618 	unsigned int offset;
1619 	int num_entries = 1;
1620 	unsigned char *dst;
1621 	int index = 0;
1622 	u8 proto = 0;
1623 
1624 	tx_scrq = adapter->tx_scrq[queue_num];
1625 	txq = netdev_get_tx_queue(netdev, queue_num);
1626 	ind_bufp = &tx_scrq->ind_buf;
1627 
1628 	if (test_bit(0, &adapter->resetting)) {
1629 		if (!netif_subqueue_stopped(netdev, skb))
1630 			netif_stop_subqueue(netdev, queue_num);
1631 		dev_kfree_skb_any(skb);
1632 
1633 		tx_send_failed++;
1634 		tx_dropped++;
1635 		ret = NETDEV_TX_OK;
1636 		ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1637 		goto out;
1638 	}
1639 
1640 	if (ibmvnic_xmit_workarounds(skb, netdev)) {
1641 		tx_dropped++;
1642 		tx_send_failed++;
1643 		ret = NETDEV_TX_OK;
1644 		ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1645 		goto out;
1646 	}
1647 	if (skb_is_gso(skb))
1648 		tx_pool = &adapter->tso_pool[queue_num];
1649 	else
1650 		tx_pool = &adapter->tx_pool[queue_num];
1651 
1652 	index = tx_pool->free_map[tx_pool->consumer_index];
1653 
1654 	if (index == IBMVNIC_INVALID_MAP) {
1655 		dev_kfree_skb_any(skb);
1656 		tx_send_failed++;
1657 		tx_dropped++;
1658 		ret = NETDEV_TX_OK;
1659 		ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1660 		goto out;
1661 	}
1662 
1663 	tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP;
1664 
1665 	offset = index * tx_pool->buf_size;
1666 	dst = tx_pool->long_term_buff.buff + offset;
1667 	memset(dst, 0, tx_pool->buf_size);
1668 	data_dma_addr = tx_pool->long_term_buff.addr + offset;
1669 
1670 	if (skb_shinfo(skb)->nr_frags) {
1671 		int cur, i;
1672 
1673 		/* Copy the head */
1674 		skb_copy_from_linear_data(skb, dst, skb_headlen(skb));
1675 		cur = skb_headlen(skb);
1676 
1677 		/* Copy the frags */
1678 		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1679 			const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1680 
1681 			memcpy(dst + cur,
1682 			       page_address(skb_frag_page(frag)) +
1683 			       skb_frag_off(frag), skb_frag_size(frag));
1684 			cur += skb_frag_size(frag);
1685 		}
1686 	} else {
1687 		skb_copy_from_linear_data(skb, dst, skb->len);
1688 	}
1689 
1690 	/* post changes to long_term_buff *dst before VIOS accessing it */
1691 	dma_wmb();
1692 
1693 	tx_pool->consumer_index =
1694 	    (tx_pool->consumer_index + 1) % tx_pool->num_buffers;
1695 
1696 	tx_buff = &tx_pool->tx_buff[index];
1697 	tx_buff->skb = skb;
1698 	tx_buff->index = index;
1699 	tx_buff->pool_index = queue_num;
1700 
1701 	memset(&tx_crq, 0, sizeof(tx_crq));
1702 	tx_crq.v1.first = IBMVNIC_CRQ_CMD;
1703 	tx_crq.v1.type = IBMVNIC_TX_DESC;
1704 	tx_crq.v1.n_crq_elem = 1;
1705 	tx_crq.v1.n_sge = 1;
1706 	tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED;
1707 
1708 	if (skb_is_gso(skb))
1709 		tx_crq.v1.correlator =
1710 			cpu_to_be32(index | IBMVNIC_TSO_POOL_MASK);
1711 	else
1712 		tx_crq.v1.correlator = cpu_to_be32(index);
1713 	tx_crq.v1.dma_reg = cpu_to_be16(tx_pool->long_term_buff.map_id);
1714 	tx_crq.v1.sge_len = cpu_to_be32(skb->len);
1715 	tx_crq.v1.ioba = cpu_to_be64(data_dma_addr);
1716 
1717 	if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) {
1718 		tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT;
1719 		tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci);
1720 	}
1721 
1722 	if (skb->protocol == htons(ETH_P_IP)) {
1723 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4;
1724 		proto = ip_hdr(skb)->protocol;
1725 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
1726 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6;
1727 		proto = ipv6_hdr(skb)->nexthdr;
1728 	}
1729 
1730 	if (proto == IPPROTO_TCP)
1731 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP;
1732 	else if (proto == IPPROTO_UDP)
1733 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP;
1734 
1735 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
1736 		tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD;
1737 		hdrs += 2;
1738 	}
1739 	if (skb_is_gso(skb)) {
1740 		tx_crq.v1.flags1 |= IBMVNIC_TX_LSO;
1741 		tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size);
1742 		hdrs += 2;
1743 	}
1744 
1745 	if ((*hdrs >> 7) & 1)
1746 		build_hdr_descs_arr(skb, indir_arr, &num_entries, *hdrs);
1747 
1748 	tx_crq.v1.n_crq_elem = num_entries;
1749 	tx_buff->num_entries = num_entries;
1750 	/* flush buffer if current entry can not fit */
1751 	if (num_entries + ind_bufp->index > IBMVNIC_MAX_IND_DESCS) {
1752 		lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1753 		if (lpar_rc != H_SUCCESS)
1754 			goto tx_flush_err;
1755 	}
1756 
1757 	indir_arr[0] = tx_crq;
1758 	memcpy(&ind_bufp->indir_arr[ind_bufp->index], &indir_arr[0],
1759 	       num_entries * sizeof(struct ibmvnic_generic_scrq));
1760 	ind_bufp->index += num_entries;
1761 	if (__netdev_tx_sent_queue(txq, skb->len,
1762 				   netdev_xmit_more() &&
1763 				   ind_bufp->index < IBMVNIC_MAX_IND_DESCS)) {
1764 		lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
1765 		if (lpar_rc != H_SUCCESS)
1766 			goto tx_err;
1767 	}
1768 
1769 	if (atomic_add_return(num_entries, &tx_scrq->used)
1770 					>= adapter->req_tx_entries_per_subcrq) {
1771 		netdev_dbg(netdev, "Stopping queue %d\n", queue_num);
1772 		netif_stop_subqueue(netdev, queue_num);
1773 	}
1774 
1775 	tx_packets++;
1776 	tx_bytes += skb->len;
1777 	txq->trans_start = jiffies;
1778 	ret = NETDEV_TX_OK;
1779 	goto out;
1780 
1781 tx_flush_err:
1782 	dev_kfree_skb_any(skb);
1783 	tx_buff->skb = NULL;
1784 	tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
1785 				  tx_pool->num_buffers - 1 :
1786 				  tx_pool->consumer_index - 1;
1787 	tx_dropped++;
1788 tx_err:
1789 	if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER)
1790 		dev_err_ratelimited(dev, "tx: send failed\n");
1791 
1792 	if (lpar_rc == H_CLOSED || adapter->failover_pending) {
1793 		/* Disable TX and report carrier off if queue is closed
1794 		 * or pending failover.
1795 		 * Firmware guarantees that a signal will be sent to the
1796 		 * driver, triggering a reset or some other action.
1797 		 */
1798 		netif_tx_stop_all_queues(netdev);
1799 		netif_carrier_off(netdev);
1800 	}
1801 out:
1802 	netdev->stats.tx_dropped += tx_dropped;
1803 	netdev->stats.tx_bytes += tx_bytes;
1804 	netdev->stats.tx_packets += tx_packets;
1805 	adapter->tx_send_failed += tx_send_failed;
1806 	adapter->tx_map_failed += tx_map_failed;
1807 	adapter->tx_stats_buffers[queue_num].packets += tx_packets;
1808 	adapter->tx_stats_buffers[queue_num].bytes += tx_bytes;
1809 	adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped;
1810 
1811 	return ret;
1812 }
1813 
1814 static void ibmvnic_set_multi(struct net_device *netdev)
1815 {
1816 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1817 	struct netdev_hw_addr *ha;
1818 	union ibmvnic_crq crq;
1819 
1820 	memset(&crq, 0, sizeof(crq));
1821 	crq.request_capability.first = IBMVNIC_CRQ_CMD;
1822 	crq.request_capability.cmd = REQUEST_CAPABILITY;
1823 
1824 	if (netdev->flags & IFF_PROMISC) {
1825 		if (!adapter->promisc_supported)
1826 			return;
1827 	} else {
1828 		if (netdev->flags & IFF_ALLMULTI) {
1829 			/* Accept all multicast */
1830 			memset(&crq, 0, sizeof(crq));
1831 			crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1832 			crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1833 			crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL;
1834 			ibmvnic_send_crq(adapter, &crq);
1835 		} else if (netdev_mc_empty(netdev)) {
1836 			/* Reject all multicast */
1837 			memset(&crq, 0, sizeof(crq));
1838 			crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1839 			crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1840 			crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL;
1841 			ibmvnic_send_crq(adapter, &crq);
1842 		} else {
1843 			/* Accept one or more multicast(s) */
1844 			netdev_for_each_mc_addr(ha, netdev) {
1845 				memset(&crq, 0, sizeof(crq));
1846 				crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
1847 				crq.multicast_ctrl.cmd = MULTICAST_CTRL;
1848 				crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC;
1849 				ether_addr_copy(&crq.multicast_ctrl.mac_addr[0],
1850 						ha->addr);
1851 				ibmvnic_send_crq(adapter, &crq);
1852 			}
1853 		}
1854 	}
1855 }
1856 
1857 static int __ibmvnic_set_mac(struct net_device *netdev, u8 *dev_addr)
1858 {
1859 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1860 	union ibmvnic_crq crq;
1861 	int rc;
1862 
1863 	if (!is_valid_ether_addr(dev_addr)) {
1864 		rc = -EADDRNOTAVAIL;
1865 		goto err;
1866 	}
1867 
1868 	memset(&crq, 0, sizeof(crq));
1869 	crq.change_mac_addr.first = IBMVNIC_CRQ_CMD;
1870 	crq.change_mac_addr.cmd = CHANGE_MAC_ADDR;
1871 	ether_addr_copy(&crq.change_mac_addr.mac_addr[0], dev_addr);
1872 
1873 	mutex_lock(&adapter->fw_lock);
1874 	adapter->fw_done_rc = 0;
1875 	reinit_completion(&adapter->fw_done);
1876 
1877 	rc = ibmvnic_send_crq(adapter, &crq);
1878 	if (rc) {
1879 		rc = -EIO;
1880 		mutex_unlock(&adapter->fw_lock);
1881 		goto err;
1882 	}
1883 
1884 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1885 	/* netdev->dev_addr is changed in handle_change_mac_rsp function */
1886 	if (rc || adapter->fw_done_rc) {
1887 		rc = -EIO;
1888 		mutex_unlock(&adapter->fw_lock);
1889 		goto err;
1890 	}
1891 	mutex_unlock(&adapter->fw_lock);
1892 	return 0;
1893 err:
1894 	ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
1895 	return rc;
1896 }
1897 
1898 static int ibmvnic_set_mac(struct net_device *netdev, void *p)
1899 {
1900 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1901 	struct sockaddr *addr = p;
1902 	int rc;
1903 
1904 	rc = 0;
1905 	if (!is_valid_ether_addr(addr->sa_data))
1906 		return -EADDRNOTAVAIL;
1907 
1908 	if (adapter->state != VNIC_PROBED) {
1909 		ether_addr_copy(adapter->mac_addr, addr->sa_data);
1910 		rc = __ibmvnic_set_mac(netdev, addr->sa_data);
1911 	}
1912 
1913 	return rc;
1914 }
1915 
1916 /*
1917  * do_reset returns zero if we are able to keep processing reset events, or
1918  * non-zero if we hit a fatal error and must halt.
1919  */
1920 static int do_reset(struct ibmvnic_adapter *adapter,
1921 		    struct ibmvnic_rwi *rwi, u32 reset_state)
1922 {
1923 	u64 old_num_rx_queues, old_num_tx_queues;
1924 	u64 old_num_rx_slots, old_num_tx_slots;
1925 	struct net_device *netdev = adapter->netdev;
1926 	int i, rc;
1927 
1928 	netdev_dbg(adapter->netdev,
1929 		   "[S:%d FOP:%d] Reset reason %d, reset_state %d\n",
1930 		   adapter->state, adapter->failover_pending,
1931 		   rwi->reset_reason, reset_state);
1932 
1933 	adapter->reset_reason = rwi->reset_reason;
1934 	/* requestor of VNIC_RESET_CHANGE_PARAM already has the rtnl lock */
1935 	if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
1936 		rtnl_lock();
1937 
1938 	/* Now that we have the rtnl lock, clear any pending failover.
1939 	 * This will ensure ibmvnic_open() has either completed or will
1940 	 * block until failover is complete.
1941 	 */
1942 	if (rwi->reset_reason == VNIC_RESET_FAILOVER)
1943 		adapter->failover_pending = false;
1944 
1945 	/* read the state and check (again) after getting rtnl */
1946 	reset_state = adapter->state;
1947 
1948 	if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
1949 		rc = -EBUSY;
1950 		goto out;
1951 	}
1952 
1953 	netif_carrier_off(netdev);
1954 
1955 	old_num_rx_queues = adapter->req_rx_queues;
1956 	old_num_tx_queues = adapter->req_tx_queues;
1957 	old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq;
1958 	old_num_tx_slots = adapter->req_tx_entries_per_subcrq;
1959 
1960 	ibmvnic_cleanup(netdev);
1961 
1962 	if (reset_state == VNIC_OPEN &&
1963 	    adapter->reset_reason != VNIC_RESET_MOBILITY &&
1964 	    adapter->reset_reason != VNIC_RESET_FAILOVER) {
1965 		if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
1966 			rc = __ibmvnic_close(netdev);
1967 			if (rc)
1968 				goto out;
1969 		} else {
1970 			adapter->state = VNIC_CLOSING;
1971 
1972 			/* Release the RTNL lock before link state change and
1973 			 * re-acquire after the link state change to allow
1974 			 * linkwatch_event to grab the RTNL lock and run during
1975 			 * a reset.
1976 			 */
1977 			rtnl_unlock();
1978 			rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
1979 			rtnl_lock();
1980 			if (rc)
1981 				goto out;
1982 
1983 			if (adapter->state == VNIC_OPEN) {
1984 				/* When we dropped rtnl, ibmvnic_open() got
1985 				 * it and noticed that we are resetting and
1986 				 * set the adapter state to OPEN. Update our
1987 				 * new "target" state, and resume the reset
1988 				 * from VNIC_CLOSING state.
1989 				 */
1990 				netdev_dbg(netdev,
1991 					   "Open changed state from %d, updating.\n",
1992 					   reset_state);
1993 				reset_state = VNIC_OPEN;
1994 				adapter->state = VNIC_CLOSING;
1995 			}
1996 
1997 			if (adapter->state != VNIC_CLOSING) {
1998 				/* If someone else changed the adapter state
1999 				 * when we dropped the rtnl, fail the reset
2000 				 */
2001 				rc = -1;
2002 				goto out;
2003 			}
2004 			adapter->state = VNIC_CLOSED;
2005 		}
2006 	}
2007 
2008 	if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2009 		release_resources(adapter);
2010 		release_sub_crqs(adapter, 1);
2011 		release_crq_queue(adapter);
2012 	}
2013 
2014 	if (adapter->reset_reason != VNIC_RESET_NON_FATAL) {
2015 		/* remove the closed state so when we call open it appears
2016 		 * we are coming from the probed state.
2017 		 */
2018 		adapter->state = VNIC_PROBED;
2019 
2020 		if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2021 			rc = init_crq_queue(adapter);
2022 		} else if (adapter->reset_reason == VNIC_RESET_MOBILITY) {
2023 			rc = ibmvnic_reenable_crq_queue(adapter);
2024 			release_sub_crqs(adapter, 1);
2025 		} else {
2026 			rc = ibmvnic_reset_crq(adapter);
2027 			if (rc == H_CLOSED || rc == H_SUCCESS) {
2028 				rc = vio_enable_interrupts(adapter->vdev);
2029 				if (rc)
2030 					netdev_err(adapter->netdev,
2031 						   "Reset failed to enable interrupts. rc=%d\n",
2032 						   rc);
2033 			}
2034 		}
2035 
2036 		if (rc) {
2037 			netdev_err(adapter->netdev,
2038 				   "Reset couldn't initialize crq. rc=%d\n", rc);
2039 			goto out;
2040 		}
2041 
2042 		rc = ibmvnic_reset_init(adapter, true);
2043 		if (rc) {
2044 			rc = IBMVNIC_INIT_FAILED;
2045 			goto out;
2046 		}
2047 
2048 		/* If the adapter was in PROBE state prior to the reset,
2049 		 * exit here.
2050 		 */
2051 		if (reset_state == VNIC_PROBED) {
2052 			rc = 0;
2053 			goto out;
2054 		}
2055 
2056 		rc = ibmvnic_login(netdev);
2057 		if (rc)
2058 			goto out;
2059 
2060 		if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2061 			rc = init_resources(adapter);
2062 			if (rc)
2063 				goto out;
2064 		} else if (adapter->req_rx_queues != old_num_rx_queues ||
2065 		    adapter->req_tx_queues != old_num_tx_queues ||
2066 		    adapter->req_rx_add_entries_per_subcrq !=
2067 		    old_num_rx_slots ||
2068 		    adapter->req_tx_entries_per_subcrq !=
2069 		    old_num_tx_slots ||
2070 		    !adapter->rx_pool ||
2071 		    !adapter->tso_pool ||
2072 		    !adapter->tx_pool) {
2073 			release_rx_pools(adapter);
2074 			release_tx_pools(adapter);
2075 			release_napi(adapter);
2076 			release_vpd_data(adapter);
2077 
2078 			rc = init_resources(adapter);
2079 			if (rc)
2080 				goto out;
2081 
2082 		} else {
2083 			rc = reset_tx_pools(adapter);
2084 			if (rc) {
2085 				netdev_dbg(adapter->netdev, "reset tx pools failed (%d)\n",
2086 					   rc);
2087 				goto out;
2088 			}
2089 
2090 			rc = reset_rx_pools(adapter);
2091 			if (rc) {
2092 				netdev_dbg(adapter->netdev, "reset rx pools failed (%d)\n",
2093 					   rc);
2094 				goto out;
2095 			}
2096 		}
2097 		ibmvnic_disable_irqs(adapter);
2098 	}
2099 	adapter->state = VNIC_CLOSED;
2100 
2101 	if (reset_state == VNIC_CLOSED) {
2102 		rc = 0;
2103 		goto out;
2104 	}
2105 
2106 	rc = __ibmvnic_open(netdev);
2107 	if (rc) {
2108 		rc = IBMVNIC_OPEN_FAILED;
2109 		goto out;
2110 	}
2111 
2112 	/* refresh device's multicast list */
2113 	ibmvnic_set_multi(netdev);
2114 
2115 	/* kick napi */
2116 	for (i = 0; i < adapter->req_rx_queues; i++)
2117 		napi_schedule(&adapter->napi[i]);
2118 
2119 	if (adapter->reset_reason == VNIC_RESET_FAILOVER ||
2120 	    adapter->reset_reason == VNIC_RESET_MOBILITY)
2121 		__netdev_notify_peers(netdev);
2122 
2123 	rc = 0;
2124 
2125 out:
2126 	/* restore the adapter state if reset failed */
2127 	if (rc)
2128 		adapter->state = reset_state;
2129 	/* requestor of VNIC_RESET_CHANGE_PARAM should still hold the rtnl lock */
2130 	if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
2131 		rtnl_unlock();
2132 
2133 	netdev_dbg(adapter->netdev, "[S:%d FOP:%d] Reset done, rc %d\n",
2134 		   adapter->state, adapter->failover_pending, rc);
2135 	return rc;
2136 }
2137 
2138 static int do_hard_reset(struct ibmvnic_adapter *adapter,
2139 			 struct ibmvnic_rwi *rwi, u32 reset_state)
2140 {
2141 	struct net_device *netdev = adapter->netdev;
2142 	int rc;
2143 
2144 	netdev_dbg(adapter->netdev, "Hard resetting driver (%d)\n",
2145 		   rwi->reset_reason);
2146 
2147 	/* read the state and check (again) after getting rtnl */
2148 	reset_state = adapter->state;
2149 
2150 	if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
2151 		rc = -EBUSY;
2152 		goto out;
2153 	}
2154 
2155 	netif_carrier_off(netdev);
2156 	adapter->reset_reason = rwi->reset_reason;
2157 
2158 	ibmvnic_cleanup(netdev);
2159 	release_resources(adapter);
2160 	release_sub_crqs(adapter, 0);
2161 	release_crq_queue(adapter);
2162 
2163 	/* remove the closed state so when we call open it appears
2164 	 * we are coming from the probed state.
2165 	 */
2166 	adapter->state = VNIC_PROBED;
2167 
2168 	reinit_completion(&adapter->init_done);
2169 	rc = init_crq_queue(adapter);
2170 	if (rc) {
2171 		netdev_err(adapter->netdev,
2172 			   "Couldn't initialize crq. rc=%d\n", rc);
2173 		goto out;
2174 	}
2175 
2176 	rc = ibmvnic_reset_init(adapter, false);
2177 	if (rc)
2178 		goto out;
2179 
2180 	/* If the adapter was in PROBE state prior to the reset,
2181 	 * exit here.
2182 	 */
2183 	if (reset_state == VNIC_PROBED)
2184 		goto out;
2185 
2186 	rc = ibmvnic_login(netdev);
2187 	if (rc)
2188 		goto out;
2189 
2190 	rc = init_resources(adapter);
2191 	if (rc)
2192 		goto out;
2193 
2194 	ibmvnic_disable_irqs(adapter);
2195 	adapter->state = VNIC_CLOSED;
2196 
2197 	if (reset_state == VNIC_CLOSED)
2198 		goto out;
2199 
2200 	rc = __ibmvnic_open(netdev);
2201 	if (rc) {
2202 		rc = IBMVNIC_OPEN_FAILED;
2203 		goto out;
2204 	}
2205 
2206 	__netdev_notify_peers(netdev);
2207 out:
2208 	/* restore adapter state if reset failed */
2209 	if (rc)
2210 		adapter->state = reset_state;
2211 	netdev_dbg(adapter->netdev, "[S:%d FOP:%d] Hard reset done, rc %d\n",
2212 		   adapter->state, adapter->failover_pending, rc);
2213 	return rc;
2214 }
2215 
2216 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter)
2217 {
2218 	struct ibmvnic_rwi *rwi;
2219 	unsigned long flags;
2220 
2221 	spin_lock_irqsave(&adapter->rwi_lock, flags);
2222 
2223 	if (!list_empty(&adapter->rwi_list)) {
2224 		rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi,
2225 				       list);
2226 		list_del(&rwi->list);
2227 	} else {
2228 		rwi = NULL;
2229 	}
2230 
2231 	spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2232 	return rwi;
2233 }
2234 
2235 static void __ibmvnic_reset(struct work_struct *work)
2236 {
2237 	struct ibmvnic_rwi *rwi;
2238 	struct ibmvnic_adapter *adapter;
2239 	bool saved_state = false;
2240 	unsigned long flags;
2241 	u32 reset_state;
2242 	int rc = 0;
2243 
2244 	adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset);
2245 
2246 	if (test_and_set_bit_lock(0, &adapter->resetting)) {
2247 		schedule_delayed_work(&adapter->ibmvnic_delayed_reset,
2248 				      IBMVNIC_RESET_DELAY);
2249 		return;
2250 	}
2251 
2252 	rwi = get_next_rwi(adapter);
2253 	while (rwi) {
2254 		spin_lock_irqsave(&adapter->state_lock, flags);
2255 
2256 		if (adapter->state == VNIC_REMOVING ||
2257 		    adapter->state == VNIC_REMOVED) {
2258 			spin_unlock_irqrestore(&adapter->state_lock, flags);
2259 			kfree(rwi);
2260 			rc = EBUSY;
2261 			break;
2262 		}
2263 
2264 		if (!saved_state) {
2265 			reset_state = adapter->state;
2266 			saved_state = true;
2267 		}
2268 		spin_unlock_irqrestore(&adapter->state_lock, flags);
2269 
2270 		if (adapter->force_reset_recovery) {
2271 			/* Since we are doing a hard reset now, clear the
2272 			 * failover_pending flag so we don't ignore any
2273 			 * future MOBILITY or other resets.
2274 			 */
2275 			adapter->failover_pending = false;
2276 
2277 			/* Transport event occurred during previous reset */
2278 			if (adapter->wait_for_reset) {
2279 				/* Previous was CHANGE_PARAM; caller locked */
2280 				adapter->force_reset_recovery = false;
2281 				rc = do_hard_reset(adapter, rwi, reset_state);
2282 			} else {
2283 				rtnl_lock();
2284 				adapter->force_reset_recovery = false;
2285 				rc = do_hard_reset(adapter, rwi, reset_state);
2286 				rtnl_unlock();
2287 			}
2288 			if (rc) {
2289 				/* give backing device time to settle down */
2290 				netdev_dbg(adapter->netdev,
2291 					   "[S:%d] Hard reset failed, waiting 60 secs\n",
2292 					   adapter->state);
2293 				set_current_state(TASK_UNINTERRUPTIBLE);
2294 				schedule_timeout(60 * HZ);
2295 			}
2296 		} else {
2297 			rc = do_reset(adapter, rwi, reset_state);
2298 		}
2299 		kfree(rwi);
2300 		adapter->last_reset_time = jiffies;
2301 
2302 		if (rc)
2303 			netdev_dbg(adapter->netdev, "Reset failed, rc=%d\n", rc);
2304 
2305 		rwi = get_next_rwi(adapter);
2306 
2307 		if (rwi && (rwi->reset_reason == VNIC_RESET_FAILOVER ||
2308 			    rwi->reset_reason == VNIC_RESET_MOBILITY))
2309 			adapter->force_reset_recovery = true;
2310 	}
2311 
2312 	if (adapter->wait_for_reset) {
2313 		adapter->reset_done_rc = rc;
2314 		complete(&adapter->reset_done);
2315 	}
2316 
2317 	clear_bit_unlock(0, &adapter->resetting);
2318 
2319 	netdev_dbg(adapter->netdev,
2320 		   "[S:%d FRR:%d WFR:%d] Done processing resets\n",
2321 		   adapter->state, adapter->force_reset_recovery,
2322 		   adapter->wait_for_reset);
2323 }
2324 
2325 static void __ibmvnic_delayed_reset(struct work_struct *work)
2326 {
2327 	struct ibmvnic_adapter *adapter;
2328 
2329 	adapter = container_of(work, struct ibmvnic_adapter,
2330 			       ibmvnic_delayed_reset.work);
2331 	__ibmvnic_reset(&adapter->ibmvnic_reset);
2332 }
2333 
2334 static int ibmvnic_reset(struct ibmvnic_adapter *adapter,
2335 			 enum ibmvnic_reset_reason reason)
2336 {
2337 	struct list_head *entry, *tmp_entry;
2338 	struct ibmvnic_rwi *rwi, *tmp;
2339 	struct net_device *netdev = adapter->netdev;
2340 	unsigned long flags;
2341 	int ret;
2342 
2343 	spin_lock_irqsave(&adapter->rwi_lock, flags);
2344 
2345 	/* If failover is pending don't schedule any other reset.
2346 	 * Instead let the failover complete. If there is already a
2347 	 * a failover reset scheduled, we will detect and drop the
2348 	 * duplicate reset when walking the ->rwi_list below.
2349 	 */
2350 	if (adapter->state == VNIC_REMOVING ||
2351 	    adapter->state == VNIC_REMOVED ||
2352 	    (adapter->failover_pending && reason != VNIC_RESET_FAILOVER)) {
2353 		ret = EBUSY;
2354 		netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n");
2355 		goto err;
2356 	}
2357 
2358 	if (adapter->state == VNIC_PROBING) {
2359 		netdev_warn(netdev, "Adapter reset during probe\n");
2360 		adapter->init_done_rc = EAGAIN;
2361 		ret = EAGAIN;
2362 		goto err;
2363 	}
2364 
2365 	list_for_each(entry, &adapter->rwi_list) {
2366 		tmp = list_entry(entry, struct ibmvnic_rwi, list);
2367 		if (tmp->reset_reason == reason) {
2368 			netdev_dbg(netdev, "Skipping matching reset, reason=%d\n",
2369 				   reason);
2370 			ret = EBUSY;
2371 			goto err;
2372 		}
2373 	}
2374 
2375 	rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC);
2376 	if (!rwi) {
2377 		ret = ENOMEM;
2378 		goto err;
2379 	}
2380 	/* if we just received a transport event,
2381 	 * flush reset queue and process this reset
2382 	 */
2383 	if (adapter->force_reset_recovery && !list_empty(&adapter->rwi_list)) {
2384 		list_for_each_safe(entry, tmp_entry, &adapter->rwi_list)
2385 			list_del(entry);
2386 	}
2387 	rwi->reset_reason = reason;
2388 	list_add_tail(&rwi->list, &adapter->rwi_list);
2389 	netdev_dbg(adapter->netdev, "Scheduling reset (reason %d)\n", reason);
2390 	schedule_work(&adapter->ibmvnic_reset);
2391 
2392 	ret = 0;
2393 err:
2394 	/* ibmvnic_close() below can block, so drop the lock first */
2395 	spin_unlock_irqrestore(&adapter->rwi_lock, flags);
2396 
2397 	if (ret == ENOMEM)
2398 		ibmvnic_close(netdev);
2399 
2400 	return -ret;
2401 }
2402 
2403 static void ibmvnic_tx_timeout(struct net_device *dev, unsigned int txqueue)
2404 {
2405 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
2406 
2407 	if (test_bit(0, &adapter->resetting)) {
2408 		netdev_err(adapter->netdev,
2409 			   "Adapter is resetting, skip timeout reset\n");
2410 		return;
2411 	}
2412 	/* No queuing up reset until at least 5 seconds (default watchdog val)
2413 	 * after last reset
2414 	 */
2415 	if (time_before(jiffies, (adapter->last_reset_time + dev->watchdog_timeo))) {
2416 		netdev_dbg(dev, "Not yet time to tx timeout.\n");
2417 		return;
2418 	}
2419 	ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT);
2420 }
2421 
2422 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter,
2423 				  struct ibmvnic_rx_buff *rx_buff)
2424 {
2425 	struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index];
2426 
2427 	rx_buff->skb = NULL;
2428 
2429 	pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff);
2430 	pool->next_alloc = (pool->next_alloc + 1) % pool->size;
2431 
2432 	atomic_dec(&pool->available);
2433 }
2434 
2435 static int ibmvnic_poll(struct napi_struct *napi, int budget)
2436 {
2437 	struct ibmvnic_sub_crq_queue *rx_scrq;
2438 	struct ibmvnic_adapter *adapter;
2439 	struct net_device *netdev;
2440 	int frames_processed;
2441 	int scrq_num;
2442 
2443 	netdev = napi->dev;
2444 	adapter = netdev_priv(netdev);
2445 	scrq_num = (int)(napi - adapter->napi);
2446 	frames_processed = 0;
2447 	rx_scrq = adapter->rx_scrq[scrq_num];
2448 
2449 restart_poll:
2450 	while (frames_processed < budget) {
2451 		struct sk_buff *skb;
2452 		struct ibmvnic_rx_buff *rx_buff;
2453 		union sub_crq *next;
2454 		u32 length;
2455 		u16 offset;
2456 		u8 flags = 0;
2457 
2458 		if (unlikely(test_bit(0, &adapter->resetting) &&
2459 			     adapter->reset_reason != VNIC_RESET_NON_FATAL)) {
2460 			enable_scrq_irq(adapter, rx_scrq);
2461 			napi_complete_done(napi, frames_processed);
2462 			return frames_processed;
2463 		}
2464 
2465 		if (!pending_scrq(adapter, rx_scrq))
2466 			break;
2467 		next = ibmvnic_next_scrq(adapter, rx_scrq);
2468 		rx_buff = (struct ibmvnic_rx_buff *)
2469 			  be64_to_cpu(next->rx_comp.correlator);
2470 		/* do error checking */
2471 		if (next->rx_comp.rc) {
2472 			netdev_dbg(netdev, "rx buffer returned with rc %x\n",
2473 				   be16_to_cpu(next->rx_comp.rc));
2474 			/* free the entry */
2475 			next->rx_comp.first = 0;
2476 			dev_kfree_skb_any(rx_buff->skb);
2477 			remove_buff_from_pool(adapter, rx_buff);
2478 			continue;
2479 		} else if (!rx_buff->skb) {
2480 			/* free the entry */
2481 			next->rx_comp.first = 0;
2482 			remove_buff_from_pool(adapter, rx_buff);
2483 			continue;
2484 		}
2485 
2486 		length = be32_to_cpu(next->rx_comp.len);
2487 		offset = be16_to_cpu(next->rx_comp.off_frame_data);
2488 		flags = next->rx_comp.flags;
2489 		skb = rx_buff->skb;
2490 		/* load long_term_buff before copying to skb */
2491 		dma_rmb();
2492 		skb_copy_to_linear_data(skb, rx_buff->data + offset,
2493 					length);
2494 
2495 		/* VLAN Header has been stripped by the system firmware and
2496 		 * needs to be inserted by the driver
2497 		 */
2498 		if (adapter->rx_vlan_header_insertion &&
2499 		    (flags & IBMVNIC_VLAN_STRIPPED))
2500 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
2501 					       ntohs(next->rx_comp.vlan_tci));
2502 
2503 		/* free the entry */
2504 		next->rx_comp.first = 0;
2505 		remove_buff_from_pool(adapter, rx_buff);
2506 
2507 		skb_put(skb, length);
2508 		skb->protocol = eth_type_trans(skb, netdev);
2509 		skb_record_rx_queue(skb, scrq_num);
2510 
2511 		if (flags & IBMVNIC_IP_CHKSUM_GOOD &&
2512 		    flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) {
2513 			skb->ip_summed = CHECKSUM_UNNECESSARY;
2514 		}
2515 
2516 		length = skb->len;
2517 		napi_gro_receive(napi, skb); /* send it up */
2518 		netdev->stats.rx_packets++;
2519 		netdev->stats.rx_bytes += length;
2520 		adapter->rx_stats_buffers[scrq_num].packets++;
2521 		adapter->rx_stats_buffers[scrq_num].bytes += length;
2522 		frames_processed++;
2523 	}
2524 
2525 	if (adapter->state != VNIC_CLOSING &&
2526 	    ((atomic_read(&adapter->rx_pool[scrq_num].available) <
2527 	      adapter->req_rx_add_entries_per_subcrq / 2) ||
2528 	      frames_processed < budget))
2529 		replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]);
2530 	if (frames_processed < budget) {
2531 		if (napi_complete_done(napi, frames_processed)) {
2532 			enable_scrq_irq(adapter, rx_scrq);
2533 			if (pending_scrq(adapter, rx_scrq)) {
2534 				if (napi_reschedule(napi)) {
2535 					disable_scrq_irq(adapter, rx_scrq);
2536 					goto restart_poll;
2537 				}
2538 			}
2539 		}
2540 	}
2541 	return frames_processed;
2542 }
2543 
2544 static int wait_for_reset(struct ibmvnic_adapter *adapter)
2545 {
2546 	int rc, ret;
2547 
2548 	adapter->fallback.mtu = adapter->req_mtu;
2549 	adapter->fallback.rx_queues = adapter->req_rx_queues;
2550 	adapter->fallback.tx_queues = adapter->req_tx_queues;
2551 	adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq;
2552 	adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq;
2553 
2554 	reinit_completion(&adapter->reset_done);
2555 	adapter->wait_for_reset = true;
2556 	rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2557 
2558 	if (rc) {
2559 		ret = rc;
2560 		goto out;
2561 	}
2562 	rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 60000);
2563 	if (rc) {
2564 		ret = -ENODEV;
2565 		goto out;
2566 	}
2567 
2568 	ret = 0;
2569 	if (adapter->reset_done_rc) {
2570 		ret = -EIO;
2571 		adapter->desired.mtu = adapter->fallback.mtu;
2572 		adapter->desired.rx_queues = adapter->fallback.rx_queues;
2573 		adapter->desired.tx_queues = adapter->fallback.tx_queues;
2574 		adapter->desired.rx_entries = adapter->fallback.rx_entries;
2575 		adapter->desired.tx_entries = adapter->fallback.tx_entries;
2576 
2577 		reinit_completion(&adapter->reset_done);
2578 		adapter->wait_for_reset = true;
2579 		rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
2580 		if (rc) {
2581 			ret = rc;
2582 			goto out;
2583 		}
2584 		rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done,
2585 						 60000);
2586 		if (rc) {
2587 			ret = -ENODEV;
2588 			goto out;
2589 		}
2590 	}
2591 out:
2592 	adapter->wait_for_reset = false;
2593 
2594 	return ret;
2595 }
2596 
2597 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu)
2598 {
2599 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2600 
2601 	adapter->desired.mtu = new_mtu + ETH_HLEN;
2602 
2603 	return wait_for_reset(adapter);
2604 }
2605 
2606 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb,
2607 						struct net_device *dev,
2608 						netdev_features_t features)
2609 {
2610 	/* Some backing hardware adapters can not
2611 	 * handle packets with a MSS less than 224
2612 	 * or with only one segment.
2613 	 */
2614 	if (skb_is_gso(skb)) {
2615 		if (skb_shinfo(skb)->gso_size < 224 ||
2616 		    skb_shinfo(skb)->gso_segs == 1)
2617 			features &= ~NETIF_F_GSO_MASK;
2618 	}
2619 
2620 	return features;
2621 }
2622 
2623 static const struct net_device_ops ibmvnic_netdev_ops = {
2624 	.ndo_open		= ibmvnic_open,
2625 	.ndo_stop		= ibmvnic_close,
2626 	.ndo_start_xmit		= ibmvnic_xmit,
2627 	.ndo_set_rx_mode	= ibmvnic_set_multi,
2628 	.ndo_set_mac_address	= ibmvnic_set_mac,
2629 	.ndo_validate_addr	= eth_validate_addr,
2630 	.ndo_tx_timeout		= ibmvnic_tx_timeout,
2631 	.ndo_change_mtu		= ibmvnic_change_mtu,
2632 	.ndo_features_check     = ibmvnic_features_check,
2633 };
2634 
2635 /* ethtool functions */
2636 
2637 static int ibmvnic_get_link_ksettings(struct net_device *netdev,
2638 				      struct ethtool_link_ksettings *cmd)
2639 {
2640 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2641 	int rc;
2642 
2643 	rc = send_query_phys_parms(adapter);
2644 	if (rc) {
2645 		adapter->speed = SPEED_UNKNOWN;
2646 		adapter->duplex = DUPLEX_UNKNOWN;
2647 	}
2648 	cmd->base.speed = adapter->speed;
2649 	cmd->base.duplex = adapter->duplex;
2650 	cmd->base.port = PORT_FIBRE;
2651 	cmd->base.phy_address = 0;
2652 	cmd->base.autoneg = AUTONEG_ENABLE;
2653 
2654 	return 0;
2655 }
2656 
2657 static void ibmvnic_get_drvinfo(struct net_device *netdev,
2658 				struct ethtool_drvinfo *info)
2659 {
2660 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2661 
2662 	strscpy(info->driver, ibmvnic_driver_name, sizeof(info->driver));
2663 	strscpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version));
2664 	strscpy(info->fw_version, adapter->fw_version,
2665 		sizeof(info->fw_version));
2666 }
2667 
2668 static u32 ibmvnic_get_msglevel(struct net_device *netdev)
2669 {
2670 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2671 
2672 	return adapter->msg_enable;
2673 }
2674 
2675 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data)
2676 {
2677 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2678 
2679 	adapter->msg_enable = data;
2680 }
2681 
2682 static u32 ibmvnic_get_link(struct net_device *netdev)
2683 {
2684 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2685 
2686 	/* Don't need to send a query because we request a logical link up at
2687 	 * init and then we wait for link state indications
2688 	 */
2689 	return adapter->logical_link_state;
2690 }
2691 
2692 static void ibmvnic_get_ringparam(struct net_device *netdev,
2693 				  struct ethtool_ringparam *ring)
2694 {
2695 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2696 
2697 	if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
2698 		ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq;
2699 		ring->tx_max_pending = adapter->max_tx_entries_per_subcrq;
2700 	} else {
2701 		ring->rx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
2702 		ring->tx_max_pending = IBMVNIC_MAX_QUEUE_SZ;
2703 	}
2704 	ring->rx_mini_max_pending = 0;
2705 	ring->rx_jumbo_max_pending = 0;
2706 	ring->rx_pending = adapter->req_rx_add_entries_per_subcrq;
2707 	ring->tx_pending = adapter->req_tx_entries_per_subcrq;
2708 	ring->rx_mini_pending = 0;
2709 	ring->rx_jumbo_pending = 0;
2710 }
2711 
2712 static int ibmvnic_set_ringparam(struct net_device *netdev,
2713 				 struct ethtool_ringparam *ring)
2714 {
2715 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2716 	int ret;
2717 
2718 	ret = 0;
2719 	adapter->desired.rx_entries = ring->rx_pending;
2720 	adapter->desired.tx_entries = ring->tx_pending;
2721 
2722 	ret = wait_for_reset(adapter);
2723 
2724 	if (!ret &&
2725 	    (adapter->req_rx_add_entries_per_subcrq != ring->rx_pending ||
2726 	     adapter->req_tx_entries_per_subcrq != ring->tx_pending))
2727 		netdev_info(netdev,
2728 			    "Could not match full ringsize request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
2729 			    ring->rx_pending, ring->tx_pending,
2730 			    adapter->req_rx_add_entries_per_subcrq,
2731 			    adapter->req_tx_entries_per_subcrq);
2732 	return ret;
2733 }
2734 
2735 static void ibmvnic_get_channels(struct net_device *netdev,
2736 				 struct ethtool_channels *channels)
2737 {
2738 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2739 
2740 	if (adapter->priv_flags & IBMVNIC_USE_SERVER_MAXES) {
2741 		channels->max_rx = adapter->max_rx_queues;
2742 		channels->max_tx = adapter->max_tx_queues;
2743 	} else {
2744 		channels->max_rx = IBMVNIC_MAX_QUEUES;
2745 		channels->max_tx = IBMVNIC_MAX_QUEUES;
2746 	}
2747 
2748 	channels->max_other = 0;
2749 	channels->max_combined = 0;
2750 	channels->rx_count = adapter->req_rx_queues;
2751 	channels->tx_count = adapter->req_tx_queues;
2752 	channels->other_count = 0;
2753 	channels->combined_count = 0;
2754 }
2755 
2756 static int ibmvnic_set_channels(struct net_device *netdev,
2757 				struct ethtool_channels *channels)
2758 {
2759 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2760 	int ret;
2761 
2762 	ret = 0;
2763 	adapter->desired.rx_queues = channels->rx_count;
2764 	adapter->desired.tx_queues = channels->tx_count;
2765 
2766 	ret = wait_for_reset(adapter);
2767 
2768 	if (!ret &&
2769 	    (adapter->req_rx_queues != channels->rx_count ||
2770 	     adapter->req_tx_queues != channels->tx_count))
2771 		netdev_info(netdev,
2772 			    "Could not match full channels request. Requested: RX %d, TX %d; Allowed: RX %llu, TX %llu\n",
2773 			    channels->rx_count, channels->tx_count,
2774 			    adapter->req_rx_queues, adapter->req_tx_queues);
2775 	return ret;
2776 }
2777 
2778 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data)
2779 {
2780 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
2781 	int i;
2782 
2783 	switch (stringset) {
2784 	case ETH_SS_STATS:
2785 		for (i = 0; i < ARRAY_SIZE(ibmvnic_stats);
2786 				i++, data += ETH_GSTRING_LEN)
2787 			memcpy(data, ibmvnic_stats[i].name, ETH_GSTRING_LEN);
2788 
2789 		for (i = 0; i < adapter->req_tx_queues; i++) {
2790 			snprintf(data, ETH_GSTRING_LEN, "tx%d_packets", i);
2791 			data += ETH_GSTRING_LEN;
2792 
2793 			snprintf(data, ETH_GSTRING_LEN, "tx%d_bytes", i);
2794 			data += ETH_GSTRING_LEN;
2795 
2796 			snprintf(data, ETH_GSTRING_LEN,
2797 				 "tx%d_dropped_packets", i);
2798 			data += ETH_GSTRING_LEN;
2799 		}
2800 
2801 		for (i = 0; i < adapter->req_rx_queues; i++) {
2802 			snprintf(data, ETH_GSTRING_LEN, "rx%d_packets", i);
2803 			data += ETH_GSTRING_LEN;
2804 
2805 			snprintf(data, ETH_GSTRING_LEN, "rx%d_bytes", i);
2806 			data += ETH_GSTRING_LEN;
2807 
2808 			snprintf(data, ETH_GSTRING_LEN, "rx%d_interrupts", i);
2809 			data += ETH_GSTRING_LEN;
2810 		}
2811 		break;
2812 
2813 	case ETH_SS_PRIV_FLAGS:
2814 		for (i = 0; i < ARRAY_SIZE(ibmvnic_priv_flags); i++)
2815 			strcpy(data + i * ETH_GSTRING_LEN,
2816 			       ibmvnic_priv_flags[i]);
2817 		break;
2818 	default:
2819 		return;
2820 	}
2821 }
2822 
2823 static int ibmvnic_get_sset_count(struct net_device *dev, int sset)
2824 {
2825 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
2826 
2827 	switch (sset) {
2828 	case ETH_SS_STATS:
2829 		return ARRAY_SIZE(ibmvnic_stats) +
2830 		       adapter->req_tx_queues * NUM_TX_STATS +
2831 		       adapter->req_rx_queues * NUM_RX_STATS;
2832 	case ETH_SS_PRIV_FLAGS:
2833 		return ARRAY_SIZE(ibmvnic_priv_flags);
2834 	default:
2835 		return -EOPNOTSUPP;
2836 	}
2837 }
2838 
2839 static void ibmvnic_get_ethtool_stats(struct net_device *dev,
2840 				      struct ethtool_stats *stats, u64 *data)
2841 {
2842 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
2843 	union ibmvnic_crq crq;
2844 	int i, j;
2845 	int rc;
2846 
2847 	memset(&crq, 0, sizeof(crq));
2848 	crq.request_statistics.first = IBMVNIC_CRQ_CMD;
2849 	crq.request_statistics.cmd = REQUEST_STATISTICS;
2850 	crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token);
2851 	crq.request_statistics.len =
2852 	    cpu_to_be32(sizeof(struct ibmvnic_statistics));
2853 
2854 	/* Wait for data to be written */
2855 	reinit_completion(&adapter->stats_done);
2856 	rc = ibmvnic_send_crq(adapter, &crq);
2857 	if (rc)
2858 		return;
2859 	rc = ibmvnic_wait_for_completion(adapter, &adapter->stats_done, 10000);
2860 	if (rc)
2861 		return;
2862 
2863 	for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++)
2864 		data[i] = be64_to_cpu(IBMVNIC_GET_STAT
2865 				      (adapter, ibmvnic_stats[i].offset));
2866 
2867 	for (j = 0; j < adapter->req_tx_queues; j++) {
2868 		data[i] = adapter->tx_stats_buffers[j].packets;
2869 		i++;
2870 		data[i] = adapter->tx_stats_buffers[j].bytes;
2871 		i++;
2872 		data[i] = adapter->tx_stats_buffers[j].dropped_packets;
2873 		i++;
2874 	}
2875 
2876 	for (j = 0; j < adapter->req_rx_queues; j++) {
2877 		data[i] = adapter->rx_stats_buffers[j].packets;
2878 		i++;
2879 		data[i] = adapter->rx_stats_buffers[j].bytes;
2880 		i++;
2881 		data[i] = adapter->rx_stats_buffers[j].interrupts;
2882 		i++;
2883 	}
2884 }
2885 
2886 static u32 ibmvnic_get_priv_flags(struct net_device *netdev)
2887 {
2888 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2889 
2890 	return adapter->priv_flags;
2891 }
2892 
2893 static int ibmvnic_set_priv_flags(struct net_device *netdev, u32 flags)
2894 {
2895 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2896 	bool which_maxes = !!(flags & IBMVNIC_USE_SERVER_MAXES);
2897 
2898 	if (which_maxes)
2899 		adapter->priv_flags |= IBMVNIC_USE_SERVER_MAXES;
2900 	else
2901 		adapter->priv_flags &= ~IBMVNIC_USE_SERVER_MAXES;
2902 
2903 	return 0;
2904 }
2905 
2906 static const struct ethtool_ops ibmvnic_ethtool_ops = {
2907 	.get_drvinfo		= ibmvnic_get_drvinfo,
2908 	.get_msglevel		= ibmvnic_get_msglevel,
2909 	.set_msglevel		= ibmvnic_set_msglevel,
2910 	.get_link		= ibmvnic_get_link,
2911 	.get_ringparam		= ibmvnic_get_ringparam,
2912 	.set_ringparam		= ibmvnic_set_ringparam,
2913 	.get_channels		= ibmvnic_get_channels,
2914 	.set_channels		= ibmvnic_set_channels,
2915 	.get_strings            = ibmvnic_get_strings,
2916 	.get_sset_count         = ibmvnic_get_sset_count,
2917 	.get_ethtool_stats	= ibmvnic_get_ethtool_stats,
2918 	.get_link_ksettings	= ibmvnic_get_link_ksettings,
2919 	.get_priv_flags		= ibmvnic_get_priv_flags,
2920 	.set_priv_flags		= ibmvnic_set_priv_flags,
2921 };
2922 
2923 /* Routines for managing CRQs/sCRQs  */
2924 
2925 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter,
2926 				   struct ibmvnic_sub_crq_queue *scrq)
2927 {
2928 	int rc;
2929 
2930 	if (!scrq) {
2931 		netdev_dbg(adapter->netdev, "Invalid scrq reset.\n");
2932 		return -EINVAL;
2933 	}
2934 
2935 	if (scrq->irq) {
2936 		free_irq(scrq->irq, scrq);
2937 		irq_dispose_mapping(scrq->irq);
2938 		scrq->irq = 0;
2939 	}
2940 
2941 	if (scrq->msgs) {
2942 		memset(scrq->msgs, 0, 4 * PAGE_SIZE);
2943 		atomic_set(&scrq->used, 0);
2944 		scrq->cur = 0;
2945 		scrq->ind_buf.index = 0;
2946 	} else {
2947 		netdev_dbg(adapter->netdev, "Invalid scrq reset\n");
2948 		return -EINVAL;
2949 	}
2950 
2951 	rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
2952 			   4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
2953 	return rc;
2954 }
2955 
2956 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter)
2957 {
2958 	int i, rc;
2959 
2960 	if (!adapter->tx_scrq || !adapter->rx_scrq)
2961 		return -EINVAL;
2962 
2963 	for (i = 0; i < adapter->req_tx_queues; i++) {
2964 		netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i);
2965 		rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]);
2966 		if (rc)
2967 			return rc;
2968 	}
2969 
2970 	for (i = 0; i < adapter->req_rx_queues; i++) {
2971 		netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i);
2972 		rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]);
2973 		if (rc)
2974 			return rc;
2975 	}
2976 
2977 	return rc;
2978 }
2979 
2980 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter,
2981 				  struct ibmvnic_sub_crq_queue *scrq,
2982 				  bool do_h_free)
2983 {
2984 	struct device *dev = &adapter->vdev->dev;
2985 	long rc;
2986 
2987 	netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n");
2988 
2989 	if (do_h_free) {
2990 		/* Close the sub-crqs */
2991 		do {
2992 			rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
2993 						adapter->vdev->unit_address,
2994 						scrq->crq_num);
2995 		} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
2996 
2997 		if (rc) {
2998 			netdev_err(adapter->netdev,
2999 				   "Failed to release sub-CRQ %16lx, rc = %ld\n",
3000 				   scrq->crq_num, rc);
3001 		}
3002 	}
3003 
3004 	dma_free_coherent(dev,
3005 			  IBMVNIC_IND_ARR_SZ,
3006 			  scrq->ind_buf.indir_arr,
3007 			  scrq->ind_buf.indir_dma);
3008 
3009 	dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3010 			 DMA_BIDIRECTIONAL);
3011 	free_pages((unsigned long)scrq->msgs, 2);
3012 	kfree(scrq);
3013 }
3014 
3015 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter
3016 							*adapter)
3017 {
3018 	struct device *dev = &adapter->vdev->dev;
3019 	struct ibmvnic_sub_crq_queue *scrq;
3020 	int rc;
3021 
3022 	scrq = kzalloc(sizeof(*scrq), GFP_KERNEL);
3023 	if (!scrq)
3024 		return NULL;
3025 
3026 	scrq->msgs =
3027 		(union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2);
3028 	if (!scrq->msgs) {
3029 		dev_warn(dev, "Couldn't allocate crq queue messages page\n");
3030 		goto zero_page_failed;
3031 	}
3032 
3033 	scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE,
3034 					 DMA_BIDIRECTIONAL);
3035 	if (dma_mapping_error(dev, scrq->msg_token)) {
3036 		dev_warn(dev, "Couldn't map crq queue messages page\n");
3037 		goto map_failed;
3038 	}
3039 
3040 	rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
3041 			   4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
3042 
3043 	if (rc == H_RESOURCE)
3044 		rc = ibmvnic_reset_crq(adapter);
3045 
3046 	if (rc == H_CLOSED) {
3047 		dev_warn(dev, "Partner adapter not ready, waiting.\n");
3048 	} else if (rc) {
3049 		dev_warn(dev, "Error %d registering sub-crq\n", rc);
3050 		goto reg_failed;
3051 	}
3052 
3053 	scrq->adapter = adapter;
3054 	scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs);
3055 	scrq->ind_buf.index = 0;
3056 
3057 	scrq->ind_buf.indir_arr =
3058 		dma_alloc_coherent(dev,
3059 				   IBMVNIC_IND_ARR_SZ,
3060 				   &scrq->ind_buf.indir_dma,
3061 				   GFP_KERNEL);
3062 
3063 	if (!scrq->ind_buf.indir_arr)
3064 		goto indir_failed;
3065 
3066 	spin_lock_init(&scrq->lock);
3067 
3068 	netdev_dbg(adapter->netdev,
3069 		   "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n",
3070 		   scrq->crq_num, scrq->hw_irq, scrq->irq);
3071 
3072 	return scrq;
3073 
3074 indir_failed:
3075 	do {
3076 		rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3077 					adapter->vdev->unit_address,
3078 					scrq->crq_num);
3079 	} while (rc == H_BUSY || rc == H_IS_LONG_BUSY(rc));
3080 reg_failed:
3081 	dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3082 			 DMA_BIDIRECTIONAL);
3083 map_failed:
3084 	free_pages((unsigned long)scrq->msgs, 2);
3085 zero_page_failed:
3086 	kfree(scrq);
3087 
3088 	return NULL;
3089 }
3090 
3091 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free)
3092 {
3093 	int i;
3094 
3095 	if (adapter->tx_scrq) {
3096 		for (i = 0; i < adapter->num_active_tx_scrqs; i++) {
3097 			if (!adapter->tx_scrq[i])
3098 				continue;
3099 
3100 			netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n",
3101 				   i);
3102 			if (adapter->tx_scrq[i]->irq) {
3103 				free_irq(adapter->tx_scrq[i]->irq,
3104 					 adapter->tx_scrq[i]);
3105 				irq_dispose_mapping(adapter->tx_scrq[i]->irq);
3106 				adapter->tx_scrq[i]->irq = 0;
3107 			}
3108 
3109 			release_sub_crq_queue(adapter, adapter->tx_scrq[i],
3110 					      do_h_free);
3111 		}
3112 
3113 		kfree(adapter->tx_scrq);
3114 		adapter->tx_scrq = NULL;
3115 		adapter->num_active_tx_scrqs = 0;
3116 	}
3117 
3118 	if (adapter->rx_scrq) {
3119 		for (i = 0; i < adapter->num_active_rx_scrqs; i++) {
3120 			if (!adapter->rx_scrq[i])
3121 				continue;
3122 
3123 			netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n",
3124 				   i);
3125 			if (adapter->rx_scrq[i]->irq) {
3126 				free_irq(adapter->rx_scrq[i]->irq,
3127 					 adapter->rx_scrq[i]);
3128 				irq_dispose_mapping(adapter->rx_scrq[i]->irq);
3129 				adapter->rx_scrq[i]->irq = 0;
3130 			}
3131 
3132 			release_sub_crq_queue(adapter, adapter->rx_scrq[i],
3133 					      do_h_free);
3134 		}
3135 
3136 		kfree(adapter->rx_scrq);
3137 		adapter->rx_scrq = NULL;
3138 		adapter->num_active_rx_scrqs = 0;
3139 	}
3140 }
3141 
3142 static int disable_scrq_irq(struct ibmvnic_adapter *adapter,
3143 			    struct ibmvnic_sub_crq_queue *scrq)
3144 {
3145 	struct device *dev = &adapter->vdev->dev;
3146 	unsigned long rc;
3147 
3148 	rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3149 				H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3150 	if (rc)
3151 		dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n",
3152 			scrq->hw_irq, rc);
3153 	return rc;
3154 }
3155 
3156 static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
3157 			   struct ibmvnic_sub_crq_queue *scrq)
3158 {
3159 	struct device *dev = &adapter->vdev->dev;
3160 	unsigned long rc;
3161 
3162 	if (scrq->hw_irq > 0x100000000ULL) {
3163 		dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq);
3164 		return 1;
3165 	}
3166 
3167 	if (test_bit(0, &adapter->resetting) &&
3168 	    adapter->reset_reason == VNIC_RESET_MOBILITY) {
3169 		u64 val = (0xff000000) | scrq->hw_irq;
3170 
3171 		rc = plpar_hcall_norets(H_EOI, val);
3172 		/* H_EOI would fail with rc = H_FUNCTION when running
3173 		 * in XIVE mode which is expected, but not an error.
3174 		 */
3175 		if (rc && rc != H_FUNCTION)
3176 			dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n",
3177 				val, rc);
3178 	}
3179 
3180 	rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
3181 				H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
3182 	if (rc)
3183 		dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n",
3184 			scrq->hw_irq, rc);
3185 	return rc;
3186 }
3187 
3188 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter,
3189 			       struct ibmvnic_sub_crq_queue *scrq)
3190 {
3191 	struct device *dev = &adapter->vdev->dev;
3192 	struct ibmvnic_tx_pool *tx_pool;
3193 	struct ibmvnic_tx_buff *txbuff;
3194 	struct netdev_queue *txq;
3195 	union sub_crq *next;
3196 	int index;
3197 	int i;
3198 
3199 restart_loop:
3200 	while (pending_scrq(adapter, scrq)) {
3201 		unsigned int pool = scrq->pool_index;
3202 		int num_entries = 0;
3203 		int total_bytes = 0;
3204 		int num_packets = 0;
3205 
3206 		next = ibmvnic_next_scrq(adapter, scrq);
3207 		for (i = 0; i < next->tx_comp.num_comps; i++) {
3208 			if (next->tx_comp.rcs[i])
3209 				dev_err(dev, "tx error %x\n",
3210 					next->tx_comp.rcs[i]);
3211 			index = be32_to_cpu(next->tx_comp.correlators[i]);
3212 			if (index & IBMVNIC_TSO_POOL_MASK) {
3213 				tx_pool = &adapter->tso_pool[pool];
3214 				index &= ~IBMVNIC_TSO_POOL_MASK;
3215 			} else {
3216 				tx_pool = &adapter->tx_pool[pool];
3217 			}
3218 
3219 			txbuff = &tx_pool->tx_buff[index];
3220 			num_packets++;
3221 			num_entries += txbuff->num_entries;
3222 			if (txbuff->skb) {
3223 				total_bytes += txbuff->skb->len;
3224 				dev_consume_skb_irq(txbuff->skb);
3225 				txbuff->skb = NULL;
3226 			} else {
3227 				netdev_warn(adapter->netdev,
3228 					    "TX completion received with NULL socket buffer\n");
3229 			}
3230 			tx_pool->free_map[tx_pool->producer_index] = index;
3231 			tx_pool->producer_index =
3232 				(tx_pool->producer_index + 1) %
3233 					tx_pool->num_buffers;
3234 		}
3235 		/* remove tx_comp scrq*/
3236 		next->tx_comp.first = 0;
3237 
3238 		txq = netdev_get_tx_queue(adapter->netdev, scrq->pool_index);
3239 		netdev_tx_completed_queue(txq, num_packets, total_bytes);
3240 
3241 		if (atomic_sub_return(num_entries, &scrq->used) <=
3242 		    (adapter->req_tx_entries_per_subcrq / 2) &&
3243 		    __netif_subqueue_stopped(adapter->netdev,
3244 					     scrq->pool_index)) {
3245 			netif_wake_subqueue(adapter->netdev, scrq->pool_index);
3246 			netdev_dbg(adapter->netdev, "Started queue %d\n",
3247 				   scrq->pool_index);
3248 		}
3249 	}
3250 
3251 	enable_scrq_irq(adapter, scrq);
3252 
3253 	if (pending_scrq(adapter, scrq)) {
3254 		disable_scrq_irq(adapter, scrq);
3255 		goto restart_loop;
3256 	}
3257 
3258 	return 0;
3259 }
3260 
3261 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance)
3262 {
3263 	struct ibmvnic_sub_crq_queue *scrq = instance;
3264 	struct ibmvnic_adapter *adapter = scrq->adapter;
3265 
3266 	disable_scrq_irq(adapter, scrq);
3267 	ibmvnic_complete_tx(adapter, scrq);
3268 
3269 	return IRQ_HANDLED;
3270 }
3271 
3272 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance)
3273 {
3274 	struct ibmvnic_sub_crq_queue *scrq = instance;
3275 	struct ibmvnic_adapter *adapter = scrq->adapter;
3276 
3277 	/* When booting a kdump kernel we can hit pending interrupts
3278 	 * prior to completing driver initialization.
3279 	 */
3280 	if (unlikely(adapter->state != VNIC_OPEN))
3281 		return IRQ_NONE;
3282 
3283 	adapter->rx_stats_buffers[scrq->scrq_num].interrupts++;
3284 
3285 	if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) {
3286 		disable_scrq_irq(adapter, scrq);
3287 		__napi_schedule(&adapter->napi[scrq->scrq_num]);
3288 	}
3289 
3290 	return IRQ_HANDLED;
3291 }
3292 
3293 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter)
3294 {
3295 	struct device *dev = &adapter->vdev->dev;
3296 	struct ibmvnic_sub_crq_queue *scrq;
3297 	int i = 0, j = 0;
3298 	int rc = 0;
3299 
3300 	for (i = 0; i < adapter->req_tx_queues; i++) {
3301 		netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n",
3302 			   i);
3303 		scrq = adapter->tx_scrq[i];
3304 		scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3305 
3306 		if (!scrq->irq) {
3307 			rc = -EINVAL;
3308 			dev_err(dev, "Error mapping irq\n");
3309 			goto req_tx_irq_failed;
3310 		}
3311 
3312 		snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-tx%d",
3313 			 adapter->vdev->unit_address, i);
3314 		rc = request_irq(scrq->irq, ibmvnic_interrupt_tx,
3315 				 0, scrq->name, scrq);
3316 
3317 		if (rc) {
3318 			dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n",
3319 				scrq->irq, rc);
3320 			irq_dispose_mapping(scrq->irq);
3321 			goto req_tx_irq_failed;
3322 		}
3323 	}
3324 
3325 	for (i = 0; i < adapter->req_rx_queues; i++) {
3326 		netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n",
3327 			   i);
3328 		scrq = adapter->rx_scrq[i];
3329 		scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
3330 		if (!scrq->irq) {
3331 			rc = -EINVAL;
3332 			dev_err(dev, "Error mapping irq\n");
3333 			goto req_rx_irq_failed;
3334 		}
3335 		snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-rx%d",
3336 			 adapter->vdev->unit_address, i);
3337 		rc = request_irq(scrq->irq, ibmvnic_interrupt_rx,
3338 				 0, scrq->name, scrq);
3339 		if (rc) {
3340 			dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n",
3341 				scrq->irq, rc);
3342 			irq_dispose_mapping(scrq->irq);
3343 			goto req_rx_irq_failed;
3344 		}
3345 	}
3346 	return rc;
3347 
3348 req_rx_irq_failed:
3349 	for (j = 0; j < i; j++) {
3350 		free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]);
3351 		irq_dispose_mapping(adapter->rx_scrq[j]->irq);
3352 	}
3353 	i = adapter->req_tx_queues;
3354 req_tx_irq_failed:
3355 	for (j = 0; j < i; j++) {
3356 		free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]);
3357 		irq_dispose_mapping(adapter->tx_scrq[j]->irq);
3358 	}
3359 	release_sub_crqs(adapter, 1);
3360 	return rc;
3361 }
3362 
3363 static int init_sub_crqs(struct ibmvnic_adapter *adapter)
3364 {
3365 	struct device *dev = &adapter->vdev->dev;
3366 	struct ibmvnic_sub_crq_queue **allqueues;
3367 	int registered_queues = 0;
3368 	int total_queues;
3369 	int more = 0;
3370 	int i;
3371 
3372 	total_queues = adapter->req_tx_queues + adapter->req_rx_queues;
3373 
3374 	allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL);
3375 	if (!allqueues)
3376 		return -1;
3377 
3378 	for (i = 0; i < total_queues; i++) {
3379 		allqueues[i] = init_sub_crq_queue(adapter);
3380 		if (!allqueues[i]) {
3381 			dev_warn(dev, "Couldn't allocate all sub-crqs\n");
3382 			break;
3383 		}
3384 		registered_queues++;
3385 	}
3386 
3387 	/* Make sure we were able to register the minimum number of queues */
3388 	if (registered_queues <
3389 	    adapter->min_tx_queues + adapter->min_rx_queues) {
3390 		dev_err(dev, "Fatal: Couldn't init  min number of sub-crqs\n");
3391 		goto tx_failed;
3392 	}
3393 
3394 	/* Distribute the failed allocated queues*/
3395 	for (i = 0; i < total_queues - registered_queues + more ; i++) {
3396 		netdev_dbg(adapter->netdev, "Reducing number of queues\n");
3397 		switch (i % 3) {
3398 		case 0:
3399 			if (adapter->req_rx_queues > adapter->min_rx_queues)
3400 				adapter->req_rx_queues--;
3401 			else
3402 				more++;
3403 			break;
3404 		case 1:
3405 			if (adapter->req_tx_queues > adapter->min_tx_queues)
3406 				adapter->req_tx_queues--;
3407 			else
3408 				more++;
3409 			break;
3410 		}
3411 	}
3412 
3413 	adapter->tx_scrq = kcalloc(adapter->req_tx_queues,
3414 				   sizeof(*adapter->tx_scrq), GFP_KERNEL);
3415 	if (!adapter->tx_scrq)
3416 		goto tx_failed;
3417 
3418 	for (i = 0; i < adapter->req_tx_queues; i++) {
3419 		adapter->tx_scrq[i] = allqueues[i];
3420 		adapter->tx_scrq[i]->pool_index = i;
3421 		adapter->num_active_tx_scrqs++;
3422 	}
3423 
3424 	adapter->rx_scrq = kcalloc(adapter->req_rx_queues,
3425 				   sizeof(*adapter->rx_scrq), GFP_KERNEL);
3426 	if (!adapter->rx_scrq)
3427 		goto rx_failed;
3428 
3429 	for (i = 0; i < adapter->req_rx_queues; i++) {
3430 		adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues];
3431 		adapter->rx_scrq[i]->scrq_num = i;
3432 		adapter->num_active_rx_scrqs++;
3433 	}
3434 
3435 	kfree(allqueues);
3436 	return 0;
3437 
3438 rx_failed:
3439 	kfree(adapter->tx_scrq);
3440 	adapter->tx_scrq = NULL;
3441 tx_failed:
3442 	for (i = 0; i < registered_queues; i++)
3443 		release_sub_crq_queue(adapter, allqueues[i], 1);
3444 	kfree(allqueues);
3445 	return -1;
3446 }
3447 
3448 static void send_request_cap(struct ibmvnic_adapter *adapter, int retry)
3449 {
3450 	struct device *dev = &adapter->vdev->dev;
3451 	union ibmvnic_crq crq;
3452 	int max_entries;
3453 
3454 	if (!retry) {
3455 		/* Sub-CRQ entries are 32 byte long */
3456 		int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4);
3457 
3458 		if (adapter->min_tx_entries_per_subcrq > entries_page ||
3459 		    adapter->min_rx_add_entries_per_subcrq > entries_page) {
3460 			dev_err(dev, "Fatal, invalid entries per sub-crq\n");
3461 			return;
3462 		}
3463 
3464 		if (adapter->desired.mtu)
3465 			adapter->req_mtu = adapter->desired.mtu;
3466 		else
3467 			adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN;
3468 
3469 		if (!adapter->desired.tx_entries)
3470 			adapter->desired.tx_entries =
3471 					adapter->max_tx_entries_per_subcrq;
3472 		if (!adapter->desired.rx_entries)
3473 			adapter->desired.rx_entries =
3474 					adapter->max_rx_add_entries_per_subcrq;
3475 
3476 		max_entries = IBMVNIC_MAX_LTB_SIZE /
3477 			      (adapter->req_mtu + IBMVNIC_BUFFER_HLEN);
3478 
3479 		if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3480 			adapter->desired.tx_entries > IBMVNIC_MAX_LTB_SIZE) {
3481 			adapter->desired.tx_entries = max_entries;
3482 		}
3483 
3484 		if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
3485 			adapter->desired.rx_entries > IBMVNIC_MAX_LTB_SIZE) {
3486 			adapter->desired.rx_entries = max_entries;
3487 		}
3488 
3489 		if (adapter->desired.tx_entries)
3490 			adapter->req_tx_entries_per_subcrq =
3491 					adapter->desired.tx_entries;
3492 		else
3493 			adapter->req_tx_entries_per_subcrq =
3494 					adapter->max_tx_entries_per_subcrq;
3495 
3496 		if (adapter->desired.rx_entries)
3497 			adapter->req_rx_add_entries_per_subcrq =
3498 					adapter->desired.rx_entries;
3499 		else
3500 			adapter->req_rx_add_entries_per_subcrq =
3501 					adapter->max_rx_add_entries_per_subcrq;
3502 
3503 		if (adapter->desired.tx_queues)
3504 			adapter->req_tx_queues =
3505 					adapter->desired.tx_queues;
3506 		else
3507 			adapter->req_tx_queues =
3508 					adapter->opt_tx_comp_sub_queues;
3509 
3510 		if (adapter->desired.rx_queues)
3511 			adapter->req_rx_queues =
3512 					adapter->desired.rx_queues;
3513 		else
3514 			adapter->req_rx_queues =
3515 					adapter->opt_rx_comp_queues;
3516 
3517 		adapter->req_rx_add_queues = adapter->max_rx_add_queues;
3518 	}
3519 
3520 	memset(&crq, 0, sizeof(crq));
3521 	crq.request_capability.first = IBMVNIC_CRQ_CMD;
3522 	crq.request_capability.cmd = REQUEST_CAPABILITY;
3523 
3524 	crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES);
3525 	crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues);
3526 	atomic_inc(&adapter->running_cap_crqs);
3527 	ibmvnic_send_crq(adapter, &crq);
3528 
3529 	crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES);
3530 	crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues);
3531 	atomic_inc(&adapter->running_cap_crqs);
3532 	ibmvnic_send_crq(adapter, &crq);
3533 
3534 	crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES);
3535 	crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues);
3536 	atomic_inc(&adapter->running_cap_crqs);
3537 	ibmvnic_send_crq(adapter, &crq);
3538 
3539 	crq.request_capability.capability =
3540 	    cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ);
3541 	crq.request_capability.number =
3542 	    cpu_to_be64(adapter->req_tx_entries_per_subcrq);
3543 	atomic_inc(&adapter->running_cap_crqs);
3544 	ibmvnic_send_crq(adapter, &crq);
3545 
3546 	crq.request_capability.capability =
3547 	    cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ);
3548 	crq.request_capability.number =
3549 	    cpu_to_be64(adapter->req_rx_add_entries_per_subcrq);
3550 	atomic_inc(&adapter->running_cap_crqs);
3551 	ibmvnic_send_crq(adapter, &crq);
3552 
3553 	crq.request_capability.capability = cpu_to_be16(REQ_MTU);
3554 	crq.request_capability.number = cpu_to_be64(adapter->req_mtu);
3555 	atomic_inc(&adapter->running_cap_crqs);
3556 	ibmvnic_send_crq(adapter, &crq);
3557 
3558 	if (adapter->netdev->flags & IFF_PROMISC) {
3559 		if (adapter->promisc_supported) {
3560 			crq.request_capability.capability =
3561 			    cpu_to_be16(PROMISC_REQUESTED);
3562 			crq.request_capability.number = cpu_to_be64(1);
3563 			atomic_inc(&adapter->running_cap_crqs);
3564 			ibmvnic_send_crq(adapter, &crq);
3565 		}
3566 	} else {
3567 		crq.request_capability.capability =
3568 		    cpu_to_be16(PROMISC_REQUESTED);
3569 		crq.request_capability.number = cpu_to_be64(0);
3570 		atomic_inc(&adapter->running_cap_crqs);
3571 		ibmvnic_send_crq(adapter, &crq);
3572 	}
3573 }
3574 
3575 static int pending_scrq(struct ibmvnic_adapter *adapter,
3576 			struct ibmvnic_sub_crq_queue *scrq)
3577 {
3578 	union sub_crq *entry = &scrq->msgs[scrq->cur];
3579 	int rc;
3580 
3581 	rc = !!(entry->generic.first & IBMVNIC_CRQ_CMD_RSP);
3582 
3583 	/* Ensure that the SCRQ valid flag is loaded prior to loading the
3584 	 * contents of the SCRQ descriptor
3585 	 */
3586 	dma_rmb();
3587 
3588 	return rc;
3589 }
3590 
3591 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter,
3592 					struct ibmvnic_sub_crq_queue *scrq)
3593 {
3594 	union sub_crq *entry;
3595 	unsigned long flags;
3596 
3597 	spin_lock_irqsave(&scrq->lock, flags);
3598 	entry = &scrq->msgs[scrq->cur];
3599 	if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) {
3600 		if (++scrq->cur == scrq->size)
3601 			scrq->cur = 0;
3602 	} else {
3603 		entry = NULL;
3604 	}
3605 	spin_unlock_irqrestore(&scrq->lock, flags);
3606 
3607 	/* Ensure that the SCRQ valid flag is loaded prior to loading the
3608 	 * contents of the SCRQ descriptor
3609 	 */
3610 	dma_rmb();
3611 
3612 	return entry;
3613 }
3614 
3615 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter)
3616 {
3617 	struct ibmvnic_crq_queue *queue = &adapter->crq;
3618 	union ibmvnic_crq *crq;
3619 
3620 	crq = &queue->msgs[queue->cur];
3621 	if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) {
3622 		if (++queue->cur == queue->size)
3623 			queue->cur = 0;
3624 	} else {
3625 		crq = NULL;
3626 	}
3627 
3628 	return crq;
3629 }
3630 
3631 static void print_subcrq_error(struct device *dev, int rc, const char *func)
3632 {
3633 	switch (rc) {
3634 	case H_PARAMETER:
3635 		dev_warn_ratelimited(dev,
3636 				     "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n",
3637 				     func, rc);
3638 		break;
3639 	case H_CLOSED:
3640 		dev_warn_ratelimited(dev,
3641 				     "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n",
3642 				     func, rc);
3643 		break;
3644 	default:
3645 		dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc);
3646 		break;
3647 	}
3648 }
3649 
3650 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter,
3651 				u64 remote_handle, u64 ioba, u64 num_entries)
3652 {
3653 	unsigned int ua = adapter->vdev->unit_address;
3654 	struct device *dev = &adapter->vdev->dev;
3655 	int rc;
3656 
3657 	/* Make sure the hypervisor sees the complete request */
3658 	dma_wmb();
3659 	rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua,
3660 				cpu_to_be64(remote_handle),
3661 				ioba, num_entries);
3662 
3663 	if (rc)
3664 		print_subcrq_error(dev, rc, __func__);
3665 
3666 	return rc;
3667 }
3668 
3669 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter,
3670 			    union ibmvnic_crq *crq)
3671 {
3672 	unsigned int ua = adapter->vdev->unit_address;
3673 	struct device *dev = &adapter->vdev->dev;
3674 	u64 *u64_crq = (u64 *)crq;
3675 	int rc;
3676 
3677 	netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n",
3678 		   (unsigned long)cpu_to_be64(u64_crq[0]),
3679 		   (unsigned long)cpu_to_be64(u64_crq[1]));
3680 
3681 	if (!adapter->crq.active &&
3682 	    crq->generic.first != IBMVNIC_CRQ_INIT_CMD) {
3683 		dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n");
3684 		return -EINVAL;
3685 	}
3686 
3687 	/* Make sure the hypervisor sees the complete request */
3688 	dma_wmb();
3689 
3690 	rc = plpar_hcall_norets(H_SEND_CRQ, ua,
3691 				cpu_to_be64(u64_crq[0]),
3692 				cpu_to_be64(u64_crq[1]));
3693 
3694 	if (rc) {
3695 		if (rc == H_CLOSED) {
3696 			dev_warn(dev, "CRQ Queue closed\n");
3697 			/* do not reset, report the fail, wait for passive init from server */
3698 		}
3699 
3700 		dev_warn(dev, "Send error (rc=%d)\n", rc);
3701 	}
3702 
3703 	return rc;
3704 }
3705 
3706 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter)
3707 {
3708 	struct device *dev = &adapter->vdev->dev;
3709 	union ibmvnic_crq crq;
3710 	int retries = 100;
3711 	int rc;
3712 
3713 	memset(&crq, 0, sizeof(crq));
3714 	crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
3715 	crq.generic.cmd = IBMVNIC_CRQ_INIT;
3716 	netdev_dbg(adapter->netdev, "Sending CRQ init\n");
3717 
3718 	do {
3719 		rc = ibmvnic_send_crq(adapter, &crq);
3720 		if (rc != H_CLOSED)
3721 			break;
3722 		retries--;
3723 		msleep(50);
3724 
3725 	} while (retries > 0);
3726 
3727 	if (rc) {
3728 		dev_err(dev, "Failed to send init request, rc = %d\n", rc);
3729 		return rc;
3730 	}
3731 
3732 	return 0;
3733 }
3734 
3735 static int send_version_xchg(struct ibmvnic_adapter *adapter)
3736 {
3737 	union ibmvnic_crq crq;
3738 
3739 	memset(&crq, 0, sizeof(crq));
3740 	crq.version_exchange.first = IBMVNIC_CRQ_CMD;
3741 	crq.version_exchange.cmd = VERSION_EXCHANGE;
3742 	crq.version_exchange.version = cpu_to_be16(ibmvnic_version);
3743 
3744 	return ibmvnic_send_crq(adapter, &crq);
3745 }
3746 
3747 struct vnic_login_client_data {
3748 	u8	type;
3749 	__be16	len;
3750 	char	name[];
3751 } __packed;
3752 
3753 static int vnic_client_data_len(struct ibmvnic_adapter *adapter)
3754 {
3755 	int len;
3756 
3757 	/* Calculate the amount of buffer space needed for the
3758 	 * vnic client data in the login buffer. There are four entries,
3759 	 * OS name, LPAR name, device name, and a null last entry.
3760 	 */
3761 	len = 4 * sizeof(struct vnic_login_client_data);
3762 	len += 6; /* "Linux" plus NULL */
3763 	len += strlen(utsname()->nodename) + 1;
3764 	len += strlen(adapter->netdev->name) + 1;
3765 
3766 	return len;
3767 }
3768 
3769 static void vnic_add_client_data(struct ibmvnic_adapter *adapter,
3770 				 struct vnic_login_client_data *vlcd)
3771 {
3772 	const char *os_name = "Linux";
3773 	int len;
3774 
3775 	/* Type 1 - LPAR OS */
3776 	vlcd->type = 1;
3777 	len = strlen(os_name) + 1;
3778 	vlcd->len = cpu_to_be16(len);
3779 	strncpy(vlcd->name, os_name, len);
3780 	vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
3781 
3782 	/* Type 2 - LPAR name */
3783 	vlcd->type = 2;
3784 	len = strlen(utsname()->nodename) + 1;
3785 	vlcd->len = cpu_to_be16(len);
3786 	strncpy(vlcd->name, utsname()->nodename, len);
3787 	vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
3788 
3789 	/* Type 3 - device name */
3790 	vlcd->type = 3;
3791 	len = strlen(adapter->netdev->name) + 1;
3792 	vlcd->len = cpu_to_be16(len);
3793 	strncpy(vlcd->name, adapter->netdev->name, len);
3794 }
3795 
3796 static int send_login(struct ibmvnic_adapter *adapter)
3797 {
3798 	struct ibmvnic_login_rsp_buffer *login_rsp_buffer;
3799 	struct ibmvnic_login_buffer *login_buffer;
3800 	struct device *dev = &adapter->vdev->dev;
3801 	struct vnic_login_client_data *vlcd;
3802 	dma_addr_t rsp_buffer_token;
3803 	dma_addr_t buffer_token;
3804 	size_t rsp_buffer_size;
3805 	union ibmvnic_crq crq;
3806 	int client_data_len;
3807 	size_t buffer_size;
3808 	__be64 *tx_list_p;
3809 	__be64 *rx_list_p;
3810 	int rc;
3811 	int i;
3812 
3813 	if (!adapter->tx_scrq || !adapter->rx_scrq) {
3814 		netdev_err(adapter->netdev,
3815 			   "RX or TX queues are not allocated, device login failed\n");
3816 		return -1;
3817 	}
3818 
3819 	release_login_buffer(adapter);
3820 	release_login_rsp_buffer(adapter);
3821 
3822 	client_data_len = vnic_client_data_len(adapter);
3823 
3824 	buffer_size =
3825 	    sizeof(struct ibmvnic_login_buffer) +
3826 	    sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) +
3827 	    client_data_len;
3828 
3829 	login_buffer = kzalloc(buffer_size, GFP_ATOMIC);
3830 	if (!login_buffer)
3831 		goto buf_alloc_failed;
3832 
3833 	buffer_token = dma_map_single(dev, login_buffer, buffer_size,
3834 				      DMA_TO_DEVICE);
3835 	if (dma_mapping_error(dev, buffer_token)) {
3836 		dev_err(dev, "Couldn't map login buffer\n");
3837 		goto buf_map_failed;
3838 	}
3839 
3840 	rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) +
3841 			  sizeof(u64) * adapter->req_tx_queues +
3842 			  sizeof(u64) * adapter->req_rx_queues +
3843 			  sizeof(u64) * adapter->req_rx_queues +
3844 			  sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS;
3845 
3846 	login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC);
3847 	if (!login_rsp_buffer)
3848 		goto buf_rsp_alloc_failed;
3849 
3850 	rsp_buffer_token = dma_map_single(dev, login_rsp_buffer,
3851 					  rsp_buffer_size, DMA_FROM_DEVICE);
3852 	if (dma_mapping_error(dev, rsp_buffer_token)) {
3853 		dev_err(dev, "Couldn't map login rsp buffer\n");
3854 		goto buf_rsp_map_failed;
3855 	}
3856 
3857 	adapter->login_buf = login_buffer;
3858 	adapter->login_buf_token = buffer_token;
3859 	adapter->login_buf_sz = buffer_size;
3860 	adapter->login_rsp_buf = login_rsp_buffer;
3861 	adapter->login_rsp_buf_token = rsp_buffer_token;
3862 	adapter->login_rsp_buf_sz = rsp_buffer_size;
3863 
3864 	login_buffer->len = cpu_to_be32(buffer_size);
3865 	login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB);
3866 	login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues);
3867 	login_buffer->off_txcomp_subcrqs =
3868 	    cpu_to_be32(sizeof(struct ibmvnic_login_buffer));
3869 	login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues);
3870 	login_buffer->off_rxcomp_subcrqs =
3871 	    cpu_to_be32(sizeof(struct ibmvnic_login_buffer) +
3872 			sizeof(u64) * adapter->req_tx_queues);
3873 	login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token);
3874 	login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size);
3875 
3876 	tx_list_p = (__be64 *)((char *)login_buffer +
3877 				      sizeof(struct ibmvnic_login_buffer));
3878 	rx_list_p = (__be64 *)((char *)login_buffer +
3879 				      sizeof(struct ibmvnic_login_buffer) +
3880 				      sizeof(u64) * adapter->req_tx_queues);
3881 
3882 	for (i = 0; i < adapter->req_tx_queues; i++) {
3883 		if (adapter->tx_scrq[i]) {
3884 			tx_list_p[i] =
3885 				cpu_to_be64(adapter->tx_scrq[i]->crq_num);
3886 		}
3887 	}
3888 
3889 	for (i = 0; i < adapter->req_rx_queues; i++) {
3890 		if (adapter->rx_scrq[i]) {
3891 			rx_list_p[i] =
3892 				cpu_to_be64(adapter->rx_scrq[i]->crq_num);
3893 		}
3894 	}
3895 
3896 	/* Insert vNIC login client data */
3897 	vlcd = (struct vnic_login_client_data *)
3898 		((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues));
3899 	login_buffer->client_data_offset =
3900 			cpu_to_be32((char *)vlcd - (char *)login_buffer);
3901 	login_buffer->client_data_len = cpu_to_be32(client_data_len);
3902 
3903 	vnic_add_client_data(adapter, vlcd);
3904 
3905 	netdev_dbg(adapter->netdev, "Login Buffer:\n");
3906 	for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) {
3907 		netdev_dbg(adapter->netdev, "%016lx\n",
3908 			   ((unsigned long *)(adapter->login_buf))[i]);
3909 	}
3910 
3911 	memset(&crq, 0, sizeof(crq));
3912 	crq.login.first = IBMVNIC_CRQ_CMD;
3913 	crq.login.cmd = LOGIN;
3914 	crq.login.ioba = cpu_to_be32(buffer_token);
3915 	crq.login.len = cpu_to_be32(buffer_size);
3916 
3917 	adapter->login_pending = true;
3918 	rc = ibmvnic_send_crq(adapter, &crq);
3919 	if (rc) {
3920 		adapter->login_pending = false;
3921 		netdev_err(adapter->netdev, "Failed to send login, rc=%d\n", rc);
3922 		goto buf_rsp_map_failed;
3923 	}
3924 
3925 	return 0;
3926 
3927 buf_rsp_map_failed:
3928 	kfree(login_rsp_buffer);
3929 	adapter->login_rsp_buf = NULL;
3930 buf_rsp_alloc_failed:
3931 	dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE);
3932 buf_map_failed:
3933 	kfree(login_buffer);
3934 	adapter->login_buf = NULL;
3935 buf_alloc_failed:
3936 	return -1;
3937 }
3938 
3939 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr,
3940 			    u32 len, u8 map_id)
3941 {
3942 	union ibmvnic_crq crq;
3943 
3944 	memset(&crq, 0, sizeof(crq));
3945 	crq.request_map.first = IBMVNIC_CRQ_CMD;
3946 	crq.request_map.cmd = REQUEST_MAP;
3947 	crq.request_map.map_id = map_id;
3948 	crq.request_map.ioba = cpu_to_be32(addr);
3949 	crq.request_map.len = cpu_to_be32(len);
3950 	return ibmvnic_send_crq(adapter, &crq);
3951 }
3952 
3953 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id)
3954 {
3955 	union ibmvnic_crq crq;
3956 
3957 	memset(&crq, 0, sizeof(crq));
3958 	crq.request_unmap.first = IBMVNIC_CRQ_CMD;
3959 	crq.request_unmap.cmd = REQUEST_UNMAP;
3960 	crq.request_unmap.map_id = map_id;
3961 	return ibmvnic_send_crq(adapter, &crq);
3962 }
3963 
3964 static void send_query_map(struct ibmvnic_adapter *adapter)
3965 {
3966 	union ibmvnic_crq crq;
3967 
3968 	memset(&crq, 0, sizeof(crq));
3969 	crq.query_map.first = IBMVNIC_CRQ_CMD;
3970 	crq.query_map.cmd = QUERY_MAP;
3971 	ibmvnic_send_crq(adapter, &crq);
3972 }
3973 
3974 /* Send a series of CRQs requesting various capabilities of the VNIC server */
3975 static void send_query_cap(struct ibmvnic_adapter *adapter)
3976 {
3977 	union ibmvnic_crq crq;
3978 
3979 	atomic_set(&adapter->running_cap_crqs, 0);
3980 	memset(&crq, 0, sizeof(crq));
3981 	crq.query_capability.first = IBMVNIC_CRQ_CMD;
3982 	crq.query_capability.cmd = QUERY_CAPABILITY;
3983 
3984 	crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES);
3985 	atomic_inc(&adapter->running_cap_crqs);
3986 	ibmvnic_send_crq(adapter, &crq);
3987 
3988 	crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES);
3989 	atomic_inc(&adapter->running_cap_crqs);
3990 	ibmvnic_send_crq(adapter, &crq);
3991 
3992 	crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES);
3993 	atomic_inc(&adapter->running_cap_crqs);
3994 	ibmvnic_send_crq(adapter, &crq);
3995 
3996 	crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES);
3997 	atomic_inc(&adapter->running_cap_crqs);
3998 	ibmvnic_send_crq(adapter, &crq);
3999 
4000 	crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES);
4001 	atomic_inc(&adapter->running_cap_crqs);
4002 	ibmvnic_send_crq(adapter, &crq);
4003 
4004 	crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES);
4005 	atomic_inc(&adapter->running_cap_crqs);
4006 	ibmvnic_send_crq(adapter, &crq);
4007 
4008 	crq.query_capability.capability =
4009 	    cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ);
4010 	atomic_inc(&adapter->running_cap_crqs);
4011 	ibmvnic_send_crq(adapter, &crq);
4012 
4013 	crq.query_capability.capability =
4014 	    cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ);
4015 	atomic_inc(&adapter->running_cap_crqs);
4016 	ibmvnic_send_crq(adapter, &crq);
4017 
4018 	crq.query_capability.capability =
4019 	    cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ);
4020 	atomic_inc(&adapter->running_cap_crqs);
4021 	ibmvnic_send_crq(adapter, &crq);
4022 
4023 	crq.query_capability.capability =
4024 	    cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ);
4025 	atomic_inc(&adapter->running_cap_crqs);
4026 	ibmvnic_send_crq(adapter, &crq);
4027 
4028 	crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD);
4029 	atomic_inc(&adapter->running_cap_crqs);
4030 	ibmvnic_send_crq(adapter, &crq);
4031 
4032 	crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED);
4033 	atomic_inc(&adapter->running_cap_crqs);
4034 	ibmvnic_send_crq(adapter, &crq);
4035 
4036 	crq.query_capability.capability = cpu_to_be16(MIN_MTU);
4037 	atomic_inc(&adapter->running_cap_crqs);
4038 	ibmvnic_send_crq(adapter, &crq);
4039 
4040 	crq.query_capability.capability = cpu_to_be16(MAX_MTU);
4041 	atomic_inc(&adapter->running_cap_crqs);
4042 	ibmvnic_send_crq(adapter, &crq);
4043 
4044 	crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS);
4045 	atomic_inc(&adapter->running_cap_crqs);
4046 	ibmvnic_send_crq(adapter, &crq);
4047 
4048 	crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION);
4049 	atomic_inc(&adapter->running_cap_crqs);
4050 	ibmvnic_send_crq(adapter, &crq);
4051 
4052 	crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION);
4053 	atomic_inc(&adapter->running_cap_crqs);
4054 	ibmvnic_send_crq(adapter, &crq);
4055 
4056 	crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES);
4057 	atomic_inc(&adapter->running_cap_crqs);
4058 	ibmvnic_send_crq(adapter, &crq);
4059 
4060 	crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED);
4061 	atomic_inc(&adapter->running_cap_crqs);
4062 	ibmvnic_send_crq(adapter, &crq);
4063 
4064 	crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES);
4065 	atomic_inc(&adapter->running_cap_crqs);
4066 	ibmvnic_send_crq(adapter, &crq);
4067 
4068 	crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES);
4069 	atomic_inc(&adapter->running_cap_crqs);
4070 	ibmvnic_send_crq(adapter, &crq);
4071 
4072 	crq.query_capability.capability =
4073 			cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q);
4074 	atomic_inc(&adapter->running_cap_crqs);
4075 	ibmvnic_send_crq(adapter, &crq);
4076 
4077 	crq.query_capability.capability =
4078 			cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ);
4079 	atomic_inc(&adapter->running_cap_crqs);
4080 	ibmvnic_send_crq(adapter, &crq);
4081 
4082 	crq.query_capability.capability =
4083 			cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ);
4084 	atomic_inc(&adapter->running_cap_crqs);
4085 	ibmvnic_send_crq(adapter, &crq);
4086 
4087 	crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ);
4088 	atomic_inc(&adapter->running_cap_crqs);
4089 	ibmvnic_send_crq(adapter, &crq);
4090 }
4091 
4092 static void send_query_ip_offload(struct ibmvnic_adapter *adapter)
4093 {
4094 	int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer);
4095 	struct device *dev = &adapter->vdev->dev;
4096 	union ibmvnic_crq crq;
4097 
4098 	adapter->ip_offload_tok =
4099 		dma_map_single(dev,
4100 			       &adapter->ip_offload_buf,
4101 			       buf_sz,
4102 			       DMA_FROM_DEVICE);
4103 
4104 	if (dma_mapping_error(dev, adapter->ip_offload_tok)) {
4105 		if (!firmware_has_feature(FW_FEATURE_CMO))
4106 			dev_err(dev, "Couldn't map offload buffer\n");
4107 		return;
4108 	}
4109 
4110 	memset(&crq, 0, sizeof(crq));
4111 	crq.query_ip_offload.first = IBMVNIC_CRQ_CMD;
4112 	crq.query_ip_offload.cmd = QUERY_IP_OFFLOAD;
4113 	crq.query_ip_offload.len = cpu_to_be32(buf_sz);
4114 	crq.query_ip_offload.ioba =
4115 	    cpu_to_be32(adapter->ip_offload_tok);
4116 
4117 	ibmvnic_send_crq(adapter, &crq);
4118 }
4119 
4120 static void send_control_ip_offload(struct ibmvnic_adapter *adapter)
4121 {
4122 	struct ibmvnic_control_ip_offload_buffer *ctrl_buf = &adapter->ip_offload_ctrl;
4123 	struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4124 	struct device *dev = &adapter->vdev->dev;
4125 	netdev_features_t old_hw_features = 0;
4126 	union ibmvnic_crq crq;
4127 
4128 	adapter->ip_offload_ctrl_tok =
4129 		dma_map_single(dev,
4130 			       ctrl_buf,
4131 			       sizeof(adapter->ip_offload_ctrl),
4132 			       DMA_TO_DEVICE);
4133 
4134 	if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) {
4135 		dev_err(dev, "Couldn't map ip offload control buffer\n");
4136 		return;
4137 	}
4138 
4139 	ctrl_buf->len = cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4140 	ctrl_buf->version = cpu_to_be32(INITIAL_VERSION_IOB);
4141 	ctrl_buf->ipv4_chksum = buf->ipv4_chksum;
4142 	ctrl_buf->ipv6_chksum = buf->ipv6_chksum;
4143 	ctrl_buf->tcp_ipv4_chksum = buf->tcp_ipv4_chksum;
4144 	ctrl_buf->udp_ipv4_chksum = buf->udp_ipv4_chksum;
4145 	ctrl_buf->tcp_ipv6_chksum = buf->tcp_ipv6_chksum;
4146 	ctrl_buf->udp_ipv6_chksum = buf->udp_ipv6_chksum;
4147 	ctrl_buf->large_tx_ipv4 = buf->large_tx_ipv4;
4148 	ctrl_buf->large_tx_ipv6 = buf->large_tx_ipv6;
4149 
4150 	/* large_rx disabled for now, additional features needed */
4151 	ctrl_buf->large_rx_ipv4 = 0;
4152 	ctrl_buf->large_rx_ipv6 = 0;
4153 
4154 	if (adapter->state != VNIC_PROBING) {
4155 		old_hw_features = adapter->netdev->hw_features;
4156 		adapter->netdev->hw_features = 0;
4157 	}
4158 
4159 	adapter->netdev->hw_features = NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
4160 
4161 	if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum)
4162 		adapter->netdev->hw_features |= NETIF_F_IP_CSUM;
4163 
4164 	if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum)
4165 		adapter->netdev->hw_features |= NETIF_F_IPV6_CSUM;
4166 
4167 	if ((adapter->netdev->features &
4168 	    (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
4169 		adapter->netdev->hw_features |= NETIF_F_RXCSUM;
4170 
4171 	if (buf->large_tx_ipv4)
4172 		adapter->netdev->hw_features |= NETIF_F_TSO;
4173 	if (buf->large_tx_ipv6)
4174 		adapter->netdev->hw_features |= NETIF_F_TSO6;
4175 
4176 	if (adapter->state == VNIC_PROBING) {
4177 		adapter->netdev->features |= adapter->netdev->hw_features;
4178 	} else if (old_hw_features != adapter->netdev->hw_features) {
4179 		netdev_features_t tmp = 0;
4180 
4181 		/* disable features no longer supported */
4182 		adapter->netdev->features &= adapter->netdev->hw_features;
4183 		/* turn on features now supported if previously enabled */
4184 		tmp = (old_hw_features ^ adapter->netdev->hw_features) &
4185 			adapter->netdev->hw_features;
4186 		adapter->netdev->features |=
4187 				tmp & adapter->netdev->wanted_features;
4188 	}
4189 
4190 	memset(&crq, 0, sizeof(crq));
4191 	crq.control_ip_offload.first = IBMVNIC_CRQ_CMD;
4192 	crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD;
4193 	crq.control_ip_offload.len =
4194 	    cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
4195 	crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok);
4196 	ibmvnic_send_crq(adapter, &crq);
4197 }
4198 
4199 static void handle_vpd_size_rsp(union ibmvnic_crq *crq,
4200 				struct ibmvnic_adapter *adapter)
4201 {
4202 	struct device *dev = &adapter->vdev->dev;
4203 
4204 	if (crq->get_vpd_size_rsp.rc.code) {
4205 		dev_err(dev, "Error retrieving VPD size, rc=%x\n",
4206 			crq->get_vpd_size_rsp.rc.code);
4207 		complete(&adapter->fw_done);
4208 		return;
4209 	}
4210 
4211 	adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len);
4212 	complete(&adapter->fw_done);
4213 }
4214 
4215 static void handle_vpd_rsp(union ibmvnic_crq *crq,
4216 			   struct ibmvnic_adapter *adapter)
4217 {
4218 	struct device *dev = &adapter->vdev->dev;
4219 	unsigned char *substr = NULL;
4220 	u8 fw_level_len = 0;
4221 
4222 	memset(adapter->fw_version, 0, 32);
4223 
4224 	dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len,
4225 			 DMA_FROM_DEVICE);
4226 
4227 	if (crq->get_vpd_rsp.rc.code) {
4228 		dev_err(dev, "Error retrieving VPD from device, rc=%x\n",
4229 			crq->get_vpd_rsp.rc.code);
4230 		goto complete;
4231 	}
4232 
4233 	/* get the position of the firmware version info
4234 	 * located after the ASCII 'RM' substring in the buffer
4235 	 */
4236 	substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len);
4237 	if (!substr) {
4238 		dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n");
4239 		goto complete;
4240 	}
4241 
4242 	/* get length of firmware level ASCII substring */
4243 	if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) {
4244 		fw_level_len = *(substr + 2);
4245 	} else {
4246 		dev_info(dev, "Length of FW substr extrapolated VDP buff\n");
4247 		goto complete;
4248 	}
4249 
4250 	/* copy firmware version string from vpd into adapter */
4251 	if ((substr + 3 + fw_level_len) <
4252 	    (adapter->vpd->buff + adapter->vpd->len)) {
4253 		strncpy((char *)adapter->fw_version, substr + 3, fw_level_len);
4254 	} else {
4255 		dev_info(dev, "FW substr extrapolated VPD buff\n");
4256 	}
4257 
4258 complete:
4259 	if (adapter->fw_version[0] == '\0')
4260 		strncpy((char *)adapter->fw_version, "N/A", 3 * sizeof(char));
4261 	complete(&adapter->fw_done);
4262 }
4263 
4264 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter)
4265 {
4266 	struct device *dev = &adapter->vdev->dev;
4267 	struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
4268 	int i;
4269 
4270 	dma_unmap_single(dev, adapter->ip_offload_tok,
4271 			 sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE);
4272 
4273 	netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n");
4274 	for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++)
4275 		netdev_dbg(adapter->netdev, "%016lx\n",
4276 			   ((unsigned long *)(buf))[i]);
4277 
4278 	netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum);
4279 	netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum);
4280 	netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n",
4281 		   buf->tcp_ipv4_chksum);
4282 	netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n",
4283 		   buf->tcp_ipv6_chksum);
4284 	netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n",
4285 		   buf->udp_ipv4_chksum);
4286 	netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n",
4287 		   buf->udp_ipv6_chksum);
4288 	netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n",
4289 		   buf->large_tx_ipv4);
4290 	netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n",
4291 		   buf->large_tx_ipv6);
4292 	netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n",
4293 		   buf->large_rx_ipv4);
4294 	netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n",
4295 		   buf->large_rx_ipv6);
4296 	netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n",
4297 		   buf->max_ipv4_header_size);
4298 	netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n",
4299 		   buf->max_ipv6_header_size);
4300 	netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n",
4301 		   buf->max_tcp_header_size);
4302 	netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n",
4303 		   buf->max_udp_header_size);
4304 	netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n",
4305 		   buf->max_large_tx_size);
4306 	netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n",
4307 		   buf->max_large_rx_size);
4308 	netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n",
4309 		   buf->ipv6_extension_header);
4310 	netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n",
4311 		   buf->tcp_pseudosum_req);
4312 	netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n",
4313 		   buf->num_ipv6_ext_headers);
4314 	netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n",
4315 		   buf->off_ipv6_ext_headers);
4316 
4317 	send_control_ip_offload(adapter);
4318 }
4319 
4320 static const char *ibmvnic_fw_err_cause(u16 cause)
4321 {
4322 	switch (cause) {
4323 	case ADAPTER_PROBLEM:
4324 		return "adapter problem";
4325 	case BUS_PROBLEM:
4326 		return "bus problem";
4327 	case FW_PROBLEM:
4328 		return "firmware problem";
4329 	case DD_PROBLEM:
4330 		return "device driver problem";
4331 	case EEH_RECOVERY:
4332 		return "EEH recovery";
4333 	case FW_UPDATED:
4334 		return "firmware updated";
4335 	case LOW_MEMORY:
4336 		return "low Memory";
4337 	default:
4338 		return "unknown";
4339 	}
4340 }
4341 
4342 static void handle_error_indication(union ibmvnic_crq *crq,
4343 				    struct ibmvnic_adapter *adapter)
4344 {
4345 	struct device *dev = &adapter->vdev->dev;
4346 	u16 cause;
4347 
4348 	cause = be16_to_cpu(crq->error_indication.error_cause);
4349 
4350 	dev_warn_ratelimited(dev,
4351 			     "Firmware reports %serror, cause: %s. Starting recovery...\n",
4352 			     crq->error_indication.flags
4353 				& IBMVNIC_FATAL_ERROR ? "FATAL " : "",
4354 			     ibmvnic_fw_err_cause(cause));
4355 
4356 	if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR)
4357 		ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4358 	else
4359 		ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL);
4360 }
4361 
4362 static int handle_change_mac_rsp(union ibmvnic_crq *crq,
4363 				 struct ibmvnic_adapter *adapter)
4364 {
4365 	struct net_device *netdev = adapter->netdev;
4366 	struct device *dev = &adapter->vdev->dev;
4367 	long rc;
4368 
4369 	rc = crq->change_mac_addr_rsp.rc.code;
4370 	if (rc) {
4371 		dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc);
4372 		goto out;
4373 	}
4374 	/* crq->change_mac_addr.mac_addr is the requested one
4375 	 * crq->change_mac_addr_rsp.mac_addr is the returned valid one.
4376 	 */
4377 	ether_addr_copy(netdev->dev_addr,
4378 			&crq->change_mac_addr_rsp.mac_addr[0]);
4379 	ether_addr_copy(adapter->mac_addr,
4380 			&crq->change_mac_addr_rsp.mac_addr[0]);
4381 out:
4382 	complete(&adapter->fw_done);
4383 	return rc;
4384 }
4385 
4386 static void handle_request_cap_rsp(union ibmvnic_crq *crq,
4387 				   struct ibmvnic_adapter *adapter)
4388 {
4389 	struct device *dev = &adapter->vdev->dev;
4390 	u64 *req_value;
4391 	char *name;
4392 
4393 	atomic_dec(&adapter->running_cap_crqs);
4394 	switch (be16_to_cpu(crq->request_capability_rsp.capability)) {
4395 	case REQ_TX_QUEUES:
4396 		req_value = &adapter->req_tx_queues;
4397 		name = "tx";
4398 		break;
4399 	case REQ_RX_QUEUES:
4400 		req_value = &adapter->req_rx_queues;
4401 		name = "rx";
4402 		break;
4403 	case REQ_RX_ADD_QUEUES:
4404 		req_value = &adapter->req_rx_add_queues;
4405 		name = "rx_add";
4406 		break;
4407 	case REQ_TX_ENTRIES_PER_SUBCRQ:
4408 		req_value = &adapter->req_tx_entries_per_subcrq;
4409 		name = "tx_entries_per_subcrq";
4410 		break;
4411 	case REQ_RX_ADD_ENTRIES_PER_SUBCRQ:
4412 		req_value = &adapter->req_rx_add_entries_per_subcrq;
4413 		name = "rx_add_entries_per_subcrq";
4414 		break;
4415 	case REQ_MTU:
4416 		req_value = &adapter->req_mtu;
4417 		name = "mtu";
4418 		break;
4419 	case PROMISC_REQUESTED:
4420 		req_value = &adapter->promisc;
4421 		name = "promisc";
4422 		break;
4423 	default:
4424 		dev_err(dev, "Got invalid cap request rsp %d\n",
4425 			crq->request_capability.capability);
4426 		return;
4427 	}
4428 
4429 	switch (crq->request_capability_rsp.rc.code) {
4430 	case SUCCESS:
4431 		break;
4432 	case PARTIALSUCCESS:
4433 		dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n",
4434 			 *req_value,
4435 			 (long)be64_to_cpu(crq->request_capability_rsp.number),
4436 			 name);
4437 
4438 		if (be16_to_cpu(crq->request_capability_rsp.capability) ==
4439 		    REQ_MTU) {
4440 			pr_err("mtu of %llu is not supported. Reverting.\n",
4441 			       *req_value);
4442 			*req_value = adapter->fallback.mtu;
4443 		} else {
4444 			*req_value =
4445 				be64_to_cpu(crq->request_capability_rsp.number);
4446 		}
4447 
4448 		send_request_cap(adapter, 1);
4449 		return;
4450 	default:
4451 		dev_err(dev, "Error %d in request cap rsp\n",
4452 			crq->request_capability_rsp.rc.code);
4453 		return;
4454 	}
4455 
4456 	/* Done receiving requested capabilities, query IP offload support */
4457 	if (atomic_read(&adapter->running_cap_crqs) == 0) {
4458 		adapter->wait_capability = false;
4459 		send_query_ip_offload(adapter);
4460 	}
4461 }
4462 
4463 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq,
4464 			    struct ibmvnic_adapter *adapter)
4465 {
4466 	struct device *dev = &adapter->vdev->dev;
4467 	struct net_device *netdev = adapter->netdev;
4468 	struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf;
4469 	struct ibmvnic_login_buffer *login = adapter->login_buf;
4470 	u64 *tx_handle_array;
4471 	u64 *rx_handle_array;
4472 	int num_tx_pools;
4473 	int num_rx_pools;
4474 	u64 *size_array;
4475 	int i;
4476 
4477 	/* CHECK: Test/set of login_pending does not need to be atomic
4478 	 * because only ibmvnic_tasklet tests/clears this.
4479 	 */
4480 	if (!adapter->login_pending) {
4481 		netdev_warn(netdev, "Ignoring unexpected login response\n");
4482 		return 0;
4483 	}
4484 	adapter->login_pending = false;
4485 
4486 	dma_unmap_single(dev, adapter->login_buf_token, adapter->login_buf_sz,
4487 			 DMA_TO_DEVICE);
4488 	dma_unmap_single(dev, adapter->login_rsp_buf_token,
4489 			 adapter->login_rsp_buf_sz, DMA_FROM_DEVICE);
4490 
4491 	/* If the number of queues requested can't be allocated by the
4492 	 * server, the login response will return with code 1. We will need
4493 	 * to resend the login buffer with fewer queues requested.
4494 	 */
4495 	if (login_rsp_crq->generic.rc.code) {
4496 		adapter->init_done_rc = login_rsp_crq->generic.rc.code;
4497 		complete(&adapter->init_done);
4498 		return 0;
4499 	}
4500 
4501 	netdev->mtu = adapter->req_mtu - ETH_HLEN;
4502 
4503 	netdev_dbg(adapter->netdev, "Login Response Buffer:\n");
4504 	for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) {
4505 		netdev_dbg(adapter->netdev, "%016lx\n",
4506 			   ((unsigned long *)(adapter->login_rsp_buf))[i]);
4507 	}
4508 
4509 	/* Sanity checks */
4510 	if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs ||
4511 	    (be32_to_cpu(login->num_rxcomp_subcrqs) *
4512 	     adapter->req_rx_add_queues !=
4513 	     be32_to_cpu(login_rsp->num_rxadd_subcrqs))) {
4514 		dev_err(dev, "FATAL: Inconsistent login and login rsp\n");
4515 		ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4516 		return -EIO;
4517 	}
4518 	size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4519 		be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size));
4520 	/* variable buffer sizes are not supported, so just read the
4521 	 * first entry.
4522 	 */
4523 	adapter->cur_rx_buf_sz = be64_to_cpu(size_array[0]);
4524 
4525 	num_tx_pools = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs);
4526 	num_rx_pools = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
4527 
4528 	tx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4529 				  be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs));
4530 	rx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
4531 				  be32_to_cpu(adapter->login_rsp_buf->off_rxadd_subcrqs));
4532 
4533 	for (i = 0; i < num_tx_pools; i++)
4534 		adapter->tx_scrq[i]->handle = tx_handle_array[i];
4535 
4536 	for (i = 0; i < num_rx_pools; i++)
4537 		adapter->rx_scrq[i]->handle = rx_handle_array[i];
4538 
4539 	adapter->num_active_tx_scrqs = num_tx_pools;
4540 	adapter->num_active_rx_scrqs = num_rx_pools;
4541 	release_login_rsp_buffer(adapter);
4542 	release_login_buffer(adapter);
4543 	complete(&adapter->init_done);
4544 
4545 	return 0;
4546 }
4547 
4548 static void handle_request_unmap_rsp(union ibmvnic_crq *crq,
4549 				     struct ibmvnic_adapter *adapter)
4550 {
4551 	struct device *dev = &adapter->vdev->dev;
4552 	long rc;
4553 
4554 	rc = crq->request_unmap_rsp.rc.code;
4555 	if (rc)
4556 		dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc);
4557 }
4558 
4559 static void handle_query_map_rsp(union ibmvnic_crq *crq,
4560 				 struct ibmvnic_adapter *adapter)
4561 {
4562 	struct net_device *netdev = adapter->netdev;
4563 	struct device *dev = &adapter->vdev->dev;
4564 	long rc;
4565 
4566 	rc = crq->query_map_rsp.rc.code;
4567 	if (rc) {
4568 		dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc);
4569 		return;
4570 	}
4571 	netdev_dbg(netdev, "page_size = %d\ntot_pages = %d\nfree_pages = %d\n",
4572 		   crq->query_map_rsp.page_size, crq->query_map_rsp.tot_pages,
4573 		   crq->query_map_rsp.free_pages);
4574 }
4575 
4576 static void handle_query_cap_rsp(union ibmvnic_crq *crq,
4577 				 struct ibmvnic_adapter *adapter)
4578 {
4579 	struct net_device *netdev = adapter->netdev;
4580 	struct device *dev = &adapter->vdev->dev;
4581 	long rc;
4582 
4583 	atomic_dec(&adapter->running_cap_crqs);
4584 	netdev_dbg(netdev, "Outstanding queries: %d\n",
4585 		   atomic_read(&adapter->running_cap_crqs));
4586 	rc = crq->query_capability.rc.code;
4587 	if (rc) {
4588 		dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc);
4589 		goto out;
4590 	}
4591 
4592 	switch (be16_to_cpu(crq->query_capability.capability)) {
4593 	case MIN_TX_QUEUES:
4594 		adapter->min_tx_queues =
4595 		    be64_to_cpu(crq->query_capability.number);
4596 		netdev_dbg(netdev, "min_tx_queues = %lld\n",
4597 			   adapter->min_tx_queues);
4598 		break;
4599 	case MIN_RX_QUEUES:
4600 		adapter->min_rx_queues =
4601 		    be64_to_cpu(crq->query_capability.number);
4602 		netdev_dbg(netdev, "min_rx_queues = %lld\n",
4603 			   adapter->min_rx_queues);
4604 		break;
4605 	case MIN_RX_ADD_QUEUES:
4606 		adapter->min_rx_add_queues =
4607 		    be64_to_cpu(crq->query_capability.number);
4608 		netdev_dbg(netdev, "min_rx_add_queues = %lld\n",
4609 			   adapter->min_rx_add_queues);
4610 		break;
4611 	case MAX_TX_QUEUES:
4612 		adapter->max_tx_queues =
4613 		    be64_to_cpu(crq->query_capability.number);
4614 		netdev_dbg(netdev, "max_tx_queues = %lld\n",
4615 			   adapter->max_tx_queues);
4616 		break;
4617 	case MAX_RX_QUEUES:
4618 		adapter->max_rx_queues =
4619 		    be64_to_cpu(crq->query_capability.number);
4620 		netdev_dbg(netdev, "max_rx_queues = %lld\n",
4621 			   adapter->max_rx_queues);
4622 		break;
4623 	case MAX_RX_ADD_QUEUES:
4624 		adapter->max_rx_add_queues =
4625 		    be64_to_cpu(crq->query_capability.number);
4626 		netdev_dbg(netdev, "max_rx_add_queues = %lld\n",
4627 			   adapter->max_rx_add_queues);
4628 		break;
4629 	case MIN_TX_ENTRIES_PER_SUBCRQ:
4630 		adapter->min_tx_entries_per_subcrq =
4631 		    be64_to_cpu(crq->query_capability.number);
4632 		netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n",
4633 			   adapter->min_tx_entries_per_subcrq);
4634 		break;
4635 	case MIN_RX_ADD_ENTRIES_PER_SUBCRQ:
4636 		adapter->min_rx_add_entries_per_subcrq =
4637 		    be64_to_cpu(crq->query_capability.number);
4638 		netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n",
4639 			   adapter->min_rx_add_entries_per_subcrq);
4640 		break;
4641 	case MAX_TX_ENTRIES_PER_SUBCRQ:
4642 		adapter->max_tx_entries_per_subcrq =
4643 		    be64_to_cpu(crq->query_capability.number);
4644 		netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n",
4645 			   adapter->max_tx_entries_per_subcrq);
4646 		break;
4647 	case MAX_RX_ADD_ENTRIES_PER_SUBCRQ:
4648 		adapter->max_rx_add_entries_per_subcrq =
4649 		    be64_to_cpu(crq->query_capability.number);
4650 		netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n",
4651 			   adapter->max_rx_add_entries_per_subcrq);
4652 		break;
4653 	case TCP_IP_OFFLOAD:
4654 		adapter->tcp_ip_offload =
4655 		    be64_to_cpu(crq->query_capability.number);
4656 		netdev_dbg(netdev, "tcp_ip_offload = %lld\n",
4657 			   adapter->tcp_ip_offload);
4658 		break;
4659 	case PROMISC_SUPPORTED:
4660 		adapter->promisc_supported =
4661 		    be64_to_cpu(crq->query_capability.number);
4662 		netdev_dbg(netdev, "promisc_supported = %lld\n",
4663 			   adapter->promisc_supported);
4664 		break;
4665 	case MIN_MTU:
4666 		adapter->min_mtu = be64_to_cpu(crq->query_capability.number);
4667 		netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
4668 		netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu);
4669 		break;
4670 	case MAX_MTU:
4671 		adapter->max_mtu = be64_to_cpu(crq->query_capability.number);
4672 		netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
4673 		netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu);
4674 		break;
4675 	case MAX_MULTICAST_FILTERS:
4676 		adapter->max_multicast_filters =
4677 		    be64_to_cpu(crq->query_capability.number);
4678 		netdev_dbg(netdev, "max_multicast_filters = %lld\n",
4679 			   adapter->max_multicast_filters);
4680 		break;
4681 	case VLAN_HEADER_INSERTION:
4682 		adapter->vlan_header_insertion =
4683 		    be64_to_cpu(crq->query_capability.number);
4684 		if (adapter->vlan_header_insertion)
4685 			netdev->features |= NETIF_F_HW_VLAN_STAG_TX;
4686 		netdev_dbg(netdev, "vlan_header_insertion = %lld\n",
4687 			   adapter->vlan_header_insertion);
4688 		break;
4689 	case RX_VLAN_HEADER_INSERTION:
4690 		adapter->rx_vlan_header_insertion =
4691 		    be64_to_cpu(crq->query_capability.number);
4692 		netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n",
4693 			   adapter->rx_vlan_header_insertion);
4694 		break;
4695 	case MAX_TX_SG_ENTRIES:
4696 		adapter->max_tx_sg_entries =
4697 		    be64_to_cpu(crq->query_capability.number);
4698 		netdev_dbg(netdev, "max_tx_sg_entries = %lld\n",
4699 			   adapter->max_tx_sg_entries);
4700 		break;
4701 	case RX_SG_SUPPORTED:
4702 		adapter->rx_sg_supported =
4703 		    be64_to_cpu(crq->query_capability.number);
4704 		netdev_dbg(netdev, "rx_sg_supported = %lld\n",
4705 			   adapter->rx_sg_supported);
4706 		break;
4707 	case OPT_TX_COMP_SUB_QUEUES:
4708 		adapter->opt_tx_comp_sub_queues =
4709 		    be64_to_cpu(crq->query_capability.number);
4710 		netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n",
4711 			   adapter->opt_tx_comp_sub_queues);
4712 		break;
4713 	case OPT_RX_COMP_QUEUES:
4714 		adapter->opt_rx_comp_queues =
4715 		    be64_to_cpu(crq->query_capability.number);
4716 		netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n",
4717 			   adapter->opt_rx_comp_queues);
4718 		break;
4719 	case OPT_RX_BUFADD_Q_PER_RX_COMP_Q:
4720 		adapter->opt_rx_bufadd_q_per_rx_comp_q =
4721 		    be64_to_cpu(crq->query_capability.number);
4722 		netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n",
4723 			   adapter->opt_rx_bufadd_q_per_rx_comp_q);
4724 		break;
4725 	case OPT_TX_ENTRIES_PER_SUBCRQ:
4726 		adapter->opt_tx_entries_per_subcrq =
4727 		    be64_to_cpu(crq->query_capability.number);
4728 		netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n",
4729 			   adapter->opt_tx_entries_per_subcrq);
4730 		break;
4731 	case OPT_RXBA_ENTRIES_PER_SUBCRQ:
4732 		adapter->opt_rxba_entries_per_subcrq =
4733 		    be64_to_cpu(crq->query_capability.number);
4734 		netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n",
4735 			   adapter->opt_rxba_entries_per_subcrq);
4736 		break;
4737 	case TX_RX_DESC_REQ:
4738 		adapter->tx_rx_desc_req = crq->query_capability.number;
4739 		netdev_dbg(netdev, "tx_rx_desc_req = %llx\n",
4740 			   adapter->tx_rx_desc_req);
4741 		break;
4742 
4743 	default:
4744 		netdev_err(netdev, "Got invalid cap rsp %d\n",
4745 			   crq->query_capability.capability);
4746 	}
4747 
4748 out:
4749 	if (atomic_read(&adapter->running_cap_crqs) == 0) {
4750 		adapter->wait_capability = false;
4751 		send_request_cap(adapter, 0);
4752 	}
4753 }
4754 
4755 static int send_query_phys_parms(struct ibmvnic_adapter *adapter)
4756 {
4757 	union ibmvnic_crq crq;
4758 	int rc;
4759 
4760 	memset(&crq, 0, sizeof(crq));
4761 	crq.query_phys_parms.first = IBMVNIC_CRQ_CMD;
4762 	crq.query_phys_parms.cmd = QUERY_PHYS_PARMS;
4763 
4764 	mutex_lock(&adapter->fw_lock);
4765 	adapter->fw_done_rc = 0;
4766 	reinit_completion(&adapter->fw_done);
4767 
4768 	rc = ibmvnic_send_crq(adapter, &crq);
4769 	if (rc) {
4770 		mutex_unlock(&adapter->fw_lock);
4771 		return rc;
4772 	}
4773 
4774 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
4775 	if (rc) {
4776 		mutex_unlock(&adapter->fw_lock);
4777 		return rc;
4778 	}
4779 
4780 	mutex_unlock(&adapter->fw_lock);
4781 	return adapter->fw_done_rc ? -EIO : 0;
4782 }
4783 
4784 static int handle_query_phys_parms_rsp(union ibmvnic_crq *crq,
4785 				       struct ibmvnic_adapter *adapter)
4786 {
4787 	struct net_device *netdev = adapter->netdev;
4788 	int rc;
4789 	__be32 rspeed = cpu_to_be32(crq->query_phys_parms_rsp.speed);
4790 
4791 	rc = crq->query_phys_parms_rsp.rc.code;
4792 	if (rc) {
4793 		netdev_err(netdev, "Error %d in QUERY_PHYS_PARMS\n", rc);
4794 		return rc;
4795 	}
4796 	switch (rspeed) {
4797 	case IBMVNIC_10MBPS:
4798 		adapter->speed = SPEED_10;
4799 		break;
4800 	case IBMVNIC_100MBPS:
4801 		adapter->speed = SPEED_100;
4802 		break;
4803 	case IBMVNIC_1GBPS:
4804 		adapter->speed = SPEED_1000;
4805 		break;
4806 	case IBMVNIC_10GBPS:
4807 		adapter->speed = SPEED_10000;
4808 		break;
4809 	case IBMVNIC_25GBPS:
4810 		adapter->speed = SPEED_25000;
4811 		break;
4812 	case IBMVNIC_40GBPS:
4813 		adapter->speed = SPEED_40000;
4814 		break;
4815 	case IBMVNIC_50GBPS:
4816 		adapter->speed = SPEED_50000;
4817 		break;
4818 	case IBMVNIC_100GBPS:
4819 		adapter->speed = SPEED_100000;
4820 		break;
4821 	case IBMVNIC_200GBPS:
4822 		adapter->speed = SPEED_200000;
4823 		break;
4824 	default:
4825 		if (netif_carrier_ok(netdev))
4826 			netdev_warn(netdev, "Unknown speed 0x%08x\n", rspeed);
4827 		adapter->speed = SPEED_UNKNOWN;
4828 	}
4829 	if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_FULL_DUPLEX)
4830 		adapter->duplex = DUPLEX_FULL;
4831 	else if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_HALF_DUPLEX)
4832 		adapter->duplex = DUPLEX_HALF;
4833 	else
4834 		adapter->duplex = DUPLEX_UNKNOWN;
4835 
4836 	return rc;
4837 }
4838 
4839 static void ibmvnic_handle_crq(union ibmvnic_crq *crq,
4840 			       struct ibmvnic_adapter *adapter)
4841 {
4842 	struct ibmvnic_generic_crq *gen_crq = &crq->generic;
4843 	struct net_device *netdev = adapter->netdev;
4844 	struct device *dev = &adapter->vdev->dev;
4845 	u64 *u64_crq = (u64 *)crq;
4846 	long rc;
4847 
4848 	netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n",
4849 		   (unsigned long)cpu_to_be64(u64_crq[0]),
4850 		   (unsigned long)cpu_to_be64(u64_crq[1]));
4851 	switch (gen_crq->first) {
4852 	case IBMVNIC_CRQ_INIT_RSP:
4853 		switch (gen_crq->cmd) {
4854 		case IBMVNIC_CRQ_INIT:
4855 			dev_info(dev, "Partner initialized\n");
4856 			adapter->from_passive_init = true;
4857 			/* Discard any stale login responses from prev reset.
4858 			 * CHECK: should we clear even on INIT_COMPLETE?
4859 			 */
4860 			adapter->login_pending = false;
4861 
4862 			if (!completion_done(&adapter->init_done)) {
4863 				complete(&adapter->init_done);
4864 				adapter->init_done_rc = -EIO;
4865 			}
4866 			rc = ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
4867 			if (rc && rc != -EBUSY) {
4868 				/* We were unable to schedule the failover
4869 				 * reset either because the adapter was still
4870 				 * probing (eg: during kexec) or we could not
4871 				 * allocate memory. Clear the failover_pending
4872 				 * flag since no one else will. We ignore
4873 				 * EBUSY because it means either FAILOVER reset
4874 				 * is already scheduled or the adapter is
4875 				 * being removed.
4876 				 */
4877 				netdev_err(netdev,
4878 					   "Error %ld scheduling failover reset\n",
4879 					   rc);
4880 				adapter->failover_pending = false;
4881 			}
4882 			break;
4883 		case IBMVNIC_CRQ_INIT_COMPLETE:
4884 			dev_info(dev, "Partner initialization complete\n");
4885 			adapter->crq.active = true;
4886 			send_version_xchg(adapter);
4887 			break;
4888 		default:
4889 			dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd);
4890 		}
4891 		return;
4892 	case IBMVNIC_CRQ_XPORT_EVENT:
4893 		netif_carrier_off(netdev);
4894 		adapter->crq.active = false;
4895 		/* terminate any thread waiting for a response
4896 		 * from the device
4897 		 */
4898 		if (!completion_done(&adapter->fw_done)) {
4899 			adapter->fw_done_rc = -EIO;
4900 			complete(&adapter->fw_done);
4901 		}
4902 		if (!completion_done(&adapter->stats_done))
4903 			complete(&adapter->stats_done);
4904 		if (test_bit(0, &adapter->resetting))
4905 			adapter->force_reset_recovery = true;
4906 		if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) {
4907 			dev_info(dev, "Migrated, re-enabling adapter\n");
4908 			ibmvnic_reset(adapter, VNIC_RESET_MOBILITY);
4909 		} else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) {
4910 			dev_info(dev, "Backing device failover detected\n");
4911 			adapter->failover_pending = true;
4912 		} else {
4913 			/* The adapter lost the connection */
4914 			dev_err(dev, "Virtual Adapter failed (rc=%d)\n",
4915 				gen_crq->cmd);
4916 			ibmvnic_reset(adapter, VNIC_RESET_FATAL);
4917 		}
4918 		return;
4919 	case IBMVNIC_CRQ_CMD_RSP:
4920 		break;
4921 	default:
4922 		dev_err(dev, "Got an invalid msg type 0x%02x\n",
4923 			gen_crq->first);
4924 		return;
4925 	}
4926 
4927 	switch (gen_crq->cmd) {
4928 	case VERSION_EXCHANGE_RSP:
4929 		rc = crq->version_exchange_rsp.rc.code;
4930 		if (rc) {
4931 			dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc);
4932 			break;
4933 		}
4934 		ibmvnic_version =
4935 			    be16_to_cpu(crq->version_exchange_rsp.version);
4936 		dev_info(dev, "Partner protocol version is %d\n",
4937 			 ibmvnic_version);
4938 		send_query_cap(adapter);
4939 		break;
4940 	case QUERY_CAPABILITY_RSP:
4941 		handle_query_cap_rsp(crq, adapter);
4942 		break;
4943 	case QUERY_MAP_RSP:
4944 		handle_query_map_rsp(crq, adapter);
4945 		break;
4946 	case REQUEST_MAP_RSP:
4947 		adapter->fw_done_rc = crq->request_map_rsp.rc.code;
4948 		complete(&adapter->fw_done);
4949 		break;
4950 	case REQUEST_UNMAP_RSP:
4951 		handle_request_unmap_rsp(crq, adapter);
4952 		break;
4953 	case REQUEST_CAPABILITY_RSP:
4954 		handle_request_cap_rsp(crq, adapter);
4955 		break;
4956 	case LOGIN_RSP:
4957 		netdev_dbg(netdev, "Got Login Response\n");
4958 		handle_login_rsp(crq, adapter);
4959 		break;
4960 	case LOGICAL_LINK_STATE_RSP:
4961 		netdev_dbg(netdev,
4962 			   "Got Logical Link State Response, state: %d rc: %d\n",
4963 			   crq->logical_link_state_rsp.link_state,
4964 			   crq->logical_link_state_rsp.rc.code);
4965 		adapter->logical_link_state =
4966 		    crq->logical_link_state_rsp.link_state;
4967 		adapter->init_done_rc = crq->logical_link_state_rsp.rc.code;
4968 		complete(&adapter->init_done);
4969 		break;
4970 	case LINK_STATE_INDICATION:
4971 		netdev_dbg(netdev, "Got Logical Link State Indication\n");
4972 		adapter->phys_link_state =
4973 		    crq->link_state_indication.phys_link_state;
4974 		adapter->logical_link_state =
4975 		    crq->link_state_indication.logical_link_state;
4976 		if (adapter->phys_link_state && adapter->logical_link_state)
4977 			netif_carrier_on(netdev);
4978 		else
4979 			netif_carrier_off(netdev);
4980 		break;
4981 	case CHANGE_MAC_ADDR_RSP:
4982 		netdev_dbg(netdev, "Got MAC address change Response\n");
4983 		adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter);
4984 		break;
4985 	case ERROR_INDICATION:
4986 		netdev_dbg(netdev, "Got Error Indication\n");
4987 		handle_error_indication(crq, adapter);
4988 		break;
4989 	case REQUEST_STATISTICS_RSP:
4990 		netdev_dbg(netdev, "Got Statistics Response\n");
4991 		complete(&adapter->stats_done);
4992 		break;
4993 	case QUERY_IP_OFFLOAD_RSP:
4994 		netdev_dbg(netdev, "Got Query IP offload Response\n");
4995 		handle_query_ip_offload_rsp(adapter);
4996 		break;
4997 	case MULTICAST_CTRL_RSP:
4998 		netdev_dbg(netdev, "Got multicast control Response\n");
4999 		break;
5000 	case CONTROL_IP_OFFLOAD_RSP:
5001 		netdev_dbg(netdev, "Got Control IP offload Response\n");
5002 		dma_unmap_single(dev, adapter->ip_offload_ctrl_tok,
5003 				 sizeof(adapter->ip_offload_ctrl),
5004 				 DMA_TO_DEVICE);
5005 		complete(&adapter->init_done);
5006 		break;
5007 	case COLLECT_FW_TRACE_RSP:
5008 		netdev_dbg(netdev, "Got Collect firmware trace Response\n");
5009 		complete(&adapter->fw_done);
5010 		break;
5011 	case GET_VPD_SIZE_RSP:
5012 		handle_vpd_size_rsp(crq, adapter);
5013 		break;
5014 	case GET_VPD_RSP:
5015 		handle_vpd_rsp(crq, adapter);
5016 		break;
5017 	case QUERY_PHYS_PARMS_RSP:
5018 		adapter->fw_done_rc = handle_query_phys_parms_rsp(crq, adapter);
5019 		complete(&adapter->fw_done);
5020 		break;
5021 	default:
5022 		netdev_err(netdev, "Got an invalid cmd type 0x%02x\n",
5023 			   gen_crq->cmd);
5024 	}
5025 }
5026 
5027 static irqreturn_t ibmvnic_interrupt(int irq, void *instance)
5028 {
5029 	struct ibmvnic_adapter *adapter = instance;
5030 
5031 	tasklet_schedule(&adapter->tasklet);
5032 	return IRQ_HANDLED;
5033 }
5034 
5035 static void ibmvnic_tasklet(struct tasklet_struct *t)
5036 {
5037 	struct ibmvnic_adapter *adapter = from_tasklet(adapter, t, tasklet);
5038 	struct ibmvnic_crq_queue *queue = &adapter->crq;
5039 	union ibmvnic_crq *crq;
5040 	unsigned long flags;
5041 	bool done = false;
5042 
5043 	spin_lock_irqsave(&queue->lock, flags);
5044 	while (!done) {
5045 		/* Pull all the valid messages off the CRQ */
5046 		while ((crq = ibmvnic_next_crq(adapter)) != NULL) {
5047 			/* This barrier makes sure ibmvnic_next_crq()'s
5048 			 * crq->generic.first & IBMVNIC_CRQ_CMD_RSP is loaded
5049 			 * before ibmvnic_handle_crq()'s
5050 			 * switch(gen_crq->first) and switch(gen_crq->cmd).
5051 			 */
5052 			dma_rmb();
5053 			ibmvnic_handle_crq(crq, adapter);
5054 			crq->generic.first = 0;
5055 		}
5056 
5057 		/* remain in tasklet until all
5058 		 * capabilities responses are received
5059 		 */
5060 		if (!adapter->wait_capability)
5061 			done = true;
5062 	}
5063 	/* if capabilities CRQ's were sent in this tasklet, the following
5064 	 * tasklet must wait until all responses are received
5065 	 */
5066 	if (atomic_read(&adapter->running_cap_crqs) != 0)
5067 		adapter->wait_capability = true;
5068 	spin_unlock_irqrestore(&queue->lock, flags);
5069 }
5070 
5071 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter)
5072 {
5073 	struct vio_dev *vdev = adapter->vdev;
5074 	int rc;
5075 
5076 	do {
5077 		rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
5078 	} while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc));
5079 
5080 	if (rc)
5081 		dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc);
5082 
5083 	return rc;
5084 }
5085 
5086 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter)
5087 {
5088 	struct ibmvnic_crq_queue *crq = &adapter->crq;
5089 	struct device *dev = &adapter->vdev->dev;
5090 	struct vio_dev *vdev = adapter->vdev;
5091 	int rc;
5092 
5093 	/* Close the CRQ */
5094 	do {
5095 		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5096 	} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5097 
5098 	/* Clean out the queue */
5099 	if (!crq->msgs)
5100 		return -EINVAL;
5101 
5102 	memset(crq->msgs, 0, PAGE_SIZE);
5103 	crq->cur = 0;
5104 	crq->active = false;
5105 
5106 	/* And re-open it again */
5107 	rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5108 				crq->msg_token, PAGE_SIZE);
5109 
5110 	if (rc == H_CLOSED)
5111 		/* Adapter is good, but other end is not ready */
5112 		dev_warn(dev, "Partner adapter not ready\n");
5113 	else if (rc != 0)
5114 		dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc);
5115 
5116 	return rc;
5117 }
5118 
5119 static void release_crq_queue(struct ibmvnic_adapter *adapter)
5120 {
5121 	struct ibmvnic_crq_queue *crq = &adapter->crq;
5122 	struct vio_dev *vdev = adapter->vdev;
5123 	long rc;
5124 
5125 	if (!crq->msgs)
5126 		return;
5127 
5128 	netdev_dbg(adapter->netdev, "Releasing CRQ\n");
5129 	free_irq(vdev->irq, adapter);
5130 	tasklet_kill(&adapter->tasklet);
5131 	do {
5132 		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5133 	} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5134 
5135 	dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE,
5136 			 DMA_BIDIRECTIONAL);
5137 	free_page((unsigned long)crq->msgs);
5138 	crq->msgs = NULL;
5139 	crq->active = false;
5140 }
5141 
5142 static int init_crq_queue(struct ibmvnic_adapter *adapter)
5143 {
5144 	struct ibmvnic_crq_queue *crq = &adapter->crq;
5145 	struct device *dev = &adapter->vdev->dev;
5146 	struct vio_dev *vdev = adapter->vdev;
5147 	int rc, retrc = -ENOMEM;
5148 
5149 	if (crq->msgs)
5150 		return 0;
5151 
5152 	crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL);
5153 	/* Should we allocate more than one page? */
5154 
5155 	if (!crq->msgs)
5156 		return -ENOMEM;
5157 
5158 	crq->size = PAGE_SIZE / sizeof(*crq->msgs);
5159 	crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE,
5160 					DMA_BIDIRECTIONAL);
5161 	if (dma_mapping_error(dev, crq->msg_token))
5162 		goto map_failed;
5163 
5164 	rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
5165 				crq->msg_token, PAGE_SIZE);
5166 
5167 	if (rc == H_RESOURCE)
5168 		/* maybe kexecing and resource is busy. try a reset */
5169 		rc = ibmvnic_reset_crq(adapter);
5170 	retrc = rc;
5171 
5172 	if (rc == H_CLOSED) {
5173 		dev_warn(dev, "Partner adapter not ready\n");
5174 	} else if (rc) {
5175 		dev_warn(dev, "Error %d opening adapter\n", rc);
5176 		goto reg_crq_failed;
5177 	}
5178 
5179 	retrc = 0;
5180 
5181 	tasklet_setup(&adapter->tasklet, (void *)ibmvnic_tasklet);
5182 
5183 	netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq);
5184 	snprintf(crq->name, sizeof(crq->name), "ibmvnic-%x",
5185 		 adapter->vdev->unit_address);
5186 	rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, crq->name, adapter);
5187 	if (rc) {
5188 		dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n",
5189 			vdev->irq, rc);
5190 		goto req_irq_failed;
5191 	}
5192 
5193 	rc = vio_enable_interrupts(vdev);
5194 	if (rc) {
5195 		dev_err(dev, "Error %d enabling interrupts\n", rc);
5196 		goto req_irq_failed;
5197 	}
5198 
5199 	crq->cur = 0;
5200 	spin_lock_init(&crq->lock);
5201 
5202 	return retrc;
5203 
5204 req_irq_failed:
5205 	tasklet_kill(&adapter->tasklet);
5206 	do {
5207 		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
5208 	} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
5209 reg_crq_failed:
5210 	dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL);
5211 map_failed:
5212 	free_page((unsigned long)crq->msgs);
5213 	crq->msgs = NULL;
5214 	return retrc;
5215 }
5216 
5217 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset)
5218 {
5219 	struct device *dev = &adapter->vdev->dev;
5220 	unsigned long timeout = msecs_to_jiffies(20000);
5221 	u64 old_num_rx_queues, old_num_tx_queues;
5222 	int rc;
5223 
5224 	adapter->from_passive_init = false;
5225 
5226 	if (reset) {
5227 		old_num_rx_queues = adapter->req_rx_queues;
5228 		old_num_tx_queues = adapter->req_tx_queues;
5229 		reinit_completion(&adapter->init_done);
5230 	}
5231 
5232 	adapter->init_done_rc = 0;
5233 	rc = ibmvnic_send_crq_init(adapter);
5234 	if (rc) {
5235 		dev_err(dev, "Send crq init failed with error %d\n", rc);
5236 		return rc;
5237 	}
5238 
5239 	if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
5240 		dev_err(dev, "Initialization sequence timed out\n");
5241 		return -1;
5242 	}
5243 
5244 	if (adapter->init_done_rc) {
5245 		release_crq_queue(adapter);
5246 		return adapter->init_done_rc;
5247 	}
5248 
5249 	if (adapter->from_passive_init) {
5250 		adapter->state = VNIC_OPEN;
5251 		adapter->from_passive_init = false;
5252 		return -1;
5253 	}
5254 
5255 	if (reset &&
5256 	    test_bit(0, &adapter->resetting) && !adapter->wait_for_reset &&
5257 	    adapter->reset_reason != VNIC_RESET_MOBILITY) {
5258 		if (adapter->req_rx_queues != old_num_rx_queues ||
5259 		    adapter->req_tx_queues != old_num_tx_queues) {
5260 			release_sub_crqs(adapter, 0);
5261 			rc = init_sub_crqs(adapter);
5262 		} else {
5263 			rc = reset_sub_crq_queues(adapter);
5264 		}
5265 	} else {
5266 		rc = init_sub_crqs(adapter);
5267 	}
5268 
5269 	if (rc) {
5270 		dev_err(dev, "Initialization of sub crqs failed\n");
5271 		release_crq_queue(adapter);
5272 		return rc;
5273 	}
5274 
5275 	rc = init_sub_crq_irqs(adapter);
5276 	if (rc) {
5277 		dev_err(dev, "Failed to initialize sub crq irqs\n");
5278 		release_crq_queue(adapter);
5279 	}
5280 
5281 	return rc;
5282 }
5283 
5284 static struct device_attribute dev_attr_failover;
5285 
5286 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
5287 {
5288 	struct ibmvnic_adapter *adapter;
5289 	struct net_device *netdev;
5290 	unsigned char *mac_addr_p;
5291 	int rc;
5292 
5293 	dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n",
5294 		dev->unit_address);
5295 
5296 	mac_addr_p = (unsigned char *)vio_get_attribute(dev,
5297 							VETH_MAC_ADDR, NULL);
5298 	if (!mac_addr_p) {
5299 		dev_err(&dev->dev,
5300 			"(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n",
5301 			__FILE__, __LINE__);
5302 		return 0;
5303 	}
5304 
5305 	netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter),
5306 				   IBMVNIC_MAX_QUEUES);
5307 	if (!netdev)
5308 		return -ENOMEM;
5309 
5310 	adapter = netdev_priv(netdev);
5311 	adapter->state = VNIC_PROBING;
5312 	dev_set_drvdata(&dev->dev, netdev);
5313 	adapter->vdev = dev;
5314 	adapter->netdev = netdev;
5315 	adapter->login_pending = false;
5316 
5317 	ether_addr_copy(adapter->mac_addr, mac_addr_p);
5318 	ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
5319 	netdev->irq = dev->irq;
5320 	netdev->netdev_ops = &ibmvnic_netdev_ops;
5321 	netdev->ethtool_ops = &ibmvnic_ethtool_ops;
5322 	SET_NETDEV_DEV(netdev, &dev->dev);
5323 
5324 	INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset);
5325 	INIT_DELAYED_WORK(&adapter->ibmvnic_delayed_reset,
5326 			  __ibmvnic_delayed_reset);
5327 	INIT_LIST_HEAD(&adapter->rwi_list);
5328 	spin_lock_init(&adapter->rwi_lock);
5329 	spin_lock_init(&adapter->state_lock);
5330 	mutex_init(&adapter->fw_lock);
5331 	init_completion(&adapter->init_done);
5332 	init_completion(&adapter->fw_done);
5333 	init_completion(&adapter->reset_done);
5334 	init_completion(&adapter->stats_done);
5335 	clear_bit(0, &adapter->resetting);
5336 
5337 	do {
5338 		rc = init_crq_queue(adapter);
5339 		if (rc) {
5340 			dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n",
5341 				rc);
5342 			goto ibmvnic_init_fail;
5343 		}
5344 
5345 		rc = ibmvnic_reset_init(adapter, false);
5346 		if (rc && rc != EAGAIN)
5347 			goto ibmvnic_init_fail;
5348 	} while (rc == EAGAIN);
5349 
5350 	rc = init_stats_buffers(adapter);
5351 	if (rc)
5352 		goto ibmvnic_init_fail;
5353 
5354 	rc = init_stats_token(adapter);
5355 	if (rc)
5356 		goto ibmvnic_stats_fail;
5357 
5358 	netdev->mtu = adapter->req_mtu - ETH_HLEN;
5359 	netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
5360 	netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
5361 
5362 	rc = device_create_file(&dev->dev, &dev_attr_failover);
5363 	if (rc)
5364 		goto ibmvnic_dev_file_err;
5365 
5366 	netif_carrier_off(netdev);
5367 	rc = register_netdev(netdev);
5368 	if (rc) {
5369 		dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc);
5370 		goto ibmvnic_register_fail;
5371 	}
5372 	dev_info(&dev->dev, "ibmvnic registered\n");
5373 
5374 	adapter->state = VNIC_PROBED;
5375 
5376 	adapter->wait_for_reset = false;
5377 	adapter->last_reset_time = jiffies;
5378 	return 0;
5379 
5380 ibmvnic_register_fail:
5381 	device_remove_file(&dev->dev, &dev_attr_failover);
5382 
5383 ibmvnic_dev_file_err:
5384 	release_stats_token(adapter);
5385 
5386 ibmvnic_stats_fail:
5387 	release_stats_buffers(adapter);
5388 
5389 ibmvnic_init_fail:
5390 	release_sub_crqs(adapter, 1);
5391 	release_crq_queue(adapter);
5392 	mutex_destroy(&adapter->fw_lock);
5393 	free_netdev(netdev);
5394 
5395 	return rc;
5396 }
5397 
5398 static void ibmvnic_remove(struct vio_dev *dev)
5399 {
5400 	struct net_device *netdev = dev_get_drvdata(&dev->dev);
5401 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5402 	unsigned long flags;
5403 
5404 	spin_lock_irqsave(&adapter->state_lock, flags);
5405 
5406 	/* If ibmvnic_reset() is scheduling a reset, wait for it to
5407 	 * finish. Then, set the state to REMOVING to prevent it from
5408 	 * scheduling any more work and to have reset functions ignore
5409 	 * any resets that have already been scheduled. Drop the lock
5410 	 * after setting state, so __ibmvnic_reset() which is called
5411 	 * from the flush_work() below, can make progress.
5412 	 */
5413 	spin_lock_irqsave(&adapter->rwi_lock, flags);
5414 	adapter->state = VNIC_REMOVING;
5415 	spin_unlock_irqrestore(&adapter->rwi_lock, flags);
5416 
5417 	spin_unlock_irqrestore(&adapter->state_lock, flags);
5418 
5419 	flush_work(&adapter->ibmvnic_reset);
5420 	flush_delayed_work(&adapter->ibmvnic_delayed_reset);
5421 
5422 	rtnl_lock();
5423 	unregister_netdevice(netdev);
5424 
5425 	release_resources(adapter);
5426 	release_sub_crqs(adapter, 1);
5427 	release_crq_queue(adapter);
5428 
5429 	release_stats_token(adapter);
5430 	release_stats_buffers(adapter);
5431 
5432 	adapter->state = VNIC_REMOVED;
5433 
5434 	rtnl_unlock();
5435 	mutex_destroy(&adapter->fw_lock);
5436 	device_remove_file(&dev->dev, &dev_attr_failover);
5437 	free_netdev(netdev);
5438 	dev_set_drvdata(&dev->dev, NULL);
5439 }
5440 
5441 static ssize_t failover_store(struct device *dev, struct device_attribute *attr,
5442 			      const char *buf, size_t count)
5443 {
5444 	struct net_device *netdev = dev_get_drvdata(dev);
5445 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5446 	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
5447 	__be64 session_token;
5448 	long rc;
5449 
5450 	if (!sysfs_streq(buf, "1"))
5451 		return -EINVAL;
5452 
5453 	rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address,
5454 			 H_GET_SESSION_TOKEN, 0, 0, 0);
5455 	if (rc) {
5456 		netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n",
5457 			   rc);
5458 		return -EINVAL;
5459 	}
5460 
5461 	session_token = (__be64)retbuf[0];
5462 	netdev_dbg(netdev, "Initiating client failover, session id %llx\n",
5463 		   be64_to_cpu(session_token));
5464 	rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
5465 				H_SESSION_ERR_DETECTED, session_token, 0, 0);
5466 	if (rc) {
5467 		netdev_err(netdev, "Client initiated failover failed, rc %ld\n",
5468 			   rc);
5469 		return -EINVAL;
5470 	}
5471 
5472 	return count;
5473 }
5474 
5475 static DEVICE_ATTR_WO(failover);
5476 
5477 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev)
5478 {
5479 	struct net_device *netdev = dev_get_drvdata(&vdev->dev);
5480 	struct ibmvnic_adapter *adapter;
5481 	struct iommu_table *tbl;
5482 	unsigned long ret = 0;
5483 	int i;
5484 
5485 	tbl = get_iommu_table_base(&vdev->dev);
5486 
5487 	/* netdev inits at probe time along with the structures we need below*/
5488 	if (!netdev)
5489 		return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl);
5490 
5491 	adapter = netdev_priv(netdev);
5492 
5493 	ret += PAGE_SIZE; /* the crq message queue */
5494 	ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl);
5495 
5496 	for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++)
5497 		ret += 4 * PAGE_SIZE; /* the scrq message queue */
5498 
5499 	for (i = 0; i < adapter->num_active_rx_pools; i++)
5500 		ret += adapter->rx_pool[i].size *
5501 		    IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl);
5502 
5503 	return ret;
5504 }
5505 
5506 static int ibmvnic_resume(struct device *dev)
5507 {
5508 	struct net_device *netdev = dev_get_drvdata(dev);
5509 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
5510 
5511 	if (adapter->state != VNIC_OPEN)
5512 		return 0;
5513 
5514 	tasklet_schedule(&adapter->tasklet);
5515 
5516 	return 0;
5517 }
5518 
5519 static const struct vio_device_id ibmvnic_device_table[] = {
5520 	{"network", "IBM,vnic"},
5521 	{"", "" }
5522 };
5523 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table);
5524 
5525 static const struct dev_pm_ops ibmvnic_pm_ops = {
5526 	.resume = ibmvnic_resume
5527 };
5528 
5529 static struct vio_driver ibmvnic_driver = {
5530 	.id_table       = ibmvnic_device_table,
5531 	.probe          = ibmvnic_probe,
5532 	.remove         = ibmvnic_remove,
5533 	.get_desired_dma = ibmvnic_get_desired_dma,
5534 	.name		= ibmvnic_driver_name,
5535 	.pm		= &ibmvnic_pm_ops,
5536 };
5537 
5538 /* module functions */
5539 static int __init ibmvnic_module_init(void)
5540 {
5541 	pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string,
5542 		IBMVNIC_DRIVER_VERSION);
5543 
5544 	return vio_register_driver(&ibmvnic_driver);
5545 }
5546 
5547 static void __exit ibmvnic_module_exit(void)
5548 {
5549 	vio_unregister_driver(&ibmvnic_driver);
5550 }
5551 
5552 module_init(ibmvnic_module_init);
5553 module_exit(ibmvnic_module_exit);
5554